<!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN" "http://www.w3.org/TR/REC-html40/loose.dtd">

<html lang="en">

<head>

<meta content="text/html; charset=US-ASCII" http-equiv="Content-Type">

<title>

GitLab

</title>

<style>img {

max-width: 100%; height: auto;

}

</style>

</head>

<body>

<div class="content">

<h3>

Matthew Pickering pushed to branch wip/ghc-debug

at <a href="https://gitlab.haskell.org/ghc/ghc">Glasgow Haskell Compiler / GHC</a>

</h3>

<h4>

Commits:

</h4>

<ul>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/f8ec32d79ff876686cb191151c5a151773ae6495">f8ec32d7</a></strong>

<div>

<span>by Alp Mestanogullari</span>

<i>at 2019-11-11T11:36:44-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">ci: push perf test metrics even when the testsuite doesn't pass

The corresponding commit might introduce a regression on a perf test, in which

case we certainly want to record it. The testsuite might also fail because

of a test unrelated to performance, in which case we want to record that the

perf test results were good.

Either way, we likely want to record them under all circumstances but

we don't without this patch.

Metric Decrease:

    T3586

Metric Increase:

    lazy-bs-alloc

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/643d42fcf68f9b96bc8829d20bcd8f5fef84a78c">643d42fc</a></strong>

<div>

<span>by Alp Mestanogullari</span>

<i>at 2019-11-12T18:40:19-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: don't collect compiler stats in collect_runtime_residency

We instead want to collect the runtime stats (with collect_stats, instead of

collect_compiler_stats).

This should fix a number of perf tests failures we have been seeing, where

we suddenly started measuring metrics we didn't intend to measure, which

tend to fall outside of the acceptance window.

Metric Decrease:

    lazy-bs-alloc

    T3586

Metric Increase:

    space_leak_001

    T4801

    T5835

    T12791

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/535d0edc11e66a9a0bdfda676dd614833d86df68">535d0edc</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2019-11-13T07:06:12-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Document CmmTopInfo type

[ci skip]

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/2d4f9ad89cb7e79c1f90983bf4898a5f4e3c7457">2d4f9ad8</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-11-13T07:06:49-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Ensure that coreView/tcView are able to inline

Previously an import cycle between Type and TyCoRep meant that several

functions in TyCoRep ended up SOURCE import coreView. This is quite

unfortunate as coreView is intended to be fused into a larger pattern

match and not incur an extra call.

Fix this with a bit of restructuring:

 * Move the functions in `TyCoRep` which depend upon things in `Type`

   into `Type`

 * Fold contents of `Kind` into `Type` and turn `Kind` into a simple

   wrapper re-exporting kind-ish things from `Type`

 * Clean up the redundant imports that popped up as a result

Closes #17441.

Metric Decrease:

    T4334

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/b795637fc74ffb3ac2f37a50f25468a67bf54bc0">b795637f</a></strong>

<div>

<span>by Alp Mestanogullari</span>

<i>at 2019-11-13T07:07:28-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">hadrian: fix Windows CI script

By only using 'export' from within bash commands.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/6885e22c83ac34e48a2abdb1958b07904f28eae6">6885e22c</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-11-13T07:08:03-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Add test for #17458

As noted in #17458, QuantifiedConstraints and UndecideableInstances

could previously be used to write programs which can loop at runtime.

This was fixed in !1870.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/b4b19d8990f345f8ae7fde130ba84327353433a9">b4b19d89</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-11-13T07:08:03-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">users guide: Fix broken link

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/9a939a6cdfb2a1357f9990af329096dff6b5be5c">9a939a6c</a></strong>

<div>

<span>by Ryan Scott</span>

<i>at 2019-11-13T07:08:40-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Print name prefixly in the Outputable instance for StandaloneKindSig

Issue #17461 was occurring because the `Outputable` instance for

standalone kind signatures was simply calling `ppr` on the name in

the kind signature, which does not add parentheses to infix names.

The solution is simple: use `pprPrefixOcc` instead.

Fixes #17461.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/a06cfb59d21c9cf6f53a8b1acedb075988a6c5ca">a06cfb59</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2019-11-13T07:09:18-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Only pass mod_location with HscRecomp instead of the entire ModSummary

HscRecomp users only need the ModLocation of the module being compiled,

so only pass that to users instead of the entire ModSummary

Metric Decrease:

    T4801

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/dd49b3f079a4ff59fd9ed0a8ea5585f0fcc0e56a">dd49b3f0</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-11-13T17:01:21-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Bump Haskeline and add exceptions as boot library

Haskeline now depends upon exceptions. See #16752.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/b06b1858f2765f27da2914bd8d17464eeb13d20c">b06b1858</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-11-14T11:30:20-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">base: Bump version to 4.14.0.0

Metric Increase:

    T4801

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/6ab80439d0c57b9a3aa03d0d57c5e095257ee2cf">6ab80439</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-11-14T23:05:30-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Allow Windows to fail again

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/46afc380b1445d6bcff0c954569a59420c847534">46afc380</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-11-15T09:45:36-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Install process to global pkgdb before starting build

This is an attempt to mitigate #17480 by ensuring that a functional version of

the process library is available before attempting the build.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/8c5cb80698da2608f08738822d97cdcb682c5544">8c5cb806</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-11-15T10:45:55-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Bump supported LLVM version to 9.0

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/8e5851f0708134efe0ff0e139a1464337410d0ce">8e5851f0</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-11-15T10:45:55-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">llvm-targets: Update with Clang 9

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/f3ffec2729abb4bccaf16e974bef376c8968516e">f3ffec27</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-11-15T11:54:26-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Increase acceptance window of T4801

This statistic is rather unstable. Hopefully fixes #17475.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/c2991f16cb6f5b4e7cff46a394dda4247d973f44">c2991f16</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-11-15T11:56:10-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">users-guide: Drop 8.6.1 release notes

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/e8da1354f0d0cac7b152cff9673f78e5156d4c0c">e8da1354</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-11-17T06:48:16-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Fix submodule linter

We ran it against the .git directory despite the fact that the linter

wants to be run against the repository.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/13290f91e10d503078787ed5b8fb0a523358f124">13290f91</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-11-17T06:48:16-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Bump version to 8.10.0

Bumps haddock submodule.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/fa98f82348b268043dce918e6eb9dab434eb3108">fa98f823</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-11-17T06:48:16-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Don't collect residency for T4801

I previously increased the size of the acceptance window from 2% to 5%

but this still isn't enough. Regardless, measuring bytes allocated

should be sufficient to catch any regressions.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/002b284269de63504ed3b0c1e22c8ecfd74e78d9">002b2842</a></strong>

<div>

<span>by Ivan Kasatenko</span>

<i>at 2019-11-17T06:49:22-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Make test 16916 more stable across runs

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/ca89dd3b591f82e73a88cf433d280ece317ace5e">ca89dd3b</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-11-17T06:58:17-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">users-guide: Address #17329

Adopts the language suggested by @JakobBruenker.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/2f5ed225b78b32c65d023072d78ae5d176e2f04b">2f5ed225</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-11-17T07:16:32-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">exceptions: Bump submodule back to master

The previous commit hasn't made it to master yet.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/34515e7c94250a8b5b85453a0481d0c8d313a7c5">34515e7c</a></strong>

<div>

<span>by nineonine</span>

<i>at 2019-11-17T13:33:22-08:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix random typos [skip ci]

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/4a37a29b67db078ba29722c5615a37fd8091989b">4a37a29b</a></strong>

<div>

<span>by Mario Blažević</span>

<i>at 2019-11-17T17:26:24-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fixed issue #17435, missing Data instances

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/97f1bcaeb05f443d267a1a0692ed45e35509e84f">97f1bcae</a></strong>

<div>

<span>by Andreas Klebinger</span>

<i>at 2019-11-17T17:26:24-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Turn some comments into GHC.Hs.Utils into haddocks

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/cf7f8e5bbec83da1bb62075968bc78c86414c245">cf7f8e5b</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-11-17T17:26:26-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Skip T17414 on Linux

It is typical for $TMP to be a small tmpfson Linux. This test will fail

in such cases since we must create a file larger than the filesystem.

See #17459.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/88013b784d77c069b7c083244d04a59ac2da2895">88013b78</a></strong>

<div>

<span>by nineonine</span>

<i>at 2019-11-19T11:53:16-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Optimize MonadUnique instances based on IO (#16843)

Metric Decrease:

    T14683

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/a8adb5b4d782b6166160a80adbf6c59452dc964d">a8adb5b4</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-11-19T11:53:55-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">desugar: Drop stale Note [Matching seqId]

The need for this note vanished in

eae703aa60f41fd232be5478e196b661839ec3de.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/08d595c026f0774c51d85cbd790518055970ff2d">08d595c0</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-11-19T11:53:55-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Give seq a more precise type and remove magic

`GHC.Prim.seq` previously had the rather plain type:

    seq :: forall a b. a -> b -> b

However, it also had a special typing rule to applications

where `b` is not of kind `Type`.

Issue #17440 noted that levity polymorphism allows us to rather give

it the more precise type:

    seq :: forall (r :: RuntimeRep) a (b :: TYPE r). a -> b -> b

This allows us to remove the special typing rule that we previously

required to allow applications on unlifted arguments. T9404 contains a

non-Type application of `seq` which should verify that this works as

expected.

Closes #17440.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/ec8a463d1ff948ba9b1b0fbb538f7d5a237bf44a">ec8a463d</a></strong>

<div>

<span>by Viktor Dukhovni</span>

<i>at 2019-11-19T11:54:45-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Enable USE_PTHREAD_FOR_ITIMER also on FreeBSD

If using a pthread instead of a timer signal is more reliable, and

has no known drawbacks, then FreeBSD is also capable of supporting

this mode of operation (tested on FreeBSD 12 with GHC 8.8.1, but

no reason why it would not also work on FreeBSD 11 or GHC 8.6).

Proposed by Kevin Zhang in:

    https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=241849

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/cd40e12ad8072c6b881c6bb650e39e4d29c9718c">cd40e12a</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2019-11-19T11:55:36-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Packages.hs: use O(n*log(n)) ordNub instead of O(n*n) nub

As reported in #8173 in some environments package lists can get quite

long, so we use more efficient ordNub instead of nub on package lists.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/2b27cc164ae5a978a8abd9966d04af74628a5396">2b27cc16</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-11-19T11:56:21-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Properly account for libdw paths in make build system

Should finally fix #17255.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/0418c38d55c7a47967187dce2db5ea2ab1021b1e">0418c38d</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-11-19T11:56:58-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">rts: Add missing include of SymbolExtras.h

This broke the Windows build.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/c819c0e413069cb4e16236fb3e6576658060a17d">c819c0e4</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-11-19T11:57:36-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">nonmoving: Use correct info table pointer accessor

Previously we used INFO_PTR_TO_STRUCT instead of

THUNK_INFO_PTR_TO_STRUCT when looking at a thunk. These two happen to be

equivalent on 64-bit architectures due to alignment considerations

however they are different on 32-bit platforms. This lead to #17487.

To fix this we also employ a small optimization: there is only one thunk

of type WHITEHOLE (namely stg_WHITEHOLE_info). Consequently, we can just

use a plain pointer comparison instead of testing against info->type.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/deed8e310cecb2491e5c333afc58cb720a39b565">deed8e31</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-11-19T11:57:36-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">nonmoving: Fix incorrect masking in mark queue type test

We were using TAG_BITS instead of TAG_MASK. This happened to work on

64-bit platforms where TAG_BITS==3 since we only use tag values 0 and

3. However, this broken on 32-bit platforms where TAG_BITS==2.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/097f807214d5f4422a1d2cbacd8bc89908eddbdb">097f8072</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-11-19T11:57:36-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">nonmoving: Rework mark queue representation

The previous representation needlessly limited the array length to

16-bits on 32-bit platforms.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/eb7b233a3db9f3958bdf8d8eb6c76c6d9f00f5b9">eb7b233a</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-11-19T11:57:36-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">nonmoving: Fix handling on large object marking on 32-bit

Previously we would reset the pointer pointing to the object to be

marked to the beginning of the block when marking a large object. This

did no harm on 64-bit but on 32-bit it broke, e.g. `arr020`, since we

align pinned ByteArray allocations such that the payload is 8

byte-aligned. This means that the object might not begin at the

beginning of the block.,

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/a7571a74df15585a48318267a3f8e995fa1270b2">a7571a74</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-11-19T11:57:36-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Increase width of stack003 test

Previously the returned tuple seemed to fit in registers on amd64. This

meant that non-moving collector bug would cause the test to fail on i386

yet not amd64.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/098d50176a2521d08efe6fdbab0ede2b7419786a">098d5017</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-11-19T11:57:36-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">nonmoving: Drop redundant write barrier on stack underflow

Previously we would push stack-carried return values to the new stack on

a stack overflow. While the precise reasoning for this barrier is

unfortunately lost to history, in hindsight I suspect it was prompted by

a missing barrier elsewhere (that has been since fixed).

Moreover, there the redundant barrier is actively harmful: the stack may

contain non-pointer values; blindly pushing these to the mark queue will

result in a crash. This is precisely what happened in the `stack003`

test. However, because of a (now fixed) deficiency in the test this

crash did not trigger on amd64.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/e57b7cc6d8b1222e0939d19c265b51d2c3c2b4c0">e57b7cc6</a></strong>

<div>

<span>by Roland Zumkeller</span>

<i>at 2019-11-19T20:39:19-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Changing Thread IDs from 32 bits to 64 bits.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/d1f3c63701b7f0fd675f792af7f33c5b11eaff83">d1f3c637</a></strong>

<div>

<span>by Roland Zumkeller</span>

<i>at 2019-11-19T20:39:19-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Use pointer equality in Eq/Ord for ThreadId

Changes (==) to use only pointer equality. This is safe because two

threads are the same iff they have the same id.

Changes `compare` to check pointer equality first and fall back on ids

only in case of inequality.

See discussion in #16761.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/ef8a08e0dd4e5b908b7fbce1b3101dc311c4d3e1">ef8a08e0</a></strong>

<div>

<span>by Alexey Kuleshevich</span>

<i>at 2019-11-19T20:39:20-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">hpc: Fix encoding issues. Add test for and fix #17073

* Make sure files are being read/written in UTF-8. Set encoding while writing

  HTML output. Also set encoding while writing and reading .tix files although

  we don't yet have a ticket complaining that this poses problems.

* Set encoding in html header to utf8

* Upgrade to new version of 'hpc' library and reuse `readFileUtf8`

  and `writeFileUtf8` functions

* Update git submodule for `hpc`

* Bump up `hpc` executable version

Co-authored-by: Ben Gamari <ben@smart-cactus.org>

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/b79e46d682d144c6a7a9ff6e7dca0d892b690654">b79e46d6</a></strong>

<div>

<span>by Vladislav Zavialov</span>

<i>at 2019-11-19T20:39:20-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Strip parentheses in expressions contexts in error messages

This makes error messages a tad less noisy.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/13bbde77ecba66c6dd960f948d726aed62c583aa">13bbde77</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-11-21T13:56:56-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Bump hsc2hs submodule

Including Phyx's backport of the process changes fixing #17480.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/d4d10501c6ec995711d626eead58b84bf413d0b3">d4d10501</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-11-23T09:42:38-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Bump hsc2hs submodule again

This fixes the Darwin build.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/889d475bca24254a42ec1178a6d68caaa4eec678">889d475b</a></strong>

<div>

<span>by nineonine</span>

<i>at 2019-11-23T18:53:29-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix typo in Note reference [skip ci]

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/8a33abfcdf5a3ae9ae1777b92891890d6a045f8b">8a33abfc</a></strong>

<div>

<span>by Ryan Scott</span>

<i>at 2019-11-23T18:54:05-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Target the IsList instance for ZipList at base-4.14.0.0 (#17489)

This moves the changelog entry about the instance from

`base-4.15.0.0` to `base-4.14.0.0`. This accomplishes part (1)

from #17489.

[ci skip]

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/e43e6ece1418f84e50d572772394ab639a083e79">e43e6ece</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-11-23T18:54:41-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">rts: Expose interface for configuring EventLogWriters

This exposes a set of interfaces from the GHC API for configuring

EventLogWriters. These can be used by consumers like

[ghc-eventlog-socket](https://github.com/bgamari/ghc-eventlog-socket).

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/de6bbdf27f1831818598c7b334cb5b247aa97af7">de6bbdf2</a></strong>

<div>

<span>by Matheus Magalhães de Alcantara</span>

<i>at 2019-11-23T18:55:23-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Take care to not eta-reduce jumps in CorePrep

CorePrep already had a check to prevent it from eta-reducing Ids that

respond true to hasNoBinding (foreign calls, constructors for unboxed

sums and products, and Ids with compulsory unfoldings). It did not,

however, consider join points as ids that 'must be saturated'.

Checking whether the Id responds True to 'isJoinId' should prevent

CorePrep from turning saturated jumps like the following (from #17429)

into undersaturated ones:

      (\ eta_XP ->

         join { mapped_s1vo _ = lvl_s1vs } in jump mapped_s1vo eta_XP)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/4a1e7e47f797fab4165b7cba05edc08d41f5d80e">4a1e7e47</a></strong>

<div>

<span>by Matheus Magalhães de Alcantara</span>

<i>at 2019-11-23T18:55:23-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Make CorePrep.tryEtaReducePrep and CoreUtils.tryEtaReduce line up

Simon PJ says he prefers this fix to #17429 over banning eta-reduction

for jumps entirely. Sure enough, this also works.

Test case: simplCore/should_compile/T17429.hs

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/15f1dc3316db0b07434862de8382b4ddd27fecca">15f1dc33</a></strong>

<div>

<span>by Ryan Scott</span>

<i>at 2019-11-23T18:56:00-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Prevent -optc arguments from being duplicated in reverse order (#17471)

This reverts a part of commit

7bc5d6c6578ab9d60a83b81c7cc14819afef32ba that causes all arguments

to `-optc` (and `-optcxx`) to be passed twice to the C/C++ compiler,

once in reverse order and then again in the correct order. While

passing duplicate arguments is usually harmless it can cause breakage

in this pattern, which is employed by Hackage libraries in the wild:

```

ghc Foo.hs foo.c -optc-D -optcFOO

```

As `FOO -D -D FOO` will cause compilers to error.

Fixes #17471.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/e85c9b22e94cdc31d59ffc5ff72fcb5b4597d683">e85c9b22</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-11-23T18:56:36-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Bump ghc version to 8.11

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/0e6c204570f77950c9019faa50ff867c158249c5">0e6c2045</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-11-23T18:57:12-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">rts: Consolidate spinlock implementation

Previously we had two distinct implementations: one with spinlock

profiling and another without. This seems like needless duplication.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/cb11fcb52ce8d5d1473a380db5fc0d32bf8ae75c">cb11fcb5</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-11-23T18:57:49-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Packages: Don't use expectJust

Throw a slightly more informative error on failure. Motivated by the

errors seen in !2160.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/5747ebe9bc9b2936e9bdae0cb752124c42d069d5">5747ebe9</a></strong>

<div>

<span>by Sebastian Graf</span>

<i>at 2019-11-23T18:58:25-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Stricten functions ins GHC.Natural

This brings `Natural` on par with `Integer` and fixes #17499.

Also does some manual CSE for 0 and 1 literals.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/c14b723f7fa24b4db0b561f9ab0d6e0f4261fd2f">c14b723f</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2019-11-23T18:59:06-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Bump exceptions submodule

Adds a few files generated by GHC's configure script to .gitignore

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/7b4c7b75be351e9b6a40093cbccab728dd89064f">7b4c7b75</a></strong>

<div>

<span>by Brian Wignall</span>

<i>at 2019-11-23T19:04:52-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix typos

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/6008206a9e050ed0e3db1f09e0b2a641420d4ffd">6008206a</a></strong>

<div>

<span>by Viktor Dukhovni</span>

<i>at 2019-11-24T14:33:18-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">On FreeBSD 12 sys/sysctl.h requires sys/types.h

Else build fails with:

    In file included from ExecutablePath.hsc:42:

    /usr/include/sys/sysctl.h:1062:25: error: unknown type name 'u_int'; did you mean 'int'?

     int sysctl(const int *, u_int, void *, size_t *, const void *, size_t);

                             ^~~~~

                             int

    compiling libraries/base/dist-install/build/System/Environment/ExecutablePath_hsc_make.c failed (exit code 1)

Perhaps also also other FreeBSD releases, but additional include

will no harm even if not needed.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/b694b5662f8c9915b1a9d072cd530cd29124194a">b694b566</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-11-24T14:33:54-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">configure: Fix HAVE_C11_ATOMICS macro

Previously we were using AC_DEFINE instead of

AC_DEFINE_UNQUOTED, resulted in the variable not being

interpolated.

Fixes #17505.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/8b8dc36653878de5556e368bd3e93abf66f839e9">8b8dc366</a></strong>

<div>

<span>by Krzysztof Gogolewski</span>

<i>at 2019-11-25T14:37:38+01:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Remove prefix arrow support for GADTs (#17211)

This reverts the change in #9096.

The specialcasing done for prefix (->) is brittle and

does not support VTA, type families, type synonyms etc.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/5a08f7d405bbedfdc20c07f64726899f594e9d07">5a08f7d4</a></strong>

<div>

<span>by Sebastian Graf</span>

<i>at 2019-11-27T00:14:59-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Make warnings for TH splices opt-in

In #17270 we have the pattern-match checker emit incorrect warnings. The

reason for that behavior is ultimately an inconsistency in whether we

treat TH splices as written by the user (`FromSource :: Origin`) or as

generated code (`Generated`). This was first reported in #14838.

The current solution is to TH splices as `Generated` by default and only

treat them as `FromSource` when the user requests so

(-fenable-th-splice-warnings). There are multiple reasons for opt-in

rather than opt-out:

  * It's not clear that the user that compiles a splice is the author of the code

    that produces the warning. Think of the situation where she just splices in

    code from a third-party library that produces incomplete pattern matches.

    In this scenario, the user isn't even able to fix that warning.

  * Gathering information for producing the warnings (pattern-match check

    warnings in particular) is costly. There's no point in doing so if the user

    is not interested in those warnings.

Fixes #17270, but not #14838, because the proper solution needs a GHC

proposal extending the TH AST syntax.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/8168b42a95ddf37c56958955eef065eb8747470f">8168b42a</a></strong>

<div>

<span>by Vladislav Zavialov</span>

<i>at 2019-11-27T11:32:18+03:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Whitespace-sensitive bang patterns (#1087, #17162)

This patch implements a part of GHC Proposal #229 that covers five

operators:

* the bang operator (!)

* the tilde operator (~)

* the at operator (@)

* the dollar operator ($)

* the double dollar operator ($$)

Based on surrounding whitespace, these operators are disambiguated into

bang patterns, lazy patterns, strictness annotations, type

applications, splices, and typed splices.

This patch doesn't cover the (-) operator or the -Woperator-whitespace

warning, which are left as future work.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/9e5477c46bbcd3beadd1a67a294056628e394945">9e5477c4</a></strong>

<div>

<span>by Ryan Scott</span>

<i>at 2019-11-27T20:01:50-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix @since annotations for isResourceVanishedError and friends (#17488)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/e122ba33e8426a7b7f18216c451f6288e90c966e">e122ba33</a></strong>

<div>

<span>by Sergei Trofimovich</span>

<i>at 2019-11-27T20:02:29-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">.gitmodules: tweak 'exception' URL to avoid redirection warnings

Avoid initial close warning of form:

```

Cloning into 'exceptions'...

warning: redirecting to https://gitlab.haskell.org/ghc/packages/exceptions.git/

```

Signed-off-by: Sergei Trofimovich <slyfox@gentoo.org>

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/5f84b52a9c439ae2739bf1899a2adbae9c6d4f67">5f84b52a</a></strong>

<div>

<span>by Philipp Krüger</span>

<i>at 2019-11-28T02:54:05-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Reduce boolean blindness in OccInfo(OneOcc) #17482

* Transformed the type aliases `InterestingCxt`, `InsideLam` and `OneBranch`

  into data types.

* Added Semigroup and Monoid instances for use in orOccInfo in OccurAnal.hs

* Simplified some usage sites by using pattern matching instead of boolean algebra.

Metric Increase:

    T12150

This increase was on a Mac-build of exactly 1%. This commit does *not* re-intruduce

the asymptotic memory usage described in T12150.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/3748ba3a01f81078025b86023833784055d1af57">3748ba3a</a></strong>

<div>

<span>by Brian Wignall</span>

<i>at 2019-11-28T02:54:52-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix typos, using Wikipedia list of common typos

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/6c59cc71dc20f26d5a6650b16a82faeae72f2065">6c59cc71</a></strong>

<div>

<span>by Stefan Schulze Frielinghaus</span>

<i>at 2019-11-28T02:55:33-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix endian handling of LLVM backend

Get rid of CPP macro WORDS_BIGENDIAN which is not defined anymore, and

replace it by DynFlag. This fixes partially #17337.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/6985e0fc4f6fb30c1effd356d87c1a0629aa9cd0">6985e0fc</a></strong>

<div>

<span>by Vladislav Zavialov</span>

<i>at 2019-11-28T15:47:53+03:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Factor out HsSCC/HsCoreAnn/HsTickPragma into HsPragE

This is a refactoring with no user-visible changes (except for GHC API

users). Consider the HsExpr constructors that correspond to user-written

pragmas:

  HsSCC         representing  {-# SCC ... #-}

  HsCoreAnn     representing  {-# CORE ... #-}

  HsTickPragma  representing  {-# GENERATED ... #-}

We can factor them out into a separate datatype, HsPragE. It makes the

code a bit tidier, especially in the parser.

Before this patch:

  hpc_annot :: { Located ( (([AddAnn],SourceText),(StringLiteral,(Int,Int),(Int,Int))),

                           ((SourceText,SourceText),(SourceText,SourceText))

                         ) }

After this patch:

  prag_hpc :: { Located ([AddAnn], HsPragE GhcPs) }

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/7f695a20f5f9fe4952ca8cde45d73f3604a7cc29">7f695a20</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2019-11-29T08:25:28-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Pass ModDetails with (partial) ModIface in HscStatus

(Partial) ModIface and ModDetails are generated at the same time, but

they're passed differently: ModIface is passed in HscStatus consturctors

while ModDetails is returned in a tuple. This refactors ModDetails

passing so that it's passed around with ModIface in HscStatus

constructors. This makes the code more consistent and hopefully easier

to understand: ModIface and ModDetails are really very closely related.

It makes sense to treat them the same way.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/e921c90fec6754414e97bfabaeeb4eaf0241472d">e921c90f</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2019-11-29T08:26:07-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Improve few Foreign.Marshal.Utils docs

In copyBytes and moveBytes mention which argument is source and which is

destination.

Also fixes some of the crazy indentation in the module and cleans

trailing whitespace.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/316f24319e151446c83cbb0f2997a73e19fe4aa3">316f2431</a></strong>

<div>

<span>by Sebastian Graf</span>

<i>at 2019-11-30T02:57:58-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Hadrian docs: Rename the second "validate" entry to "slow-validate" [ci skip]

That would be in line with the implementation.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/5aba5d3218330f8ce127aa7767efcbb6f63a2db1">5aba5d32</a></strong>

<div>

<span>by Vladislav Zavialov</span>

<i>at 2019-11-30T02:58:34-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Remove HasSrcSpan (#17494)

Metric Decrease:

    haddock.compiler

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/d1de5c227160f555ea6d5679791388a980910c88">d1de5c22</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2019-11-30T02:59:13-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Use Hadrian by default in validate script (#17527)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/3a96a0b6db6a32457ae2f91bb711c2481c767656">3a96a0b6</a></strong>

<div>

<span>by Sebastian Graf</span>

<i>at 2019-11-30T02:59:55-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Simpler Semigroup instance for InsideLam and InterestingCtxt

This mirrors the definition of `(&&)` and `(||)` now, relieving the

Simplifier of a marginal amount of pressure.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/f8cfe81aed556d00eac58e3ad0f5e27c503b636a">f8cfe81a</a></strong>

<div>

<span>by Roland Senn</span>

<i>at 2019-11-30T20:33:49+01:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Improve tests for #17171

While backporting MR !1806 to 8.8.2 (!1885) I learnt the following:

* Tests with `expect_fail` do not compare `*.stderr` output files. So a test using `expect_fail` will not detect future regressions on the `stderr` output.

* To compare the `*.stderr` output files, I have to use the `exit_code(n)` function.

* When a release is made, tests with `makefile_test` are converted to use `run_command`.

* For the test `T17171a` the return code is `1` when running `makefile_test`, however it's `2` when running `run_command`.

Therefore I decided:

* To improve my tests for #17171

* To change test T17171a from `expect_fail` to `exit_code(2)`

* To change both tests from  `makefile_test` to `run_command`

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/2b113fc957a3fb6eafd10ad41a7caf11009ef4eb">2b113fc9</a></strong>

<div>

<span>by Vladislav Zavialov</span>

<i>at 2019-12-01T08:17:05-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Update DisambECP-related comments

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/beed7c3e7c0703089a9ba4321bd3a2e442213fad">beed7c3e</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-02T03:41:37-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Fix location of typing_stubs module

This should fix the build on Debian 8.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/53251413348186c8fbe2bbe9c2702d274e469324">53251413</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-02T03:42:14-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Don't override LD_LIBRARY_PATH, only prepend

NixOS development environments often require that LD_LIBRARY_PATH

be set in order to find system libraries. T1407 was overriding

LD_LIBRARY_PATH, dropping these directories. Now it merely prepends,

its directory.</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/65400314244b2486dc4740a4aadd4e1225d38e3c">65400314</a></strong>

<div>

<span>by Krzysztof Gogolewski</span>

<i>at 2019-12-02T03:42:57-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Convert warnings into assertions

Since the invariants always hold in the testsuite, we can convert

them to asserts.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/18baed641fccc7fd91988334544d47505b577810">18baed64</a></strong>

<div>

<span>by Alan Zimmerman</span>

<i>at 2019-12-02T03:43:37-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">API Annotations: Unicode '->' on HsForallTy

The code fragment

  type family Proxy2' ∷ ∀ k → k → Type where

    Proxy2' = Proxy'

Generates AnnRarrow instead of AnnRarrowU for the first →.

Fixes #17519

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/717f323609857a40fdbefe7a5ba426747512414a">717f3236</a></strong>

<div>

<span>by Brian Wignall</span>

<i>at 2019-12-02T03:44:16-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix more typos

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/bde48f8e5bad1427a530610796d85d299997f80c">bde48f8e</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-02T11:55:34-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">More Haddock syntax in GHC.Hs.Utils

As suggested by RyanGlScott in !2163.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/038bedbc8daa842a4cd91a0da780d09c78c4984f">038bedbc</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-02T11:56:18-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Simplify: Fix pretty-printing of strictness

A colleague recently hit the panic in Simplify.addEvals and I noticed

that the message is quite unreadable due to incorrect pretty-printing.

Fix this.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/c500f652470d012fa432632ab122a12674bd9788">c500f652</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-02T11:56:54-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Fix changelog linting logic

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/8ead967d48b48020f6ce621f9b27363aca45c363">8ead967d</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-02T11:56:54-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">win32-init: Drop workaround for #17480

The `process` changes have now been merged into `hsc2hs`.

(cherry picked from commit fa029f53132ad59f847ed012d3b835452cf16615)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/d402209ace97212b59ad6f9d7cdd1e1bb15fc425">d402209a</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-02T11:56:54-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Disable Sphinx build on Debian 8

The docutils version available appears to be too old to support the

`table` directive's `:widths:` options.

(cherry picked from commit 75764487a96a7a026948b5af5022781872d12baa)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/f1f688244d7a766a2b71fbe172c86e3cb4c87539">f1f68824</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-02T11:56:54-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">base: Fix <unistd.h> #include

Previously we were including <sys/unistd.h> which is available on glibc

but not musl.

(cherry picked from commit e44b695ca7cb5f3f99eecfba05c9672c6a22205e)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/37eb94b3b060ef83685127ec10d14a9675cf508a">37eb94b3</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-02T11:56:54-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Bump Docker images

Installs pxz on Centos7

(cherry picked from commit 86960e691f7a600be247c32a7cf795bf9abf7cc4)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/aec98a79a7cc04c3fa8d0f5dfd6cb462daa4e612">aec98a79</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-02T11:56:54-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: pxz is unavailable on CentOS 7

Fall back to xz

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/6708b8e5b814bf9797e1bbac82622d815e7669b1">6708b8e5</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-02T11:56:54-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Set LANG on CentOS 7

It otherwise seems to default to ascii

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/470ef0e7d0d6c07b17252a660d72691dd02a1dad">470ef0e7</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-02T11:56:54-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Consolidate release build configuration

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/38338757e0854a0fc6c3a99d1c35392227623887">38338757</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-02T11:56:54-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Add Debian 10 builds

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/012f13b5051ff4385542f90c9692ef4ba6450db9">012f13b5</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-02T11:56:54-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Fix Windows bindist collection

Apparently variable interpolation in the `artifacts.paths` key of

`gitlab-ci.yml` doesn't work on Windows as it does on WIndows.

(cherry picked from commit 100cc756faa4468ed6950116bae30609c1c3468b)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/a0f09e23dd19f0cf223034f9b787a4f038cd995d">a0f09e23</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-02T11:56:54-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Simplify Python <3.5 fallback for TextIO

(cherry picked from commit d092d8598694c23bc07cdcc504dff52fa5f33be1)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/2b2370ec82294c527f9fbe2d1d66daff9816f837">2b2370ec</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-02T11:56:54-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Add release-x86_64-linux-deb9 job

(cherry picked from commit cbedb3c4a90649f474cb716842ba53afc5a642ca)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/b1c206fd58a8ff365d4b28dff41fec343884972b">b1c206fd</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-02T11:56:54-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Always build source tarball

(cherry picked from commit 67b5de88ef923971f1980335137e3c7193213abd)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/4cbd5b47a00a29b7835710f1b91bb93ac8e3f790">4cbd5b47</a></strong>

<div>

<span>by Sergei Trofimovich</span>

<i>at 2019-12-02T11:57:33-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">configure.ac: make cross-compiler detection stricter

Be more precise at detecting cross-compilation case.

Before the change configuration

  $ ./configure --host=x86_64-pc-linux-gnu --target=x86_64-gentoo-linux-musl

was not considered a cross-target. Even though libcs are different (`glibc` vs. `musl`).

Without this patch build fails as:

```

"inplace/bin/ghc-cabal" check libraries/integer-gmp

"inplace/bin/ghc-cabal" configure libraries/integer-gmp dist-install \

    --with-ghc="/home/slyfox/dev/git/ghc/inplace/bin/ghc-stage1" \

    --with-ghc-pkg="/home/slyfox/dev/git/ghc/inplace/bin/ghc-pkg" \

    --disable-library-for-ghci --enable-library-vanilla --enable-library-for-ghci \

    --enable-library-profiling --enable-shared --with-hscolour="/usr/bin/HsColour" \

    --configure-option=CFLAGS="-Wall \

    -Werror=unused-but-set-variable -Wno-error=inline \

    -iquote /home/slyfox/dev/git/ghc/libraries/integer-gmp" \

    --configure-option=LDFLAGS="  " --configure-option=CPPFLAGS="   \

    " --gcc-options="-Wall     -Werror=unused-but-set-variable -Wno-error=inline -iquote /home/slyfox/dev/git/ghc/libraries/integer-gmp   \

    " --with-gcc="x86_64-gentoo-linux-musl-gcc" --with-ld="x86_64-gentoo-linux-musl-ld.gold" --with-ar="x86_64-gentoo-linux-musl-ar" \

    --with-alex="/usr/bin/alex" --with-happy="/usr/bin/happy"

Configuring integer-gmp-1.0.2.0...

configure: WARNING: unrecognized options: --with-compiler

checking build system type... x86_64-pc-linux-gnu

checking host system type... x86_64-pc-linux-gnu

checking target system type... x86_64-pc-linux-gnu

checking for gcc... /usr/lib/ccache/bin/x86_64-gentoo-linux-musl-gcc

checking whether the C compiler works... yes

checking for C compiler default output file name... a.out

checking for suffix of executables...

checking whether we are cross compiling... configure: error: in `/home/slyfox/dev/git/ghc/libraries/integer-gmp/dist-install/build':

configure: error: cannot run C compiled programs.

If you meant to cross compile, use `--host'.

See `config.log' for more details

make[1]: *** [libraries/integer-gmp/ghc.mk:5: libraries/integer-gmp/dist-install/package-data.mk] Error 1

make: *** [Makefile:126: all] Error 2

```

Note: here `ghc-stage1` is assumed to target `musl` target but is passed

`glibc` toolchain. It happens because initial ./configure phase did not

detect host/target as different.

Signed-off-by: Sergei Trofimovich <slyfox@gentoo.org>

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/5f7cb423d7703788014b675af0cbbd611d19310b">5f7cb423</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2019-12-02T23:59:29-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add `timesInt2#` primop

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/fbbe18a274cd094a0582396e83619378a7071386">fbbe18a2</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2019-12-02T23:59:29-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Use the new timesInt2# primop in integer-gmp (#9431)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/5a4b8d0cf2ff83d1a04826b9624fffec7b9a5683">5a4b8d0c</a></strong>

<div>

<span>by Athas</span>

<i>at 2019-12-03T00:00:09-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Document RTS behaviour upon encountering '--'.</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/705a16df02411ec2445c9a254396a93cabe559ef">705a16df</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-03T07:11:33-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Make BCO# lifted

In #17424 Simon PJ noted that there is a potentially unsafe occurrence

of unsafeCoerce#, coercing from an unlifted to lifted type. However,

nowhere in the compiler do we assume that a BCO# is not a thunk.

Moreover, in the case of a CAF the result returned by `createBCO` *will*

be a thunk (as noted in [Updatable CAF BCOs]).  Consequently it seems

better to rather make BCO# a lifted type and rename it to BCO.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/35afe4f3b1591740ab8454ebe4c0ec206d2b7e14">35afe4f3</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2019-12-03T07:12:13-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Use Int# primops in `Bits Int{8,16,32,64}` instances

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/7a51b587ad6a154f89542641593c2b5b2c3f3dde">7a51b587</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2019-12-03T07:12:13-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add constant folding rule (#16402)

    narrowN (x .&. m)

    m .&. (2^N-1) = 2^N-1

    ==> narrowN x

e.g.  narrow16 (x .&. 0x12FFFF) ==> narrow16 x

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/10caee7fd3f048ead668e8bc70d855d9a55f89a5">10caee7f</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-03T21:04:50-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">users-guide: Add 8.12.1 release notes

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/25019d18109cd620a2cf6ab0e7d417d14935e8a5">25019d18</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-03T21:04:50-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Drop Uniquable constraint for AnnTarget

This relied on deriveUnique, which was far too subtle to be safely

applied. Thankfully the instance doesn't appear to be used so let's just

drop it.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/78b67ad0e891fc3b66df72643fb173dc985e8306">78b67ad0</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-03T21:04:50-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Simplify uniqAway

This does two things:

 * Eliminate all uses of Unique.deriveUnique, which was quite easy to

   mis-use and extremely subtle.

 * Rename the previous "derived unique" notion to "local unique". This

   is possible because the only places where `uniqAway` can be safely

   used are those where local uniqueness (with respect to some

   InScopeSet) is sufficient.

 * Rework the implementation of VarEnv.uniqAway, as discussed in #17462.

   This should make the operation significantly more efficient than its

   previous iterative implementation..

Metric Decrease:

    T9872c

    T12227

    T9233

    T14683

    T5030

    T12545

    hie002

Metric Increase:

    T9961

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/f03a41d4bf9418ee028ecb51654c928b2da74edd">f03a41d4</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-03T21:05:27-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Elf: Fix link info note generation

Previously we would use the `.int` assembler directive to generate

32-bit words in the note section. However, `.int` is note guaranteed to

produce 4-bytes; in fact, on some platforms (e.g. AArch64) it produces

8-bytes. Use the `.4bytes` directive to avoid this.

Moreover, we used the `.align` directive, which is quite platform

dependent. On AArch64 it appears to not even be idempotent (despite what

the documentation claims). `.balign` is consequentially preferred as it

offers consistent behavior across platforms.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/84585e5e7c5d729ce38fa47ebaa7518acd14c2f1">84585e5e</a></strong>

<div>

<span>by Vladislav Zavialov</span>

<i>at 2019-12-05T16:07:44-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Meaning-preserving SCC annotations (#15730)

This patch implements GHC Proposal #176:

  https://github.com/ghc-proposals/ghc-proposals/blob/master/proposals/0176-scc-parsing.rst

Before the change:

  1 /                    2 / 2 = 0.25

  1 / {-# SCC "name" #-} 2 / 2 = 1.0

After the change:

  1 /                    2 / 2 = 0.25

  1 / {-# SCC "name" #-} 2 / 2 = parse error

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/e49e54705dee111b64c061a01b1e193d89cf84ed">e49e5470</a></strong>

<div>

<span>by Vladislav Zavialov</span>

<i>at 2019-12-05T16:07:44-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Improve error messages for SCC pragmas

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/a2b535d9ef8864b2d314c739a0e788a6df911aaf">a2b535d9</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-05T16:07:45-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">users guide: Try to silence underfull \hbox warnings

We use two tricks, as suggested here [1]:

 * Use microtype to try to reduce the incidence of underfull boxes

 * Bump up \hbadness to eliminate the warnings

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/4e47217f85b2006f03e72d0a550fe2cca26cfd93">4e47217f</a></strong>

<div>

<span>by Bodigrim</span>

<i>at 2019-12-05T16:07:47-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Make sameNat and sameSymbol proxy-polymorphic

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/8324f0b7357c428f505dccbc84bb7dde897b509c">8324f0b7</a></strong>

<div>

<span>by Bodigrim</span>

<i>at 2019-12-05T16:07:47-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Test proxy-polymorphic sameNat and sameSymbol

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/69001f54279a55bc4e5e5883c675e5ba6fe49a22">69001f54</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-05T16:07:48-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">nonmoving: Clear segment bitmaps during sweep

Previously we would clear the bitmaps of segments which we are going to

sweep during the preparatory pause. However, this is unnecessary: the

existence of the mark epoch ensures that the sweep will correctly

identify non-reachable objects, even if we do not clear the bitmap.

We now defer clearing the bitmap to sweep, which happens concurrently

with mutation.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/58a9c4295b9f457ca465415b7017f8e343d54339">58a9c429</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-05T16:07:48-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Disable divByZero on non-NCG targets

The LLVM backend does not guarantee any particular semantics for

division by zero, making this test unreliable across platforms.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/8280bd8ab7a20a0a78ca53a06279cf8f07864eb5">8280bd8a</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-05T16:07:49-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Factor out terminal coloring

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/92a52aaaf48420335c4beda45cf40388d305c3f2">92a52aaa</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-05T16:07:49-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Make performance metric summary more readable

Along with some refactoring.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/c4ca29c796fa86ad9d5cd4dfa1a5cdd4e0565fb0">c4ca29c7</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-05T16:07:49-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Use colors more consistently

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/3354c68ec6c90bbccc0f361aa7973eeb75ea229c">3354c68e</a></strong>

<div>

<span>by Vladislav Zavialov</span>

<i>at 2019-12-05T16:07:49-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Pretty-printing of the * kind

Before this patch, GHC always printed the * kind unparenthesized.

This led to two issues:

1. Sometimes GHC printed invalid or incorrect code.

   For example, GHC would print:  type F @*   x = x

         when it meant to print:  type F @(*) x = x

   In the former case, instead of a kind application we were getting a

   type operator (@*).

2. Sometimes GHC printed kinds that were correct but hard to read.

   Should  Either * Int  be read as  Either (*) Int

                              or as  (*) Either Int  ?

   This depends on whether -XStarIsType is enabled, but it would be

   easier if we didn't have to check for the flag when reading the code.

We can solve both problems by assigning (*) a different precedence. Note

that Haskell98 kinds are not affected:

  ((* -> *) -> *) -> *  does NOT become  (((*) -> (*)) -> (*)) -> (*)

The parentheses are added when (*) is used in a function argument

position:

   F * * *   becomes  F (*) (*) (*)

   F A * B   becomes  F A (*) B

   Proxy *   becomes  Proxy (*)

   a * -> *  becomes  a (*) -> *

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/70dd0e4b7051e674479729916285b12fc14c696f">70dd0e4b</a></strong>

<div>

<span>by Vladislav Zavialov</span>

<i>at 2019-12-05T16:07:49-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Parenthesize the * kind in TH.Ppr

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/a7a4efbff1badf7daa41256cb506915e4fddee5c">a7a4efbf</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-05T16:07:49-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">rts/NonMovingSweep: Fix locking of new mutable list allocation

Previously we used allocBlockOnNode_sync in nonmovingSweepMutLists

despite the fact that we aren't in the GC and therefore the allocation

spinlock isn't in use. This meant that sweep would end up spinning until

the next minor GC, when the SM lock was moved away from the SM_MUTEX to

the spinlock. This isn't a correctness issue but it sure isn't good for

performance.

Found thanks for Ward.

Fixes #17539.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/f171b3582d44746bf8b08897a3b23bc97e5dbdda">f171b358</a></strong>

<div>

<span>by Matthias Braun</span>

<i>at 2019-12-05T16:07:51-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix typo in documentation of Base.hs.</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/9897e8c8ef0b19a9571ef97a1d9bb050c1ee9121">9897e8c8</a></strong>

<div>

<span>by Gabor Greif</span>

<i>at 2019-12-06T21:20:38-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Implement pointer tagging for big families (#14373)

Formerly we punted on these and evaluated constructors always got a tag

of 1.

We now cascade switches because we have to check the tag first and when

it is MAX_PTR_TAG then get the precise tag from the info table and

switch on that. The only technically tricky part is that the default

case needs (logical) duplication. To do this we emit an extra label for

it and branch to that from the second switch. This avoids duplicated

codegen.

Here's a simple example of the new code gen:

    data D = D1 | D2 | D3 | D4 | D5 | D6 | D7 | D8

On a 64-bit system previously all constructors would be tagged 1. With

the new code gen D7 and D8 are tagged 7:

    [Lib.D7_con_entry() {

         ...

         {offset

           c1eu: // global

               R1 = R1 + 7;

               call (P64[Sp])(R1) args: 8, res: 0, upd: 8;

         }

     }]

    [Lib.D8_con_entry() {

         ...

         {offset

           c1ez: // global

               R1 = R1 + 7;

               call (P64[Sp])(R1) args: 8, res: 0, upd: 8;

         }

     }]

When switching we now look at the info table only when the tag is 7. For

example, if we derive Enum for the type above, the Cmm looks like this:

    c2Le:

        _s2Js::P64 = R1;

        _c2Lq::P64 = _s2Js::P64 & 7;

        switch [1 .. 7] _c2Lq::P64 {

            case 1 : goto c2Lk;

            case 2 : goto c2Ll;

            case 3 : goto c2Lm;

            case 4 : goto c2Ln;

            case 5 : goto c2Lo;

            case 6 : goto c2Lp;

            case 7 : goto c2Lj;

        }

    // Read info table for tag

    c2Lj:

        _c2Lv::I64 = %MO_UU_Conv_W32_W64(I32[I64[_s2Js::P64 & (-8)] - 4]);

        if (_c2Lv::I64 != 6) goto c2Lu; else goto c2Lt;

Generated Cmm sizes do not change too much, but binaries are very

slightly larger, due to the fact that the new instructions are longer in

encoded form. E.g. previously entry code for D8 above would be

    00000000000001c0 <Lib_D8_con_info>:

     1c0:       48 ff c3                inc    %rbx

     1c3:       ff 65 00                jmpq   *0x0(%rbp)

With this patch

    00000000000001d0 <Lib_D8_con_info>:

     1d0:       48 83 c3 07             add    $0x7,%rbx

     1d4:       ff 65 00                jmpq   *0x0(%rbp)

This is one byte longer.

Secondly, reading info table directly and then switching is shorter

    _c1co:

            movq -1(%rbx),%rax

            movl -4(%rax),%eax

            // Switch on info table tag

            jmp *_n1d5(,%rax,8)

than doing the same switch, and then for the tag 7 doing another switch:

    // When tag is 7

    _c1ct:

            andq $-8,%rbx

            movq (%rbx),%rax

            movl -4(%rax),%eax

            // Switch on info table tag

            ...

Some changes of binary sizes in actual programs:

- In NoFib the worst case is 0.1% increase in benchmark "parser" (see

  NoFib results below). All programs get slightly larger.

- Stage 2 compiler size does not change.

- In "containers" (the library) size of all object files increases

  0.0005%. Size of the test program "bitqueue-properties" increases

  0.03%.

nofib benchmarks kindly provided by Ömer (@osa1):

NoFib Results

=============

--------------------------------------------------------------------------------

        Program           Size    Allocs    Instrs     Reads    Writes

--------------------------------------------------------------------------------

             CS          +0.0%      0.0%     -0.0%     -0.0%     -0.0%

            CSD          +0.0%      0.0%      0.0%     +0.0%     +0.0%

             FS          +0.0%      0.0%      0.0%     +0.0%      0.0%

              S          +0.0%      0.0%     -0.0%      0.0%      0.0%

             VS          +0.0%      0.0%     -0.0%     +0.0%     +0.0%

            VSD          +0.0%      0.0%     -0.0%     +0.0%     -0.0%

            VSM          +0.0%      0.0%      0.0%      0.0%      0.0%

           anna          +0.0%      0.0%     +0.1%     -0.9%     -0.0%

           ansi          +0.0%      0.0%     -0.0%     +0.0%     +0.0%

           atom          +0.0%      0.0%      0.0%      0.0%      0.0%

         awards          +0.0%      0.0%     -0.0%     +0.0%      0.0%

         banner          +0.0%      0.0%     -0.0%     +0.0%      0.0%

     bernouilli          +0.0%      0.0%     +0.0%     +0.0%     +0.0%

   binary-trees          +0.0%      0.0%     -0.0%     -0.0%     -0.0%

          boyer          +0.0%      0.0%     +0.0%      0.0%     -0.0%

         boyer2          +0.0%      0.0%     +0.0%      0.0%     -0.0%

           bspt          +0.0%      0.0%     +0.0%     +0.0%      0.0%

      cacheprof          +0.0%      0.0%     +0.1%     -0.8%      0.0%

       calendar          +0.0%      0.0%     -0.0%     +0.0%     -0.0%

       cichelli          +0.0%      0.0%     +0.0%      0.0%      0.0%

        circsim          +0.0%      0.0%     -0.0%     -0.1%     -0.0%

       clausify          +0.0%      0.0%     +0.0%     +0.0%      0.0%

  comp_lab_zift          +0.0%      0.0%     +0.0%      0.0%     -0.0%

       compress          +0.0%      0.0%     +0.0%     +0.0%      0.0%

      compress2          +0.0%      0.0%      0.0%      0.0%      0.0%

    constraints          +0.0%      0.0%     -0.0%     -0.0%     -0.0%

   cryptarithm1          +0.0%      0.0%     +0.0%      0.0%      0.0%

   cryptarithm2          +0.0%      0.0%     +0.0%     -0.0%      0.0%

            cse          +0.0%      0.0%     +0.0%     +0.0%      0.0%

   digits-of-e1          +0.0%      0.0%     -0.0%     -0.0%     -0.0%

   digits-of-e2          +0.0%      0.0%     +0.0%     -0.0%     -0.0%

         dom-lt          +0.0%      0.0%     +0.0%     +0.0%      0.0%

          eliza          +0.0%      0.0%     -0.0%     +0.0%      0.0%

          event          +0.0%      0.0%     -0.0%     -0.0%     -0.0%

    exact-reals          +0.0%      0.0%     +0.0%     +0.0%     +0.0%

         exp3_8          +0.0%      0.0%     -0.0%     -0.0%     -0.0%

         expert          +0.0%      0.0%     +0.0%     +0.0%     +0.0%

 fannkuch-redux          +0.0%      0.0%     +0.0%      0.0%      0.0%

          fasta          +0.0%      0.0%     -0.0%     -0.0%     -0.0%

            fem          +0.0%      0.0%     +0.0%     +0.0%     +0.0%

            fft          +0.0%      0.0%     +0.0%     -0.0%     -0.0%

           fft2          +0.0%      0.0%     +0.0%     +0.0%     +0.0%

       fibheaps          +0.0%      0.0%     +0.0%     +0.0%      0.0%

           fish          +0.0%      0.0%     +0.0%     +0.0%      0.0%

          fluid          +0.0%      0.0%     +0.0%     +0.0%     +0.0%

         fulsom          +0.0%      0.0%     +0.0%     -0.0%     +0.0%

         gamteb          +0.0%      0.0%     +0.0%     -0.0%     -0.0%

            gcd          +0.0%      0.0%     +0.0%     +0.0%      0.0%

    gen_regexps          +0.0%      0.0%     +0.0%     -0.0%     -0.0%

         genfft          +0.0%      0.0%     -0.0%     -0.0%     -0.0%

             gg          +0.0%      0.0%      0.0%     -0.0%      0.0%

           grep          +0.0%      0.0%     +0.0%     +0.0%     +0.0%

         hidden          +0.0%      0.0%     +0.0%     -0.0%     -0.0%

            hpg          +0.0%      0.0%     +0.0%     -0.1%     -0.0%

            ida          +0.0%      0.0%     +0.0%     -0.0%     -0.0%

          infer          +0.0%      0.0%     -0.0%     -0.0%     -0.0%

        integer          +0.0%      0.0%     -0.0%     -0.0%     -0.0%

      integrate          +0.0%      0.0%      0.0%     +0.0%      0.0%

   k-nucleotide          +0.0%      0.0%     -0.0%     -0.0%     -0.0%

          kahan          +0.0%      0.0%     -0.0%     -0.0%     -0.0%

        knights          +0.0%      0.0%     +0.0%     -0.0%     -0.0%

         lambda          +0.0%      0.0%     +1.2%     -6.1%     -0.0%

     last-piece          +0.0%      0.0%     +0.0%     -0.0%     -0.0%

           lcss          +0.0%      0.0%     +0.0%     -0.0%     -0.0%

           life          +0.0%      0.0%     +0.0%     -0.0%     -0.0%

           lift          +0.0%      0.0%     +0.0%     +0.0%      0.0%

         linear          +0.0%      0.0%     +0.0%     +0.0%     +0.0%

      listcompr          +0.0%      0.0%     -0.0%     -0.0%     -0.0%

       listcopy          +0.0%      0.0%     -0.0%     -0.0%     -0.0%

       maillist          +0.0%      0.0%     +0.0%     -0.0%     -0.0%

         mandel          +0.0%      0.0%     +0.0%     +0.0%     +0.0%

        mandel2          +0.0%      0.0%     +0.0%     +0.0%     -0.0%

           mate          +0.0%      0.0%     +0.0%     +0.0%     +0.0%

        minimax          +0.0%      0.0%     -0.0%     +0.0%     -0.0%

        mkhprog          +0.0%      0.0%     +0.0%     +0.0%     +0.0%

     multiplier          +0.0%      0.0%      0.0%     +0.0%     -0.0%

         n-body          +0.0%      0.0%     +0.0%     -0.0%     -0.0%

       nucleic2          +0.0%      0.0%     +0.0%     +0.0%     -0.0%

           para          +0.0%      0.0%     +0.0%     +0.0%     +0.0%

      paraffins          +0.0%      0.0%     +0.0%     +0.0%     +0.0%

         parser          +0.1%      0.0%     +0.4%     -1.7%     -0.0%

        parstof          +0.0%      0.0%     -0.0%     -0.0%     -0.0%

            pic          +0.0%      0.0%     +0.0%      0.0%     -0.0%

       pidigits          +0.0%      0.0%     -0.0%     -0.0%     -0.0%

          power          +0.0%      0.0%     +0.0%     -0.0%     -0.0%

         pretty          +0.0%      0.0%     +0.0%     +0.0%     +0.0%

         primes          +0.0%      0.0%     +0.0%      0.0%      0.0%

      primetest          +0.0%      0.0%     +0.0%     +0.0%     +0.0%

         prolog          +0.0%      0.0%     +0.0%     +0.0%     +0.0%

         puzzle          +0.0%      0.0%     +0.0%     +0.0%     +0.0%

         queens          +0.0%      0.0%      0.0%     +0.0%     +0.0%

        reptile          +0.0%      0.0%     +0.0%     +0.0%      0.0%

reverse-complem          +0.0%      0.0%     -0.0%     -0.0%     -0.0%

        rewrite          +0.0%      0.0%     +0.0%      0.0%     -0.0%

           rfib          +0.0%      0.0%     +0.0%     +0.0%     +0.0%

            rsa          +0.0%      0.0%     +0.0%     +0.0%     +0.0%

            scc          +0.0%      0.0%     +0.0%     +0.0%     +0.0%

          sched          +0.0%      0.0%     +0.0%     +0.0%     +0.0%

            scs          +0.0%      0.0%     +0.0%     +0.0%      0.0%

         simple          +0.0%      0.0%     +0.0%     +0.0%     +0.0%

          solid          +0.0%      0.0%     +0.0%     +0.0%      0.0%

        sorting          +0.0%      0.0%     +0.0%     -0.0%      0.0%

  spectral-norm          +0.0%      0.0%     -0.0%     -0.0%     -0.0%

         sphere          +0.0%      0.0%     +0.0%     -1.0%      0.0%

         symalg          +0.0%      0.0%     +0.0%     +0.0%     +0.0%

            tak          +0.0%      0.0%     +0.0%     +0.0%     +0.0%

      transform          +0.0%      0.0%     +0.4%     -1.3%     +0.0%

       treejoin          +0.0%      0.0%     +0.0%     -0.0%      0.0%

      typecheck          +0.0%      0.0%     -0.0%     +0.0%      0.0%

        veritas          +0.0%      0.0%     +0.0%     -0.1%     +0.0%

           wang          +0.0%      0.0%     +0.0%     +0.0%     +0.0%

      wave4main          +0.0%      0.0%     +0.0%      0.0%     -0.0%

   wheel-sieve1          +0.0%      0.0%     +0.0%     +0.0%     +0.0%

   wheel-sieve2          +0.0%      0.0%     +0.0%     +0.0%      0.0%

           x2n1          +0.0%      0.0%     +0.0%     +0.0%      0.0%

--------------------------------------------------------------------------------

            Min          +0.0%      0.0%     -0.0%     -6.1%     -0.0%

            Max          +0.1%      0.0%     +1.2%     +0.0%     +0.0%

 Geometric Mean          +0.0%     -0.0%     +0.0%     -0.1%     -0.0%

NoFib GC Results

================

--------------------------------------------------------------------------------

        Program           Size    Allocs    Instrs     Reads    Writes

--------------------------------------------------------------------------------

        circsim          +0.0%      0.0%     -0.0%     -0.0%     -0.0%

    constraints          +0.0%      0.0%     -0.0%      0.0%     -0.0%

       fibheaps          +0.0%      0.0%      0.0%     -0.0%     -0.0%

         fulsom          +0.0%      0.0%      0.0%     -0.6%     -0.0%

       gc_bench          +0.0%      0.0%      0.0%      0.0%     -0.0%

           hash          +0.0%      0.0%     -0.0%     -0.0%     -0.0%

           lcss          +0.0%      0.0%      0.0%     -0.0%      0.0%

      mutstore1          +0.0%      0.0%      0.0%     -0.0%     -0.0%

      mutstore2          +0.0%      0.0%     +0.0%     -0.0%     -0.0%

          power          +0.0%      0.0%     -0.0%      0.0%     -0.0%

     spellcheck          +0.0%      0.0%     -0.0%     -0.0%     -0.0%

--------------------------------------------------------------------------------

            Min          +0.0%      0.0%     -0.0%     -0.6%     -0.0%

            Max          +0.0%      0.0%     +0.0%      0.0%      0.0%

 Geometric Mean          +0.0%     +0.0%     +0.0%     -0.1%     +0.0%

Fixes #14373

These performance regressions appear to be a fluke in CI. See the

discussion in !1742 for details.

Metric Increase:

    T6048

    T12234

    T12425

    Naperian

    T12150

    T5837

    T13035

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/ee07421fcf99189de6506cf8d17185ed540ea2b3">ee07421f</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2019-12-06T21:21:14-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Work in progress on coercionLKind, coercionRKind

This is a preliminary patch for #17515

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/0a4ca9eb152c6bfbc3aad71c180a38bbfeca5bfb">0a4ca9eb</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2019-12-06T21:21:14-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Split up coercionKind

This patch implements the idea in #17515, splitting `coercionKind` into:

 * `coercion{Left,Right}Kind`, which computes the left/right side of the

    pair

 * `coercionKind`, which computes the pair of coercible types

This is reduces allocation since we frequently only need only one side

of the pair. Specifically, we see the following improvements on x86-64

Debian 9:

| test     | new        | old           | relative chg. |

| :------- | ---------: | ------------: | ------------: |

| T5030    | 695537752  | 747641152.0   | -6.97%        |

| T5321Fun | 449315744  | 474009040.0   | -5.21%        |

| T9872a   | 2611071400 | 2645040952.0  | -1.28%        |

| T9872c   | 2957097904 | 2994260264.0  | -1.24%        |

| T12227   | 773435072  | 812367768.0   | -4.79%        |

| T12545   | 3142687224 | 3215714752.0  | -2.27%        |

| T14683   | 9392407664 | 9824775000.0  | -4.40%        |

Metric Decrease:

    T12545

    T9872a

    T14683

    T5030

    T12227

    T9872c

    T5321Fun

    T9872b

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/d46a72e19e1b508358827e7270139f3273915697">d46a72e1</a></strong>

<div>

<span>by Gabor Greif</span>

<i>at 2019-12-09T12:05:15-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix comment typos

The below is only necessary to fix the CI perf fluke that

happened in 9897e8c8ef0b19a9571ef97a1d9bb050c1ee9121:

-------------------------

Metric Decrease:

    T5837

    T6048

    T9020

    T12425

    T12234

    T13035

    T12150

    Naperian

-------------------------

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/e3bba7e4b221139090edf54a18c7e0e1ec3f9be8">e3bba7e4</a></strong>

<div>

<span>by Micha Wiedenmann</span>

<i>at 2019-12-10T19:52:44-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">users guide: Motivation of DefaultSignatures

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/843ceb3892432b28628e4b4bad2c712ca371d02b">843ceb38</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-10T19:53:54-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">rts: Add a long form flag to enable the non-moving GC

The old flag, `-xn`, was quite cryptic. Here we add `--nonmoving-gc` in

addition.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/921d32381a0bdd693935aa2bada3cd76870bf2f9">921d3238</a></strong>

<div>

<span>by Ryan Scott</span>

<i>at 2019-12-10T19:54:34-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Ignore unary constraint tuples during typechecking (#17511)

We deliberately avoid defining a magical `Unit%` class, for reasons

that I have expounded upon in the newly added

`Note [Ignore unary constraint tuples]` in `TcHsType`. However, a

sneaky user could try to insert `Unit%` into their program by way of

Template Haskell, leading to the interface-file error observed

in #17511. To avoid this, any time we encounter a unary constraint

tuple during typechecking, we drop the surrounding constraint tuple

application. This is safe to do since `Unit% a` and `a` would be

semantically equivalent (unlike other forms of unary tuples).

Fixes #17511.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/436ec9f3b31f9bc7ccf8c685afcda9fc424e45e9">436ec9f3</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-10T19:55:37-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Move changelog linting logic to shell script

Allowing it to be easily used locally.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/2f6b434fc96b1f48315e103f4a77fecc2da89471">2f6b434f</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-10T19:55:37-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Move changelog linting logic to shell script

Allowing it to be easily used locally.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/7a5a6e0718144786c4e0eb7395ede271e57ff4b7">7a5a6e07</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-10T19:56:25-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">base: Fix incorrect @since in GHC.Natural

Fixes #17547.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/2bbfaf8abac0854a5b7f5485aa830fec669fa602">2bbfaf8a</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-10T19:57:01-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">hadrian: AArch64 supports the GHCi interpreter and SMP

I'm not sure how this was omitted from the list of supported

architectures.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/8f1ceb67331cfbe075c484c8d1d6b6c9f6fc1e92">8f1ceb67</a></strong>

<div>

<span>by John Ericson</span>

<i>at 2019-12-10T19:57:39-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Move Int# section of primops.txt.pp

This matches the organization of the fixed-sized ones, and keeps each

Int* next to its corresponding Word*.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/7a823b0ffcbd5b5c4c03334e72c8a36460b62ff7">7a823b0f</a></strong>

<div>

<span>by John Ericson</span>

<i>at 2019-12-10T19:57:39-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Move Int64# and Word64# sections of primops.txt.pp

This way it is next to the other fixed-sized ones.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/8dd9929ad29fd200405d068463af64dafff6b402">8dd9929a</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-10T19:58:19-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Add (broken) test for #17510

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/6e47a76a3d0a7b3d424442914478de579a49363c">6e47a76a</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-10T19:58:59-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Re-layout validate script

This script was previously a whitespace nightmare.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/f80c4a66ae219afa7bd4172441f4e94ba649c9d9">f80c4a66</a></strong>

<div>

<span>by Crazycolorz5</span>

<i>at 2019-12-11T14:12:17-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">rts: Specialize hashing at call site rather than in struct.

Separate word and string hash tables on the type level, and do not store

the hashing function.  Thus when a different hash function is desire it

is provided upon accessing the table. This is worst case the same as

before the change, and in the majority of cases is better. Also mark the

functions for aggressive inlining to improve performance.  {F1686506}

Reviewers: bgamari, erikd, simonmar

Subscribers: rwbarton, thomie, carter

GHC Trac Issues: #13165

Differential Revision: https://phabricator.haskell.org/D4889

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/2d1b9619e56e1a83fba9a630e7735adb8ac7c16a">2d1b9619</a></strong>

<div>

<span>by Richard Eisenberg</span>

<i>at 2019-12-11T14:12:55-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Warn on inferred polymorphic recursion

Silly users sometimes try to use visible dependent quantification

and polymorphic recursion without a CUSK or SAK. This causes

unexpected errors. So we now adjust expectations with a bit

of helpful messaging.

Closes #17541 and closes #17131.

test cases: dependent/should_fail/T{17541{,b},17131}

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/4dde485eddacc7a38c47ff1054234e2896c214d8">4dde485e</a></strong>

<div>

<span>by Oleg Grenrus</span>

<i>at 2019-12-12T02:24:46-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add --show-unit-ids flag to ghc-pkg

I only added it into --simple-output and ghc-pkg check output;

there are probably other places where it can be adopted.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/e6e1ec08e79950c8cadd1f7a2e01292d6142fbfc">e6e1ec08</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-12T02:25:33-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Simplify and clarify performance test baseline search

The previous implementation was extremely complicated, seemingly to

allow the local and CI namespaces to be searched incrementally. However,

it's quite unclear why this is needed and moreover the implementation

seems to have had quadratic runtime cost in the search depth(!).

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/29c4609c58005aa4dad887b7396f0e95dfe0c68d">29c4609c</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-12T02:26:19-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Add test for #17549

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/9f0ee2534fcfd95b0870d09718ac3c31704f5753">9f0ee253</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-12T02:26:56-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Move -dwarf and -debug jobs to full-build stage

This sacrifices some precision in favor of improving parallelism.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/7179b968c709eb0f5413c8e3847a8a593c596185">7179b968</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-12T02:27:34-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Revert "rts: Drop redundant flags for libffi"

This seems to have regressed builds using `--with-system-libffi`

(#17520).

This reverts commit 3ce18700f80a12c48a029b49c6201ad2410071bb.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/cc7d56506cb0ab49c385498a2139954397e3c782">cc7d5650</a></strong>

<div>

<span>by Oleg Grenrus</span>

<i>at 2019-12-16T10:20:56+02:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Having no shake upper bound is irresposible

Given that shake is far from "done" API wise,

and is central component to the build system.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/9431f905a1eabcb823b65460560558c08fb55cd9">9431f905</a></strong>

<div>

<span>by Oleg Grenrus</span>

<i>at 2019-12-16T10:55:50+02:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add index-state to hadrian/cabal.project

Then one is freer to omit upper bounds, as we won't pick

any new entries on Hackage while building hadrian itself.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/3e17a866fecebc5f80b4e7da93a73803b86499ca">3e17a866</a></strong>

<div>

<span>by Krzysztof Gogolewski</span>

<i>at 2019-12-16T19:31:44-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Remove dataConSig

As suggested in #17291

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/75355fdef61da44a395ee9bfa2b9dca0eecea58a">75355fde</a></strong>

<div>

<span>by Krzysztof Gogolewski</span>

<i>at 2019-12-16T19:31:44-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Use "OrCoVar" functions less

As described in #17291, we'd like to separate coercions and expressions

in a more robust fashion.

This is a small step in this direction.

- `mkLocalId` now panicks on a covar.

  Calls where this was not the case were changed to `mkLocalIdOrCoVar`.

- Don't use "OrCoVar" functions in places where we know the type is

  not a coercion.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/f9686e132a7fe5bbe517a4748fef6094fe74d43d">f9686e13</a></strong>

<div>

<span>by Richard Eisenberg</span>

<i>at 2019-12-16T19:32:21-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Do more validity checks for quantified constraints

Close #17583.

Test case: typecheck/should_fail/T17563

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/af7637651bbc55db81e8cef387faf93951d831c2">af763765</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-16T19:33:01-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Fix Windows artifact collection

Variable interpolation in gitlab-ci.yml apparently doesn't work. Sigh.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/e6d4b902fe2072c4cf75cfc2ec3d9f595938b3a1">e6d4b902</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-16T19:33:01-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Use xz --threads on Debian 10

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/8ba650e9f50724d5bbec6cdd7e1d462cf8b7895a">8ba650e9</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-16T19:33:01-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Allow debian 8 build to fail

The python release shipped with deb8 (3.3) is too old for our testsuite

driver.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/ac25a3f637d894e9b972600e8942a91570e2dcd0">ac25a3f6</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-16T19:33:01-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Use xz --threads on Alpine

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/cc6280885f15c00703c4fc8305f6d47747910fc2">cc628088</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-16T19:33:01-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Another approach for xz detection

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/37d788ab42703682ed22642d1a9f97351a12c7cd">37d788ab</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-16T19:33:01-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Re-add release-x86_64-deb9 job

Also eliminate some redundancy.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/f82791382806dcaa0ea2241d97b17e1d3369c0b6">f8279138</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-16T19:33:01-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Drop redundant release-x86_64-linux-deb9 job

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/8148ff06d13edc99e6b00c45554f61279cfdee2d">8148ff06</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-17T07:24:40-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Mark cgrun057 as broken on ARMv7

Due to #17554. It's very surprising that this only occurs on ARMv7 but

this is the only place I've seen this failure thusfar.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/85e5696d1e244a88053675280c9c28669246983f">85e5696d</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-17T07:24:40-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Mark prog001 as fragile on ARMv7

Due to #17555.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/a5f0aab05866ff178952cf023e7b0e18b5db9562">a5f0aab0</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-17T07:24:40-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Mark T10272 as broken on ARMv7

Due to #17556.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/1e6827c6e4e083eafccabb7a90a38d9d1262bbed">1e6827c6</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-17T07:24:40-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Mark T13825-debugger as broken on ARMv7

Due to #17557.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/7cef0b7d707c0b242ee788919f76a73d1a3d2c96">7cef0b7d</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-17T07:24:40-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Mark T14028 as broken on ARMv7

Due to #17558.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/6ea4eb4babf090e447277a3e4b0dea31d79e9acf">6ea4eb4b</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-17T07:24:40-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Make ghc_built_by_llvm check more precise

Previously it would hackily look at the flavour name to determine

whether LLVM was used to build stage2 ghc. However, this didn't work at

all with Hadrian and would miss cases like ARM where we use the LLVM

backend by default.

See #16087 for the motivation for why ghc_built_by_llvm is needed at

all. This should catch one of the ARMv7 failures described in #17555.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/c3e82bf7d74b96f2efc10a2a83c52395aaee400d">c3e82bf7</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-17T07:24:40-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Mark T5435_* tests as broken on ARM

`T5435_v_asm_a`, `T5435_v_asm_b`, and `T5435_v_gcc` all fail on ARMv7.

See #17559.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/eb2aa851bc22f5445963b909b2105009a7abb949">eb2aa851</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-17T07:24:40-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Don't allow armv7 jobs to fail

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/efc92216f0f961dd86f1fcbc20fd79b4af41e95b">efc92216</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-17T07:24:40-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Revert "testsuite: Mark cgrun057 as broken on ARMv7"

This reverts commit 6cfc47ec8a478e1751cb3e7338954da1853c3996.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/1d2bb9eb828f8daac552a9494746b6d6af60b515">1d2bb9eb</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-17T07:24:40-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Mark print002 as fragile on ARM

Due to #17557.

Also accepting spurious performance change.

Metric Decrease:

    T1969

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/41f4e4fbe637d22bfded6f1597af57337ae5cf15">41f4e4fb</a></strong>

<div>

<span>by Josh Meredith</span>

<i>at 2019-12-17T07:25:17-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix ambiguous occurence error when building Hadrian

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/4374983a60d533768d7cc8a3b29a6e9a2e6babdb">4374983a</a></strong>

<div>

<span>by Josh Meredith</span>

<i>at 2019-12-17T07:25:17-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Rename SphinxMode constructors

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/a8f7ecd54821493dc061c55ceebb7e271b17056e">a8f7ecd5</a></strong>

<div>

<span>by Josh Meredith</span>

<i>at 2019-12-17T07:25:17-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Use *Mode suffix instead of *M

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/58655b9da7599135395417a042f53cfa13b2151d">58655b9d</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2019-12-18T13:43:37+01:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add GHC-API logging hooks

* Add 'dumpAction' hook to DynFlags.

It allows GHC API users to catch dumped intermediate codes and

information. The format of the dump (Core, Stg, raw text, etc.) is now

reported allowing easier automatic handling.

* Add 'traceAction' hook to DynFlags.

Some dumps go through the trace mechanism (for instance unfoldings that

have been considered for inlining). This is problematic because:

1) dumps aren't written into files even with -ddump-to-file on

2) dumps are written on stdout even with GHC API

3) in this specific case, dumping depends on unsafe globally stored

DynFlags which is bad for GHC API users

We introduce 'traceAction' hook which allows GHC API to catch those

traces and to avoid using globally stored DynFlags.

* Avoid dumping empty logs via dumpAction/traceAction (but still write

empty files to keep the existing behavior)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/fad866e028e577b55510a3f9a2faf26d6fdd7bce">fad866e0</a></strong>

<div>

<span>by Moritz Kiefer</span>

<i>at 2019-12-19T11:15:39-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Avoid race condition in hDuplicateTo

In our codebase we have some code along the lines of

```

newStdout <- hDuplicate stdout

stderr `hDuplicateTo` stdout

```

to avoid stray `putStrLn`s from corrupting a protocol (LSP) that is

run over stdout.

On CI we have seen a bunch of issues where `dup2` returned `EBUSY` so

this fails with `ResourceExhausted` in Haskell.

I’ve spent some time looking at the docs for `dup2` and the code in

`base` and afaict the following race condition is being triggered

here:

1. The user calls `hDuplicateTo stderr stdout`.

2. `hDuplicateTo` calls `hClose_help stdout_`, this closes the file

handle for stdout.

3. The file handle for stdout is now free, so another thread

allocating a file might get stdout.

4. If `dup2` is called while `stdout` (now pointing to something

else) is half-open, it returns EBUSY.

I think there might actually be an even worse case where `dup2` is run

after FD 1 is fully open again. In that case, you will end up not just

redirecting the original stdout to stderr but also the whatever

resulted in that file handle being allocated.

As far as I can tell, `dup2` takes care of closing the file handle

itself so there is no reason to do this in `hDuplicateTo`. So this PR

replaces the call to `hClose_help` by the only part of `hClose_help`

that we actually care about, namely, `flushWriteBuffer`.

I tested this on our codebase fairly extensively and haven’t been able

to reproduce the issue with this patch.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/0c114c6599c1df93b208c5f2b1754523858d80ee">0c114c65</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2019-12-19T11:16:17-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Handle large ARR_WORDS in heap census (fix #17572)

We can do a heap census with a non-profiling RTS. With a non-profiling

RTS we don't zero superfluous bytes of shrunk arrays hence a need to

handle the case specifically to avoid a crash.

Revert part of a586b33f8e8ad60b5c5ef3501c89e9b71794bbed

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/1a0d1a6583cc39a31d6947eda1d4998c4fb53c4f">1a0d1a65</a></strong>

<div>

<span>by John Ericson</span>

<i>at 2019-12-20T10:50:22-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Deduplicate copied monad failure handler code

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/70e56b272492b65e41a149ec39a939e794fea66b">70e56b27</a></strong>

<div>

<span>by Ryan Scott</span>

<i>at 2019-12-20T10:50:57-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">lookupBindGroupOcc: recommend names in the same namespace (#17593)

Previously, `lookupBindGroupOcc`'s error message would recommend all

similar names in scope, regardless of whether they were type

constructors, data constructors, or functions, leading to the

confusion witnessed in #17593. This is easily fixed by only

recommending names in the same namespace, using the

`nameSpacesRelated` function.

Fixes #17593.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/3c12355ece0fa14e534c71a9f7327eda9773fad9">3c12355e</a></strong>

<div>

<span>by Stefan Schulze Frielinghaus</span>

<i>at 2019-12-24T01:03:44-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix endian handling w.r.t. CPP macro WORDS_BIGENDIAN

Include header file `ghcautoconf.h` where the CPP macro

`WORDS_BIGENDIAN` is defined. This finally fixes #17337 (in conjunction

with commit 6c59cc71dc).

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/11f8eef5a9625c77f78a829406c446b615ed6168">11f8eef5</a></strong>

<div>

<span>by Stefan Schulze Frielinghaus</span>

<i>at 2019-12-24T01:03:44-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">fixup! Fix endian handling w.r.t. CPP macro WORDS_BIGENDIAN

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/40327b037f7115f7b05cc0265acb787671bea294">40327b03</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2019-12-24T01:04:24-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Remove outdated comment

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/aeea92ef6ffa514793b1d37b38aaed3616c5c24a">aeea92ef</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2019-12-25T19:23:54-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Switch to ReadTheDocs theme for the user-guide

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/26493eab8c64776b3133a3c0b3bac33fbd28546c">26493eab</a></strong>

<div>

<span>by Gabor Greif</span>

<i>at 2019-12-25T19:24:32-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix copy-paste error in comment</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/776df719905b2d39845268a87026bfec90b2795f">776df719</a></strong>

<div>

<span>by Gabor Greif</span>

<i>at 2019-12-25T19:24:32-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix comment about minimal gcc version

to be consistent what FP_GCC_VERSION requires</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/3b17114dc360689651f2005126f5dc25510a7263">3b17114d</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2019-12-26T14:09:11-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Minor refactor in ghc.cabal.in:

- Remove outdated comments

- Move cutils.c from parser to cbits

- Remove unused cutils.h

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/334290b6681796dd141c964b88c541da13ce03c7">334290b6</a></strong>

<div>

<span>by Ryan Scott</span>

<i>at 2019-12-26T14:09:48-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Replace panic/notHandled with noExtCon in DsMeta

There are many spots in `DsMeta` where `panic` or `notHandled` is

used after pattern-matching on a TTG extension constructor. This is

overkill, however, as using `noExtCon` would work just as well. This

patch switches out these panics for `noExtCon`.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/68252aa3eb833c9747c59de3a7655a1def1b7a76">68252aa3</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-27T15:11:38-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Skip T17499 when built against integer-simple

Since it routinely times out in CI.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/0c51aeebc79edd9b317109019522b5ee557c92df">0c51aeeb</a></strong>

<div>

<span>by Gabor Greif</span>

<i>at 2019-12-27T15:12:17-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">suppress popup dialog about missing Xcode at configure

tested with `bash` and `zsh`.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/8d76bcc2c0111fdd5ee0ed3e60cdce91ca2172c5">8d76bcc2</a></strong>

<div>

<span>by Gabor Greif</span>

<i>at 2019-12-27T15:12:17-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">while at it rename XCode to the official Xcode

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/47a68205d5781a65504b255f1e8e08b162f0c6eb">47a68205</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-27T15:12:55-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Mark cgrun057 as fragile on ARM

As reported in #17554. Only marking on ARM for now although there is

evidence to suggest that the issue may occur on other platforms as well.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/d03dec8ff7681423b9cafa2ce1b5b5f624e664c7">d03dec8f</a></strong>

<div>

<span>by Gabor Greif</span>

<i>at 2019-12-27T15:13:32-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">use shell variable CcLlvmBackend for test

Previously we used `AC_DEFINE`d variable `CC_LLVM_BACKEND` which has an empty shell expansion.</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/2528e6847d6ab11b452998e5511d60b26453f022">2528e684</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-30T06:51:32-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">driver: Include debug level in the recompilation check hash

Fixes #17586.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/f14bb50b76a60253a090797c652a9d29120bf875">f14bb50b</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-30T06:52:09-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">rts: Ensure that nonmoving gc isn't used with profiling

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/b426de376e278af5fb7a8ce0ef7b1e245847c662">b426de37</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-30T06:52:45-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">llvmGen: Ensure that entry labels don't have predecessors

The LLVM IR forbids the entry label of a procedure from having any

predecessors. In the case of a simple looping function the LLVM code

generator broke this invariant, as noted in #17589. Fix this by

moving the function prologue to its own basic block, as suggested by

@kavon in #11649.

Fixes #11649 and #17589.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/613f7265c7985b779877e6897118c53d5a333f10">613f7265</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-30T06:52:45-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">llvmGen: Drop old fix for #11649

This was a hack which is no longer necessary now since we introduce a

dedicated entry block for each procedure.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/fdeffa5e56abccc64df750ba67a39b4af99adac5">fdeffa5e</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-30T06:53:23-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">rts: Error on invalid --numa flags

Previously things like `+RTS --numa-debug` would enable NUMA support,

despite being an invalid flag.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/9ce3ba68ad30d2173983be309ecfc08737e1dc38">9ce3ba68</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-30T06:53:23-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">rts: Fix --debug-numa mode under Docker

As noted in #17606, Docker disallows the get_mempolicy syscall by

default. This caused numerous tests to fail under CI in the `debug_numa`

way. Avoid this by disabling the NUMA probing logic when --debug-numa is

in use, instead setting n_numa_nodes in RtsFlags.c.

Fixes #17606.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/5baa2a4326700c3044904ca0535d45d373ef3b89">5baa2a43</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-30T06:54:01-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Disable derefnull when built with LLVM

LLVM does not guarantee any particular semantics when dereferencing null

pointers. Consequently, this test actually passes when built with the

LLVM backend.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/bd544d3df7ebe6326e38d532ddc4a3756c8fe029">bd544d3d</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-30T06:54:38-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">hadrian: Track hash of Cabal Setup builder arguments

Lest we fail to rebuild when they change. Fixes #17611.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/6e2c495e3c600710bb6fc07dfb3cf0af41b362da">6e2c495e</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-30T06:55:19-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">TcIface: Fix inverted logic in typechecking of source ticks

Previously we would throw away source ticks when the debug level was

non-zero. This is precisely the opposite of what was intended.

Fixes #17616.

Metric Decrease:

    T13056

    T9020

    T9961

    T12425

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/7fad387d4c2ab8031eb28cf6b02b4e35f7143374">7fad387d</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-30T06:55:55-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">perf_notes: Add --zero-y argument

This makes it easier to see the true magnitude of fluctuations.

Also do some house-keeping in the argument parsing department.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/0d42b287c3fe2510433a7fb744531a0765ad8ac8">0d42b287</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2019-12-30T06:55:55-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Enlarge acceptance window for T1969

As noted in #17624, it's quite unstable, especially, for some reason, on

i386 and armv7 (something about 32-bit platforms perhaps?).

Metric Increase:

    T1969

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/eb6082358cdb5f271a8e4c74044a12f97352c52f">eb608235</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2019-12-31T14:22:32-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Module hierarchy (#13009): Stg

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/d710fd666a70ae100120042c8549634537001609">d710fd66</a></strong>

<div>

<span>by Vladislav Zavialov</span>

<i>at 2019-12-31T14:23:10-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Testsuite: update some Haddock tests

Fixed tests:

* haddockA039: added to all.T

* haddockE004: replaced with T17561 (marked as expect_broken)

New tests:

* haddockA040: deriving clause for a data instance

* haddockA041: haddock and CPP #include

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/859ebdd446eda446d38708a587503c255b58c4c6">859ebdd4</a></strong>

<div>

<span>by Kevin Buhr</span>

<i>at 2019-12-31T23:44:39-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add "-Iw" RTS flag for minimum wait between idle GCs (#11134)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/dd4b65519f99e85a7e8f5c2dcb654a8d463c8858">dd4b6551</a></strong>

<div>

<span>by Kevin Buhr</span>

<i>at 2019-12-31T23:44:39-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add additional Note explaining the -Iw flag

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/c4279ff18070c398b5ddc677f9c5a915de68dafc">c4279ff1</a></strong>

<div>

<span>by Kevin Buhr</span>

<i>at 2019-12-31T23:44:39-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix some sloppy indentation

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/b84c09d533faf576c406ce9f7163efecf3037787">b84c09d5</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2019-12-31T23:45:19-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Tweak Cmm dumps to avoid generating sections for empty groups

When dumping Cmm groups check if the group is empty, to avoid generating

empty sections in dump files like

    ==================== Output Cmm ====================

    []

Also fixes a few bad indentation in the code around changes.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/b2e0323f318959c879629ef277f6433b44473c4b">b2e0323f</a></strong>

<div>

<span>by Gabor Greif</span>

<i>at 2020-01-03T21:22:36-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Simplify mrStr

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/3c9dc06ba2034e867c9169e60e854539875654fd">3c9dc06b</a></strong>

<div>

<span>by Brian Wignall</span>

<i>at 2020-01-04T15:55:06-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix typos, via a Levenshtein-style corrector

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/d561c8f6244f8280a2483e8753c38e39d34c1f01">d561c8f6</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-01-04T15:55:46-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add Cmm related hooks

* stgToCmm hook

* cmmToRawCmm hook

These hooks are used by Asterius and could be useful to other clients of

the GHC API.

It increases the Parser dependencies (test CountParserDeps) to 184. It's

still less than 200 which was the initial request (cf

https://mail.haskell.org/pipermail/ghc-devs/2019-September/018122.html)

so I think it's ok to merge this.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/ae6b62765b12b44163612488a94a9ac270b4dcc5">ae6b6276</a></strong>

<div>

<span>by Oleg Grenrus</span>

<i>at 2020-01-04T15:56:22-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Update to Cabal submodule to v3.2.0.0-alpha3

Metric Increase:

    haddock.Cabal

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/073f7cfdf374846e287f826cf6a83e7b3fec55a6">073f7cfd</a></strong>

<div>

<span>by Vladislav Zavialov</span>

<i>at 2020-01-04T15:56:59-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add lexerDbg to dump the tokens fed to the parser

This a small utility function that comes in handy when debugging the

lexer and the parser.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/558d4d4a2b5927875dceb42e120026e75e313f33">558d4d4a</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-01-04T15:57:38-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Split integerGmpInternals test in several parts

This is to prepare for ghc-bignum which implements some but not all of

gmp functions.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/4056b9666a0fc5865a82fbba03e8312d7900ac5b">4056b966</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-01-04T15:58:15-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Mark cgrun057 as fragile on all platforms

I have seen this fail both on x86-64/Debian 9 and armv7/Debian 9

See #17554.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/5ffea0c6c6a2670fd6819540f3ea61ce6620caaa">5ffea0c6</a></strong>

<div>

<span>by Tamar Christina</span>

<i>at 2020-01-06T18:38:37-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix overflow.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/99a9f51bf8207c79241fc0b685fadeb222a61292">99a9f51b</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-01-06T18:39:22-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Module hierarchy: Iface (cf #13009)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/7aa4a0615629f36e520c11c7e40db7b1475b6402">7aa4a061</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-01-07T13:11:48-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">configure: Only check GCC version if CC is GCC

Also refactor FP_GCC_EXTRA_FLAGS in a few ways:

 * We no longer support compilers which lack support for -fno-builtin

   and -fwrapv so remove the condition on GccVersion

 * These flags are only necessary when using the via-C backend

   so make them conditional on Unregisterised.

Fixes #15742.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/0805ed7ef32891685109b6cf3461f8adb4ca6981">0805ed7e</a></strong>

<div>

<span>by John Ericson</span>

<i>at 2020-01-07T13:12:25-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Use non-empty lists to remove partiality in matching code

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/7844f3a84fca2a79910f164f5a9d2d44b589bb44">7844f3a8</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-01-07T13:13:02-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Mark T17073 as broken on Windows

Due to #17607.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/acf40cae24dab7be7c719dba191daa5cf9f7e676">acf40cae</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-01-07T13:13:02-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Disallow Windows from failing

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/34bc02c7bc8b7e8c267e6da9a8c35a6c25f1b72e">34bc02c7</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-01-07T13:13:02-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">configure: Find Python3 for testsuite

In addition, we prefer the Mingw64 Python distribution on Windows due

to #17483.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/e35fe8d58f18bd179efdc848c617dc9eddf4478b">e35fe8d5</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-01-07T13:13:02-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Fix Windows platform test

Previously we used platform.system() and while this worked fine (e.g.

returned `Windows`, as expected) locally under both msys and MingW64

Python distributions, it inexplicably returned `MINGW64_NT-10.0`

under MingW64 Python on CI. It seems os.name is more reliable so we now

use that instead..

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/48ef62175d4e65d3248f2e743750ec2f4d421f3a">48ef6217</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-01-07T13:13:39-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Rename push-test-metrics.sh to test-metrics.sh

Refactoring to follow.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/2234fa9252b5da2e4423d1ec52c9830130ecfe35">2234fa92</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-01-07T13:13:39-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Pull test metrics before running testsuite

Otherwise the testsuite driver may not have an up-to-date

baseline.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/1ca9adbc88903afe49de0d063ccd35daf43f7d9e">1ca9adbc</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-01-07T13:14:18-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Remove `parallel` check from configure.ac

`parallel` is no longer a submodule since 3cb063c805ec841ca33b8371ef8aba9329221b6c

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/b69a3460d11cba49e861f708100801c8e25efa3e">b69a3460</a></strong>

<div>

<span>by Ryan Scott</span>

<i>at 2020-01-07T13:14:57-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Monomorphize HsModule to GhcPs (#17642)

Analyzing the call sites for `HsModule` reveals that it is only ever

used with parsed code (i.e., `GhcPs`). This simplifies `HsModule` by

concretizing its `pass` parameter to always be `GhcPs`.

Fixes #17642.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/d491a6795d507eabe35d8aec63c534d29f2d305b">d491a679</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-01-08T06:16:31-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Module hierarchy: Renamer (cf #13009)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/d589410f6bf394c8aca53f18848bbe4dfca92f23">d589410f</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-01-08T06:17:09-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Bump haskeline submodule to 0.8.0.1

(cherry picked from commit feb3b955402d53c3875dd7a9a39f322827e5bd69)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/923a127205dd60147453f4420614efd1be29f070">923a1272</a></strong>

<div>

<span>by Ryan Scott</span>

<i>at 2020-01-08T06:17:47-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Print Core type applications with no whitespace after @ (#17643)

This brings the pretty-printer for Core in line with how visible

type applications are normally printed: namely, with no whitespace

after the `@` character (i.e., `f @a` instead of `f @ a`). While I'm

in town, I also give the same treatment to type abstractions (i.e.,

`\(@a)` instead of `\(@ a)`) and coercion applications (i.e.,

`f @~x` instead of `f @~ x`).

Fixes #17643.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/49f83a0de12a7c02f4a6e99d26eaa362a373afa5">49f83a0d</a></strong>

<div>

<span>by Adam Sandberg Eriksson</span>

<i>at 2020-01-12T21:28:09-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">improve docs for HeaderInfo.getImports

[skip ci]

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/9129210f7e9937c1065330295f06524661575839">9129210f</a></strong>

<div>

<span>by Matthew Pickering</span>

<i>at 2020-01-12T21:28:47-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Overloaded Quotation Brackets (#246)

This patch implements overloaded quotation brackets which generalise the

desugaring of all quotation forms in terms of a new minimal interface.

The main change is that a quotation, for example, [e| 5 |], will now

have type `Quote m => m Exp` rather than `Q Exp`. The `Quote` typeclass

contains a single method for generating new names which is used when

desugaring binding structures.

The return type of functions from the `Lift` type class, `lift` and `liftTyped` have

been restricted to `forall m . Quote m => m Exp` rather than returning a

result in a Q monad.

More details about the feature can be read in the GHC proposal.

https://github.com/ghc-proposals/ghc-proposals/blob/master/proposals/0246-overloaded-bracket.rst

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/350e2b78788d47255d27489dfc62d664498b5de4">350e2b78</a></strong>

<div>

<span>by Richard Eisenberg</span>

<i>at 2020-01-12T21:29:27-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Don't zap to Any; error instead

This changes GHC's treatment of so-called Naughty Quantification

Candidates to issue errors, instead of zapping to Any.

Close #16775.

No new test cases, because existing ones cover this well.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/0b5ddc7f2c10ee84631dd6cb5f6368afbc389449">0b5ddc7f</a></strong>

<div>

<span>by Brian Wignall</span>

<i>at 2020-01-12T21:30:08-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix more typos, via an improved Levenshtein-style corrector

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/f732dbec2159e314619d372aa9cc34d6ab8fadf5">f732dbec</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-01-12T21:30:49-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Retain bindists used by head.hackage for longer

Previously we would keep them for two weeks. However, on the stable

branches two weeks can easily elapse with no pushes.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/c8636da51449ab13399cb029e3e55d7a246ac5a6">c8636da5</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-01-12T21:31:30-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix LANG=C for readelf invocation in T14999

The test fails when used with LANG=fr_FR.UTF-8

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/077a88de5c0981badbb61b06f32073d7f3a49b08">077a88de</a></strong>

<div>

<span>by Jean-Baptiste Mazon</span>

<i>at 2020-01-12T21:32:08-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">users-guide/debug-info: typo “behivior”</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/61916c5d038c3b0e87148d3b46149a32f7379ea9">61916c5d</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2020-01-12T21:32:44-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add comments about TH levels

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/1fd766ca7ac1feea1e9c752f8ce438fab8e01d35">1fd766ca</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2020-01-12T21:32:44-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Comments about constraint floating

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/de01427e6b1d45122c84eaf03122fda26c4dc688">de01427e</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2020-01-12T21:32:45-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Minor refactor around quantified constraints

This patch clarifies a dark corner of quantified

constraints.

* See Note [Yukky eq_sel for a HoleDest] in TcSMonad

* Minor refactor, breaking out new function

     TcInteract.doTopReactEqPred

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/30be3bf13a6e72247ff561df1f291370dad79ef9">30be3bf1</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2020-01-12T21:32:45-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Comments in TcHsType

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/c5977d4dacc43e76438acb316d12575e0ead18e2">c5977d4d</a></strong>

<div>

<span>by Sebastian Graf</span>

<i>at 2020-01-16T05:58:58-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Better documentation for mkEtaWW [skip ci]

So that hopefully I understand it faster next time. Also got rid of the

confusing `orig_expr`, which makes the call site in `etaExpand` look out

of sync with the passed `n` (which is not the original `n`).

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/22c0bdc3f3a4e3b787b64ec0f3adba5c7c5b5d04">22c0bdc3</a></strong>

<div>

<span>by John Ericson</span>

<i>at 2020-01-16T05:59:37-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Handle TagToEnum in the same big case as the other primops

Before, it was a panic because it was handled above. But there must have

been an error in my reasoning (another caller?) because #17442 reported

the panic was hit.

But, rather than figuring out what happened, I can just make it

impossible by construction. By adding just a bit more bureaucracy in the

return types, I can handle TagToEnum in the same case as all the others,

so the big case is is now total, and the panic is removed.

Fixes #17442

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/ee5d63f40c2f507d09a16377a5b35c4c8669a028">ee5d63f4</a></strong>

<div>

<span>by John Ericson</span>

<i>at 2020-01-16T05:59:37-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Get rid of OpDest

`OpDest` was basically a defunctionalization. Just turn the code that

cased on it into those functions, and call them directly.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/1ff552265bbbcc0dba272d4742e8be79baad2c51">1ff55226</a></strong>

<div>

<span>by John Ericson</span>

<i>at 2020-01-16T06:00:16-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Remove special case case of bool during STG -> C--

Allow removing the no longer needed cgPrimOp, getting rid of a small a

small layer violation too.

Change which made the special case no longer needed was #6135 /

6579a6c73082387f82b994305011f011d9d8382b, which dates back to 2013,

making me feel better.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/f416fe6432c0f2f2237f096a9fb0e58e897b4ba5">f416fe64</a></strong>

<div>

<span>by Adam Wespiser</span>

<i>at 2020-01-16T06:00:53-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">replace dead html link (fixes #17661)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/f6bf2ce88a13bab21ac6da289f3d32dd8f811048">f6bf2ce8</a></strong>

<div>

<span>by Sebastian Graf</span>

<i>at 2020-01-16T06:01:32-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Revert "`exprOkForSpeculation` for Note [IO hack in the demand analyser]"

This reverts commit ce64b397777408731c6dd3f5c55ea8415f9f565b on the

grounds of the regression it would introduce in a couple of packages.

Fixes #17653.

Also undoes a slight metric increase in #13701 introduced by that commit

that we didn't see prior to !1983.

Metric Decrease:

    T13701

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/a71323ffebf7663c50025d2731bf9de2d04f82c3">a71323ff</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-01-17T08:43:16-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Don't FORCE_SYMLINKS on Windows

Not all runners have symlink permissions enabled.</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/0499e3bcd0ea04b7371cee56a8aea1781e9e371f">0499e3bc</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2020-01-20T15:31:33-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix +RTS -Z flag documentation

Stack squeezing is done on context switch, not on GC or stack overflow.

Fix the documentation.

Fixes #17685

[ci skip]

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/a661df91da5d867ab3e6a912e03a9e1756e59cb6">a661df91</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2020-01-20T15:32:13-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Document Stg.FVs module

Fixes #17662

[ci skip]

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/db24e4803fe9fb13b85fc83193ff4afc407702f6">db24e480</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-01-20T15:32:52-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">llvmGen: Don't trash STG registers

Fixes #13904.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/f3d7fdb3437a6161cf7ff1f30cd0a6563bf07db0">f3d7fdb3</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-01-20T15:32:52-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">llvmGen: Fix typo in readnone attribute

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/442751c605d8a9792014a9e3e694db5b312e80c3">442751c6</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-01-20T15:32:52-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">llvmGen: Add lower-expect to the -O0 optimisation set

@kavon says that this will improve block layout for stack checks.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/e90ecc932c4857c774405dc398c940290f70f371">e90ecc93</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-01-20T15:32:52-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">llvmGen: Fix #14251

Fixes the calling convention for functions passing raw SSE-register

values by adding padding as needed to get the values in the right

registers. This problem cropped up when some args were unused an dropped

from the live list.

This folds together 2e23e1c7de01c92b038e55ce53d11bf9db993dd4 and

73273be476a8cc6c13368660b042b3b0614fd928 previously from @kavon.

Metric Increase:

    T12707

    ManyConstructors

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/66e511a416b80585995d2b4a669eb037471cb57d">66e511a4</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-01-20T15:33:28-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Preserve more information in framework failures

Namely print the entire exception in hopes that this will help track

down #17649.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/b62b8cea0e19073133bf3b6757500d1723d02159">b62b8cea</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2020-01-20T15:34:06-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Remove deprecated -smp flag

It was deprecated in 2012 with 46258b40

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/0c04a86af8bc04ee6f966a17381b88a297a54836">0c04a86a</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-01-20T15:34:43-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Reenable submodule linter

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/2bfabd2282e9ced19a273aa94b5d8dd1a27d6bfa">2bfabd22</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-01-20T15:34:43-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Allow submodule cleaning to fail on Windows

Currently CI is inexplicably failing with

```

$ git submodule foreach git clean -xdf

fatal: not a git repository: libffi-tarballs/../.git/modules/libffi-tarballs

```

I have no idea how this working tree got into such a state but we do

need to fail more gracefully when it happens. Consequently, we allow the

cleaning step to fail.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/14bced999dc8c826a9a10dfd922dadf2d681ffe3">14bced99</a></strong>

<div>

<span>by Xavier Denis</span>

<i>at 2020-01-20T15:35:21-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Put the docs for :instances in alphabetical position

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/7e0bb82b5e267596c18e1086afd6007a00d2c072">7e0bb82b</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-01-20T15:35:57-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add missing Note [Improvement from Ground Wanteds]

Closes #17659.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/17e43a7cd48d862f38d3944d44024646f2e97b87">17e43a7c</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-01-20T15:36:32-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">unregisterised: Fix declaration for stg_NO_FINALIZER

Previously it had a redundant _entry suffix. We never noticed this

previously presumably because we never generated references to it

(however hard to believe this may be). However, it did start failing in

!1304.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/3dae006fc424e768bb43fc73851a08fefcb732a5">3dae006f</a></strong>

<div>

<span>by PHO</span>

<i>at 2020-01-20T15:37:08-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Avoid ./configure failure on NetBSD

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/738e2912bfa2122074630d38693a260e4b58aeaf">738e2912</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-01-24T13:42:56-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Widen acceptance window of T1969

I have seen >20% fluctuations in this number, leading to spurious

failures.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/ad4eb7a76f6f1fb18e3ea47665feb3a9893427de">ad4eb7a7</a></strong>

<div>

<span>by Gabor Greif</span>

<i>at 2020-01-25T05:19:07-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Document the fact, that openFileBlocking can consume an OS thread indefinitely.

Also state that a deadlock can happen with the non-threaded runtime.

[ci skip]</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/be910728d78fdf2ee800828ecdc8a11bd64fdad0">be910728</a></strong>

<div>

<span>by Sebastian Graf</span>

<i>at 2020-01-25T05:19:46-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">`-ddump-str-signatures` dumps Text, not STG [skip ci]

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/0e57d8a106a61cac11bacb43633b8b4af12d7fdb">0e57d8a1</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2020-01-25T05:20:27-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix chaining tagged and untagged ptrs in compacting GC

Currently compacting GC has the invariant that in a chain all fields are tagged

the same. However this does not really hold: root pointers are not tagged, so

when we thread a root we initialize a chain without a tag. When the pointed

objects is evaluated and we have more pointers to it from the heap, we then add

*tagged* fields to the chain (because pointers to it from the heap are tagged),

ending up chaining fields with different tags (pointers from roots are NOT

tagged, pointers from heap are). This breaks the invariant and as a result

compacting GC turns tagged pointers into non-tagged.

This later causes problem in the generated code where we do reads assuming that

the pointer is aligned, e.g.

    0x7(%rax) -- assumes that pointer is tagged 1

which causes misaligned reads. This caused #17088.

We fix this using the "pointer tagging for large families" patch (#14373,

!1742):

- With the pointer tagging patch the GC can know what the tagged pointer to a

  CONSTR should be (previously we'd need to know the family size -- large

  families are always tagged 1, small families are tagged depending on the

  constructor).

- Since we now know what the tags should be we no longer need to store the

  pointer tag in the info table pointers when forming chains in the compacting

  GC.

As a result we no longer need to tag pointers in chains with 1/2 depending on

whether the field points to an info table pointer, or to another field: an info

table pointer is always tagged 0, everything else in the chain is tagged 1. The

lost tags in pointers can be retrieved by looking at the info table.

Finally, instead of using tag 1 for fields and tag 0 for info table pointers, we

use two different tags for fields:

- 1 for fields that have untagged pointers

- 2 for fields that have tagged pointers

When unchaining we then look at the pointer to a field, and depending on its tag

we either leave a tagged pointer or an untagged pointer in the field.

This allows chaining untagged and tagged fields together in compacting GC.

Fixes #17088

Nofib results

-------------

Binaries are smaller because of smaller `Compact.c` code.

make mode=fast EXTRA_RUNTEST_OPTS="-cachegrind" EXTRA_HC_OPTS="-with-rtsopts=-c" NoFibRuns=1

    --------------------------------------------------------------------------------

            Program           Size    Allocs    Instrs     Reads    Writes

    --------------------------------------------------------------------------------

                 CS          -0.3%      0.0%     +0.0%     +0.0%     +0.0%

                CSD          -0.3%      0.0%     +0.0%     +0.0%     +0.0%

                 FS          -0.3%      0.0%     +0.0%     -0.0%     -0.0%

                  S          -0.3%      0.0%     +5.4%     +0.8%     +3.9%

                 VS          -0.3%      0.0%     +0.0%     -0.0%     -0.0%

                VSD          -0.3%      0.0%     -0.0%     -0.0%     -0.2%

                VSM          -0.3%      0.0%     +0.0%     +0.0%     +0.0%

               anna          -0.1%      0.0%     +0.0%     +0.0%     +0.0%

               ansi          -0.3%      0.0%     +0.1%     +0.0%     +0.0%

               atom          -0.2%      0.0%     +0.0%     +0.0%     +0.0%

             awards          -0.2%      0.0%     +0.0%      0.0%     -0.0%

             banner          -0.3%      0.0%     +0.0%     +0.0%     +0.0%

         bernouilli          -0.3%      0.0%     +0.1%     +0.0%     +0.0%

       binary-trees          -0.2%      0.0%     +0.0%      0.0%     +0.0%

              boyer          -0.3%      0.0%     +0.2%     +0.0%     +0.0%

             boyer2          -0.2%      0.0%     +0.2%     +0.1%     +0.0%

               bspt          -0.2%      0.0%     +0.0%     +0.0%     +0.0%

          cacheprof          -0.2%      0.0%     +0.0%     +0.0%     +0.0%

           calendar          -0.3%      0.0%     +0.0%     +0.0%     +0.0%

           cichelli          -0.3%      0.0%     +1.1%     +0.2%     +0.5%

            circsim          -0.2%      0.0%     +0.0%     -0.0%     -0.0%

           clausify          -0.3%      0.0%     +0.0%     -0.0%     -0.0%

      comp_lab_zift          -0.2%      0.0%     +0.0%     +0.0%     +0.0%

           compress          -0.3%      0.0%     +0.0%     +0.0%     +0.0%

          compress2          -0.3%      0.0%     +0.0%     -0.0%     -0.0%

        constraints          -0.3%      0.0%     +0.2%     +0.1%     +0.1%

       cryptarithm1          -0.3%      0.0%     +0.0%     -0.0%      0.0%

       cryptarithm2          -0.3%      0.0%     +0.0%     +0.0%     +0.0%

                cse          -0.3%      0.0%     +0.0%     +0.0%     +0.0%

       digits-of-e1          -0.3%      0.0%     +0.0%     +0.0%     +0.0%

       digits-of-e2          -0.3%      0.0%     +0.0%     +0.0%     -0.0%

             dom-lt          -0.2%      0.0%     +0.0%     +0.0%     +0.0%

              eliza          -0.2%      0.0%     +0.0%     +0.0%     +0.0%

              event          -0.3%      0.0%     +0.1%     +0.0%     -0.0%

        exact-reals          -0.2%      0.0%     +0.0%     +0.0%     +0.0%

             exp3_8          -0.3%      0.0%     +0.0%     +0.0%     +0.0%

             expert          -0.2%      0.0%     +0.0%     +0.0%     +0.0%

     fannkuch-redux          -0.3%      0.0%     -0.0%     -0.0%     -0.0%

              fasta          -0.3%      0.0%     +0.0%     +0.0%     +0.0%

                fem          -0.2%      0.0%     +0.1%     +0.0%     +0.0%

                fft          -0.2%      0.0%     +0.0%     -0.0%     -0.0%

               fft2          -0.2%      0.0%     +0.0%     -0.0%     +0.0%

           fibheaps          -0.3%      0.0%     +0.0%     -0.0%     -0.0%

               fish          -0.3%      0.0%     +0.0%     +0.0%     +0.0%

              fluid          -0.2%      0.0%     +0.4%     +0.1%     +0.1%

             fulsom          -0.2%      0.0%     +0.0%     +0.0%     +0.0%

             gamteb          -0.2%      0.0%     +0.1%     +0.0%     +0.0%

                gcd          -0.3%      0.0%     +0.0%     +0.0%     +0.0%

        gen_regexps          -0.3%      0.0%     +0.0%     -0.0%     -0.0%

             genfft          -0.3%      0.0%     +0.0%     +0.0%     +0.0%

                 gg          -0.2%      0.0%     +0.7%     +0.3%     +0.2%

               grep          -0.2%      0.0%     +0.0%     +0.0%     +0.0%

             hidden          -0.2%      0.0%     +0.0%     +0.0%     +0.0%

                hpg          -0.2%      0.0%     +0.1%     +0.0%     +0.0%

                ida          -0.3%      0.0%     +0.0%     +0.0%     +0.0%

              infer          -0.2%      0.0%     +0.0%     -0.0%     -0.0%

            integer          -0.3%      0.0%     +0.0%     +0.0%     +0.0%

          integrate          -0.2%      0.0%     +0.0%     +0.0%     +0.0%

       k-nucleotide          -0.2%      0.0%     +0.0%     +0.0%     -0.0%

              kahan          -0.3%      0.0%     -0.0%     -0.0%     -0.0%

            knights          -0.3%      0.0%     +0.0%     -0.0%     -0.0%

             lambda          -0.3%      0.0%     +0.0%     -0.0%     -0.0%

         last-piece          -0.3%      0.0%     +0.0%     +0.0%     +0.0%

               lcss          -0.3%      0.0%     +0.0%     +0.0%      0.0%

               life          -0.3%      0.0%     +0.0%     -0.0%     -0.0%

               lift          -0.2%      0.0%     +0.0%     +0.0%     +0.0%

             linear          -0.2%      0.0%     +0.0%     +0.0%     +0.0%

          listcompr          -0.3%      0.0%     +0.0%     +0.0%     +0.0%

           listcopy          -0.3%      0.0%     +0.0%     +0.0%     +0.0%

           maillist          -0.3%      0.0%     +0.0%     -0.0%     -0.0%

             mandel          -0.2%      0.0%     +0.0%     +0.0%     +0.0%

            mandel2          -0.3%      0.0%     +0.0%     +0.0%     +0.0%

               mate          -0.2%      0.0%     +0.0%     +0.0%     +0.0%

            minimax          -0.3%      0.0%     +0.0%     +0.0%     +0.0%

            mkhprog          -0.2%      0.0%     +0.0%     +0.0%     +0.0%

         multiplier          -0.3%      0.0%     +0.0%     -0.0%     -0.0%

             n-body          -0.2%      0.0%     -0.0%     -0.0%     -0.0%

           nucleic2          -0.2%      0.0%     +0.0%     +0.0%     +0.0%

               para          -0.2%      0.0%     +0.0%     -0.0%     -0.0%

          paraffins          -0.3%      0.0%     +0.0%     -0.0%     -0.0%

             parser          -0.2%      0.0%     +0.0%     +0.0%     +0.0%

            parstof          -0.2%      0.0%     +0.8%     +0.2%     +0.2%

                pic          -0.2%      0.0%     +0.1%     -0.1%     -0.1%

           pidigits          -0.3%      0.0%     +0.0%     +0.0%     +0.0%

              power          -0.2%      0.0%     +0.0%     -0.0%     -0.0%

             pretty          -0.3%      0.0%     -0.0%     -0.0%     -0.1%

             primes          -0.3%      0.0%     +0.0%     +0.0%     -0.0%

          primetest          -0.2%      0.0%     +0.0%     -0.0%     -0.0%

             prolog          -0.3%      0.0%     +0.0%     -0.0%     -0.0%

             puzzle          -0.3%      0.0%     +0.0%     +0.0%     +0.0%

             queens          -0.3%      0.0%     +0.0%     +0.0%     +0.0%

            reptile          -0.2%      0.0%     +0.2%     +0.1%     +0.0%

    reverse-complem          -0.3%      0.0%     +0.0%     +0.0%     +0.0%

            rewrite          -0.3%      0.0%     +0.0%     -0.0%     -0.0%

               rfib          -0.2%      0.0%     +0.0%     +0.0%     -0.0%

                rsa          -0.2%      0.0%     +0.0%     +0.0%     +0.0%

                scc          -0.3%      0.0%     -0.0%     -0.0%     -0.1%

              sched          -0.3%      0.0%     +0.0%     +0.0%     +0.0%

                scs          -0.2%      0.0%     +0.1%     +0.0%     +0.0%

             simple          -0.2%      0.0%     +3.4%     +1.0%     +1.8%

              solid          -0.2%      0.0%     +0.0%     +0.0%     +0.0%

            sorting          -0.3%      0.0%     +0.0%     +0.0%     +0.0%

      spectral-norm          -0.2%      0.0%     -0.0%     -0.0%     -0.0%

             sphere          -0.2%      0.0%     +0.0%     +0.0%     +0.0%

             symalg          -0.2%      0.0%     +0.0%     +0.0%     +0.0%

                tak          -0.3%      0.0%     +0.0%     +0.0%     -0.0%

          transform          -0.2%      0.0%     +0.2%     +0.1%     +0.1%

           treejoin          -0.3%      0.0%     +0.2%     -0.0%     -0.1%

          typecheck          -0.3%      0.0%     +0.0%     +0.0%     +0.0%

            veritas          -0.1%      0.0%     +0.0%     +0.0%     +0.0%

               wang          -0.2%      0.0%     +0.0%     -0.0%     -0.0%

          wave4main          -0.3%      0.0%     +0.0%     -0.0%     -0.0%

       wheel-sieve1          -0.3%      0.0%     +0.0%     -0.0%     -0.0%

       wheel-sieve2          -0.3%      0.0%     +0.0%     -0.0%     -0.0%

               x2n1          -0.3%      0.0%     +0.0%     +0.0%     +0.0%

    --------------------------------------------------------------------------------

                Min          -0.3%      0.0%     -0.0%     -0.1%     -0.2%

                Max          -0.1%      0.0%     +5.4%     +1.0%     +3.9%

     Geometric Mean          -0.3%     -0.0%     +0.1%     +0.0%     +0.1%

    --------------------------------------------------------------------------------

            Program           Size    Allocs    Instrs     Reads    Writes

    --------------------------------------------------------------------------------

            circsim          -0.2%      0.0%     +1.6%     +0.4%     +0.7%

        constraints          -0.3%      0.0%     +4.3%     +1.5%     +2.3%

           fibheaps          -0.3%      0.0%     +3.5%     +1.2%     +1.3%

             fulsom          -0.2%      0.0%     +3.6%     +1.2%     +1.8%

           gc_bench          -0.3%      0.0%     +4.1%     +1.3%     +2.3%

               hash          -0.3%      0.0%     +6.6%     +2.2%     +3.6%

               lcss          -0.3%      0.0%     +0.7%     +0.2%     +0.7%

          mutstore1          -0.3%      0.0%     +4.8%     +1.4%     +2.8%

          mutstore2          -0.3%      0.0%     +3.4%     +1.0%     +1.7%

              power          -0.2%      0.0%     +2.7%     +0.6%     +1.9%

         spellcheck          -0.3%      0.0%     +1.1%     +0.4%     +0.4%

    --------------------------------------------------------------------------------

                Min          -0.3%      0.0%     +0.7%     +0.2%     +0.4%

                Max          -0.2%      0.0%     +6.6%     +2.2%     +3.6%

     Geometric Mean          -0.3%     +0.0%     +3.3%     +1.0%     +1.8%

Metric changes

--------------

While it sounds ridiculous, this change causes increased allocations in

the following tests. We concluded that this change can't cause a

difference in allocations and decided to land this patch. Fluctuations

in "bytes allocated" metric is tracked in #17686.

Metric Increase:

    Naperian

    T10547

    T12150

    T12234

    T12425

    T13035

    T5837

    T6048

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/8038cbd96f444fdba18e8c9fb292c565738b774d">8038cbd9</a></strong>

<div>

<span>by Sebastian Graf</span>

<i>at 2020-01-25T05:21:05-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">PmCheck: Formulate as translation between Clause Trees

We used to check `GrdVec`s arising from multiple clauses and guards in

isolation. That resulted in a split between `pmCheck` and

`pmCheckGuards`, the implementations of which were similar, but subtly

different in detail. Also the throttling mechanism described in

`Note [Countering exponential blowup]` ultimately got quite complicated

because it had to cater for both checking functions.

This patch realises that pattern match checking doesn't just consider

single guarded RHSs, but that it's always a whole set of clauses, each

of which can have multiple guarded RHSs in turn. We do so by

translating a list of `Match`es to a `GrdTree`:

```haskell

data GrdTree

  = Rhs !RhsInfo

  | Guard !PmGrd !GrdTree      -- captures lef-to-right  match semantics

  | Sequence !GrdTree !GrdTree -- captures top-to-bottom match semantics

  | Empty                      -- For -XEmptyCase, neutral element of Sequence

```

Then we have a function `checkGrdTree` that matches a given `GrdTree`

against an incoming set of values, represented by `Deltas`:

```haskell

checkGrdTree :: GrdTree -> Deltas -> CheckResult

...

```

Throttling is isolated to the `Sequence` case and becomes as easy as one

would expect: When the union of uncovered values becomes too big, just

return the original incoming `Deltas` instead (which is always a

superset of the union, thus a sound approximation).

The returned `CheckResult` contains two things:

1. The set of values that were not covered by any of the clauses, for

   exhaustivity warnings.

2. The `AnnotatedTree` that enriches the syntactic structure of the

   input program with divergence and inaccessibility information.

This is `AnnotatedTree`:

```haskell

data AnnotatedTree

  = AccessibleRhs !RhsInfo

  | InaccessibleRhs !RhsInfo

  | MayDiverge !AnnotatedTree

  | SequenceAnn !AnnotatedTree !AnnotatedTree

  | EmptyAnn

```

Crucially, `MayDiverge` asserts that the tree may force diverging

values, so not all of its wrapped clauses can be redundant.

While the set of uncovered values can be used to generate the missing

equations for warning messages, redundant and proper inaccessible

equations can be extracted from `AnnotatedTree` by

`redundantAndInaccessibleRhss`.

For this to work properly, the interface to the Oracle had to change.

There's only `addPmCts` now, which takes a bag of `PmCt`s. There's a

whole bunch of `PmCt` variants to replace the different oracle functions

from before.

The new `AnnotatedTree` structure allows for more accurate warning

reporting (as evidenced by a number of changes spread throughout GHC's

code base), thus we fix #17465.

Fixes #17646 on the go.

Metric Decrease:

    T11822

    T9233

    PmSeriesS

    haddock.compiler

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/86966d48954db4a8bd40046af259ed60aed535eb">86966d48</a></strong>

<div>

<span>by Sebastian Graf</span>

<i>at 2020-01-25T05:21:05-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">PmCheck: Properly handle constructor-bound type variables

In https://gitlab.haskell.org/ghc/ghc/merge_requests/2192#note_246551

Simon convinced me that ignoring type variables existentially bound by

data constructors have to be the same way as value binders.

Sadly I couldn't think of a regression test, but I'm confident that this

change strictly improves on the status quo.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/c3fde723633d1788e4ded8c6f59eb7cef1ae95fd">c3fde723</a></strong>

<div>

<span>by Ryan Scott</span>

<i>at 2020-01-25T05:21:40-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Handle local fixity declarations in DsMeta properly

`DsMeta.rep_sig` used to skip over `FixSig` entirely, which had the

effect of causing local fixity declarations to be dropped when quoted

in Template Haskell. But there is no good reason for this state of

affairs, as the code in `DsMeta.repFixD` (which handles top-level

fixity declarations) handles local fixity declarations just fine.

This patch factors out the necessary parts of `repFixD` so that they

can be used in `rep_sig` as well.

There was one minor complication: the fixity signatures for class

methods in each `HsGroup` were stored both in `FixSig`s _and_ the

list of `LFixitySig`s for top-level fixity signatures, so I needed

to take action to prevent fixity signatures for class methods being

converted to `Dec`s twice. I tweaked `RnSource.add` to avoid putting

these fixity signatures in two places and added

`Note [Top-level fixity signatures in an HsGroup]` in `GHC.Hs.Decls`

to explain the new design.

Fixes #17608. Bumps the Haddock submodule.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/6e2d9ee25bce06ae51d2f1cf8df4f7422106a383">6e2d9ee2</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-01-25T05:22:20-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Module hierarchy: Cmm (cf #13009)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/8b726534fa4764ad423eff9bdfb04305f330a00f">8b726534</a></strong>

<div>

<span>by PHO</span>

<i>at 2020-01-25T05:23:01-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix rts allocateExec() on NetBSD

Similar to SELinux, NetBSD "PaX mprotect" prohibits marking a page

mapping both writable and executable at the same time. Use libffi

which knows how to work around it.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/6eb566a0cf43aefaf451f5885ca9b56686840a23">6eb566a0</a></strong>

<div>

<span>by Xavier Denis</span>

<i>at 2020-01-25T05:23:39-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add ghc-in-ghci for stack based builds

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/b1a32170a1b8e7a31c2dd28eb1b02f1375ea3998">b1a32170</a></strong>

<div>

<span>by Xavier Denis</span>

<i>at 2020-01-25T05:23:39-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Create ghci.cabal.sh

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/0a5e4f5f7d6cef16b6b11ac8d3a269b92016ed5d">0a5e4f5f</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-01-25T05:24:19-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Split glasgow_exts into several files (#17316)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/b3e5c678851ed73897b0eb337e656ff377d242c9">b3e5c678</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-01-25T05:24:57-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">hadrian: Throw error on duplicate-named flavours

Throw an error if the user requests a flavour for which there is more

than one match.

Fixes #17156.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/0940b59accb6926aaede045bcd5f5bdc77c7075d">0940b59a</a></strong>

<div>

<span>by Ryan Scott</span>

<i>at 2020-01-25T08:15:05-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Do not bring visible foralls into scope in hsScopedTvs

Previously, `hsScopedTvs` (and its cousin `hsWcScopedTvs`) pretended

that visible dependent quantification could not possibly happen at

the term level, and cemented that assumption with an `ASSERT`:

```hs

    hsScopedTvs (HsForAllTy { hst_fvf = vis_flag, ... }) =

      ASSERT( vis_flag == ForallInvis )

      ...

```

It turns out that this assumption is wrong. You can end up tripping

this `ASSERT` if you stick it to the man and write a type for a term

that uses visible dependent quantification anyway, like in this

example:

```hs

{-# LANGUAGE ScopedTypeVariables #-}

x :: forall a -> a -> a

x = x

```

That won't typecheck, but that's not the point. Before the

typechecker has a chance to reject this, the renamer will try

to use `hsScopedTvs` to bring `a` into scope over the body of `x`,

since `a` is quantified by a `forall`. This, in turn, causes the

`ASSERT` to fail. Bummer.

Instead of walking on this dangerous ground, this patch makes GHC

adopt a more hardline stance by pattern-matching directly on

`ForallInvis` in `hsScopedTvs`:

```hs

    hsScopedTvs (HsForAllTy { hst_fvf = ForallInvis, ... }) = ...

```

Now `a` will not be brought over the body of `x` at all (which is how

it should be), there's no chance of the `ASSERT` failing anymore (as

it's gone), and best of all, the behavior of `hsScopedTvs` does not

change. Everyone wins!

Fixes #17687.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/1132602f764ef4694b52abeaeeaa8da544915134">1132602f</a></strong>

<div>

<span>by Ryan Scott</span>

<i>at 2020-01-27T10:03:42-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Use splitLHs{ForAll,Sigma}TyInvis throughout the codebase

Richard points out in #17688 that we use `splitLHsForAllTy` and

`splitLHsSigmaTy` in places that we ought to be using the

corresponding `-Invis` variants instead, identifying two bugs

that are caused by this oversight:

* Certain TH-quoted type signatures, such as those that appear in

  quoted `SPECIALISE` pragmas, silently turn visible `forall`s into

  invisible `forall`s.

* When quoted, the type `forall a -> (a ~ a) => a` will turn into

  `forall a -> a` due to a bug in `DsMeta.repForall` that drops

  contexts that follow visible `forall`s.

These are both ultimately caused by the fact that `splitLHsForAllTy`

and `splitLHsSigmaTy` split apart visible `forall`s in addition to

invisible ones. This patch cleans things up:

* We now use `splitLHsForAllTyInvis` and `splitLHsSigmaTyInvis`

  throughout the codebase. Relatedly, the `splitLHsForAllTy` and

  `splitLHsSigmaTy` have been removed, as they are easy to misuse.

* `DsMeta.repForall` now only handles invisible `forall`s to reduce

  the chance for confusion with visible `forall`s, which need to be

  handled differently. I also renamed it from `repForall` to

  `repForallT` to emphasize that its distinguishing characteristic

  is the fact that it desugars down to `L.H.TH.Syntax.ForallT`.

Fixes #17688.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/97d0b0a367e4c6a52a17c3299439ac7de129da24">97d0b0a3</a></strong>

<div>

<span>by Matthew Pickering</span>

<i>at 2020-01-27T10:04:19-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Make Block.h compile with c++ compilers

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/4bada77d5882974514d85d4bd0fd4e1801dad755">4bada77d</a></strong>

<div>

<span>by Tom Ellis</span>

<i>at 2020-01-27T12:30:46-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Disable two warnings for files that trigger them

incomplete-uni-patterns and incomplete-record-updates will be in -Wall at a

future date, so prepare for that by disabling those warnings on files that

trigger them.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/0188404ab096747f41fdf6c92fd1e312e5575dc6">0188404a</a></strong>

<div>

<span>by Tom Ellis</span>

<i>at 2020-01-27T12:30:46-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add two warnings to stage 2 build

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/acae02c1ae8fe5fdb9966abc019ae98a3b2e51c3">acae02c1</a></strong>

<div>

<span>by Tom Ellis</span>

<i>at 2020-01-27T12:30:46-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add two warnings to Hadrian

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/bf38a20eefcaaaac404a1818c3eff8273dc67dd9">bf38a20e</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-01-31T02:46:15-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Call `interpretPackageEnv` from `setSessionDynFlags`

interpretPackageEnv modifies the flags by reading the dreaded package

environments. It is much less surprising to call it from

`setSessionDynFlags` instead of reading package environments as a

side-effect of `initPackages`.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/29c701c154cafa4844cf3c1bd4a93cacfa6b1ee1">29c701c1</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-01-31T02:46:15-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Refactor package related code

The package terminology is a bit of a mess. Cabal packages contain

components. Instances of these components when built with some

flags/options/dependencies are called units. Units are registered into

package databases and their metadata are called PackageConfig.

GHC only knows about package databases containing units. It is a sad

mismatch not fixed by this patch (we would have to rename parameters

such as `package-id <unit-id>` which would affect users).

This patch however fixes the following internal names:

- Renames PackageConfig into UnitInfo.

- Rename systemPackageConfig into globalPackageDatabase[Path]

- Rename PkgConfXX into PkgDbXX

- Rename pkgIdMap into unitIdMap

- Rename ModuleToPkgDbAll into ModuleNameProvidersMap

- Rename lookupPackage into lookupUnit

- Add comments on DynFlags package related fields

It also introduces a new `PackageDatabase` datatype instead of

explicitly passing the following tuple: `(FilePath,[PackageConfig])`.

The `pkgDatabase` field in `DynFlags` now contains the unit info for

each unit of each package database exactly as they have been read from

disk. Previously the command-line flag `-distrust-all-packages` would

modify these unit info. Now this flag only affects the "dynamic"

consolidated package state found in `pkgState` field. It makes sense

because `initPackages` could be called first with this

`distrust-all-packages` flag set and then again (using ghc-api) without

and it should work (package databases are not read again from disk when

`initPackages` is called the second time).

Bump haddock submodule

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/942c714830c73ac40e1c02bff37e2850c6794081">942c7148</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-01-31T02:46:54-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">rename: Eliminate usage of mkVarOccUnique

Replacing it with `newSysName`. Fixes #17061.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/41117d71bb58e001f6a2b6a11c9314d5b70b9182">41117d71</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-01-31T02:47:31-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">base: Use one-shot kqueue on macOS

The underlying reason requiring that one-shot usage be disabled (#13903)

has been fixed.

Closes #15768.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/01b15b835a7555c501df862b4dc8cc8eaff86afc">01b15b83</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-01-31T02:48:08-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Don't crash on encoding failure in print

If the user doesn't use a Unicode locale then the testsuite driver would

previously throw framework failures due to encoding failures. We now

rather use the `replace` error-handling strategy.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/c846618ae0f8601515683a4c7677c20c3272a50f">c846618a</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2020-01-31T12:21:10+03:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Do CafInfo/SRT analysis in Cmm

This patch removes all CafInfo predictions and various hacks to preserve

predicted CafInfos from the compiler and assigns final CafInfos to

interface Ids after code generation. SRT analysis is extended to support

static data, and Cmm generator is modified to allow generating

static_link fields after SRT analysis.

This also fixes `-fcatch-bottoms`, which introduces error calls in case

expressions in CorePrep, which runs *after* CoreTidy (which is where we

decide on CafInfos) and turns previously non-CAFFY things into CAFFY.

Fixes #17648

Fixes #9718

Evaluation

==========

NoFib

-----

Boot with: `make boot mode=fast`

Run: `make mode=fast EXTRA_RUNTEST_OPTS="-cachegrind" NoFibRuns=1`

--------------------------------------------------------------------------------

        Program           Size    Allocs    Instrs     Reads    Writes

--------------------------------------------------------------------------------

             CS          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

            CSD          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

             FS          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

              S          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

             VS          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

            VSD          -0.0%      0.0%     -0.0%     -0.0%     -0.5%

            VSM          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

           anna          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

           ansi          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

           atom          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

         awards          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

         banner          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

     bernouilli          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

   binary-trees          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

          boyer          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

         boyer2          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

           bspt          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

      cacheprof          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

       calendar          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

       cichelli          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

        circsim          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

       clausify          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

  comp_lab_zift          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

       compress          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

      compress2          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

    constraints          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

   cryptarithm1          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

   cryptarithm2          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

            cse          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

   digits-of-e1          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

   digits-of-e2          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

         dom-lt          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

          eliza          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

          event          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

    exact-reals          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

         exp3_8          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

         expert          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

 fannkuch-redux          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

          fasta          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

            fem          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

            fft          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

           fft2          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

       fibheaps          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

           fish          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

          fluid          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

         fulsom          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

         gamteb          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

            gcd          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

    gen_regexps          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

         genfft          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

             gg          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

           grep          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

         hidden          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

            hpg          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

            ida          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

          infer          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

        integer          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

      integrate          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

   k-nucleotide          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

          kahan          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

        knights          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

         lambda          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

     last-piece          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

           lcss          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

           life          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

           lift          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

         linear          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

      listcompr          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

       listcopy          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

       maillist          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

         mandel          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

        mandel2          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

           mate          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

        minimax          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

        mkhprog          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

     multiplier          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

         n-body          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

       nucleic2          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

           para          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

      paraffins          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

         parser          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

        parstof          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

            pic          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

       pidigits          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

          power          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

         pretty          -0.0%      0.0%     -0.3%     -0.4%     -0.4%

         primes          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

      primetest          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

         prolog          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

         puzzle          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

         queens          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

        reptile          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

reverse-complem          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

        rewrite          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

           rfib          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

            rsa          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

            scc          -0.0%      0.0%     -0.3%     -0.5%     -0.4%

          sched          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

            scs          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

         simple          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

          solid          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

        sorting          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

  spectral-norm          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

         sphere          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

         symalg          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

            tak          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

      transform          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

       treejoin          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

      typecheck          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

        veritas          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

           wang          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

      wave4main          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

   wheel-sieve1          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

   wheel-sieve2          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

           x2n1          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

--------------------------------------------------------------------------------

            Min          -0.1%      0.0%     -0.3%     -0.5%     -0.5%

            Max          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

 Geometric Mean          -0.0%     -0.0%     -0.0%     -0.0%     -0.0%

--------------------------------------------------------------------------------

        Program           Size    Allocs    Instrs     Reads    Writes

--------------------------------------------------------------------------------

        circsim          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

    constraints          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

       fibheaps          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

       gc_bench          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

           hash          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

           lcss          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

          power          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

     spellcheck          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

--------------------------------------------------------------------------------

            Min          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

            Max          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

 Geometric Mean          -0.0%     +0.0%     -0.0%     -0.0%     -0.0%

Manual inspection of programs in testsuite/tests/programs

---------------------------------------------------------

I built these programs with a bunch of dump flags and `-O` and compared

STG, Cmm, and Asm dumps and file sizes.

(Below the numbers in parenthesis show number of modules in the program)

These programs have identical compiler (same .hi and .o sizes, STG, and

Cmm and Asm dumps):

- Queens (1), andre_monad (1), cholewo-eval (2), cvh_unboxing (3),

  andy_cherry (7), fun_insts (1), hs-boot (4), fast2haskell (2),

  jl_defaults (1), jq_readsPrec (1), jules_xref (1), jtod_circint (4),

  jules_xref2 (1), lennart_range (1), lex (1), life_space_leak (1),

  bargon-mangler-bug (7), record_upd (1), rittri (1), sanders_array (1),

  strict_anns (1), thurston-module-arith (2), okeefe_neural (1),

  joao-circular (6), 10queens (1)

Programs with different compiler outputs:

- jl_defaults (1): For some reason GHC HEAD marks a lot of top-level

  `[Int]` closures as CAFFY for no reason. With this patch we no longer

  make them CAFFY and generate less SRT entries. For some reason Main.o

  is slightly larger with this patch (1.3%) and the executable sizes are

  the same. (I'd expect both to be smaller)

- launchbury (1): Same as jl_defaults: top-level `[Int]` closures marked

  as CAFFY for no reason. Similarly `Main.o` is 1.4% larger but the

  executable sizes are the same.

- galois_raytrace (13): Differences are in the Parse module. There are a

  lot, but some of the changes are caused by the fact that for some

  reason (I think a bug) GHC HEAD marks the dictionary for `Functor

  Identity` as CAFFY. Parse.o is 0.4% larger, the executable size is the

  same.

- north_array: We now generate less SRT entries because some of array

  primops used in this program like `NewArrayOp` get eliminated during

  Stg-to-Cmm and turn some CAFFY things into non-CAFFY. Main.o gets 24%

  larger (9224 bytes from 9000 bytes), executable sizes are the same.

- seward-space-leak: Difference in this program is better shown by this

  smaller example:

      module Lib where

      data CDS

        = Case [CDS] [(Int, CDS)]

        | Call CDS CDS

      instance Eq CDS where

        Case sels1 rets1 == Case sels2 rets2 =

            sels1 == sels2 && rets1 == rets2

        Call a1 b1 == Call a2 b2 =

            a1 == a2 && b1 == b2

        _ == _ =

            False

   In this program GHC HEAD builds a new SRT for the recursive group of

   `(==)`, `(/=)` and the dictionary closure. Then `/=` points to `==`

   in its SRT field, and `==` uses the SRT object as its SRT. With this

   patch we use the closure for `/=` as the SRT and add `==` there. Then

   `/=` gets an empty SRT field and `==` points to `/=` in its SRT

   field.

   This change looks fine to me.

   Main.o gets 0.07% larger, executable sizes are identical.

head.hackage

------------

head.hackage's CI script builds 428 packages from Hackage using this

patch with no failures.

Compiler performance

--------------------

The compiler perf tests report that the compiler allocates slightly more

(worst case observed so far is 4%). However most programs in the test

suite are small, single file programs. To benchmark compiler performance

on something more realistic I build Cabal (the library, 236 modules)

with different optimisation levels. For the "max residency" row I run

GHC with `+RTS -s -A100k -i0 -h` for more accurate numbers. Other rows

are generated with just `-s`. (This is because `-i0` causes running GC

much more frequently and as a result "bytes copied" gets inflated by

more than 25x in some cases)

* -O0

|                 | GHC HEAD       | This MR        | Diff   |

| --------------- | -------------- | -------------- | ------ |

| Bytes allocated | 54,413,350,872 | 54,701,099,464 | +0.52% |

| Bytes copied    |  4,926,037,184 |  4,990,638,760 | +1.31% |

| Max residency   |    421,225,624 |    424,324,264 | +0.73% |

* -O1

|                 | GHC HEAD        | This MR         | Diff   |

| --------------- | --------------- | --------------- | ------ |

| Bytes allocated | 245,849,209,992 | 246,562,088,672 | +0.28% |

| Bytes copied    |  26,943,452,560 |  27,089,972,296 | +0.54% |

| Max residency   |     982,643,440 |     991,663,432 | +0.91% |

* -O2

|                 | GHC HEAD        | This MR         | Diff   |

| --------------- | --------------- | --------------- | ------ |

| Bytes allocated | 291,044,511,408 | 291,863,910,912 | +0.28% |

| Bytes copied    |  37,044,237,616 |  36,121,690,472 | -2.49% |

| Max residency   |   1,071,600,328 |   1,086,396,256 | +1.38% |

Extra compiler allocations

--------------------------

Runtime allocations of programs are as reported above (NoFib section).

The compiler now allocates more than before. Main source of allocation

in this patch compared to base commit is the new SRT algorithm

(GHC.Cmm.Info.Build). Below is some of the extra work we do with this

patch, numbers generated by profiled stage 2 compiler when building a

pathological case (the test 'ManyConstructors') with '-O2':

- We now sort the final STG for a module, which means traversing the

  entire program, generating free variable set for each top-level

  binding, doing SCC analysis, and re-ordering the program. In

  ManyConstructors this step allocates 97,889,952 bytes.

- We now do SRT analysis on static data, which in a program like

  ManyConstructors causes analysing 10,000 bindings that we would

  previously just skip. This step allocates 70,898,352 bytes.

- We now maintain an SRT map for the entire module as we compile Cmm

  groups:

      data ModuleSRTInfo = ModuleSRTInfo

        { ...

        , moduleSRTMap :: SRTMap

        }

   (SRTMap is just a strict Map from the 'containers' library)

   This map gets an entry for most bindings in a module (exceptions are

   THUNKs and CAFFY static functions). For ManyConstructors this map

   gets 50015 entries.

- Once we're done with code generation we generate a NameSet from SRTMap

  for the non-CAFFY names in the current module. This set gets the same

  number of entries as the SRTMap.

- Finally we update CafInfos in ModDetails for the non-CAFFY Ids, using

  the NameSet generated in the previous step. This usually does the

  least amount of allocation among the work listed here.

Only place with this patch where we do less work in the CAF analysis in

the tidying pass (CoreTidy). However that doesn't save us much, as the

pass still needs to traverse the whole program and update IdInfos for

other reasons. Only thing we don't here do is the `hasCafRefs` pass over

the RHS of bindings, which is a stateless pass that returns a boolean

value, so it doesn't allocate much.

(Metric changes blow are all increased allocations)

Metric changes

--------------

Metric Increase:

    ManyAlternatives

    ManyConstructors

    T13035

    T14683

    T1969

    T9961

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/2a87a565365d1724a83cd0d5c5fc3b696210c4f2">2a87a565</a></strong>

<div>

<span>by Andreas Klebinger</span>

<i>at 2020-01-31T12:21:10+03:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">A few optimizations in STG and Cmm parts:

(Guided by the profiler output)

- Add a few bang patterns, INLINABLE annotations, and a seqList in a few

  places in Cmm and STG parts.

- Do not add external variables as dependencies in STG dependency

  analysis (GHC.Stg.DepAnal).

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/bef704b6263f49c96af4553407259d36494be3f0">bef704b6</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2020-02-01T02:28:45-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Improve skolemisation

This patch avoids skolemiseUnboundMetaTyVar making

up a fresh Name when it doesn't need to.

See Note [Skolemising and identity]

Improves error messsages for partial type signatures.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/cd11042337a829da1dfbd19ca1a46feabdd23147">cd110423</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2020-02-01T02:28:45-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Improve pretty-printing for TyConBinders

In particular, show their kinds.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/913287a0fa5370a2488ce560f2dfba61db51055d">913287a0</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2020-02-01T02:28:45-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix scoping of TyCon binders in TcTyClsDecls

This patch fixes #17566 by refactoring the way we decide the final

identity of the tyvars in the TyCons of a possibly-recursive nest

of type and class decls, possibly with associated types.

It's all laid out in

  Note [Swizzling the tyvars before generaliseTcTyCon]

Main changes:

* We have to generalise each decl (with its associated types)

  all at once: TcTyClsDecls.generaliseTyClDecl

* The main new work is done in TcTyClsDecls.swizzleTcTyConBndrs

* The mysterious TcHsSyn.zonkRecTyVarBndrs dies altogether

Other smaller things:

* A little refactoring, moving bindTyClTyVars from tcTyClDecl1

  to tcDataDefn, tcSynRhs, etc.  Clearer, reduces the number of

  parameters

* Reduce the amount of swizzling required.

  Specifically, bindExplicitTKBndrs_Q_Tv doesn't need

    to clone a new Name for the TyVarTv, and not

    cloning means that in the vasly common case,

    swizzleTyConBndrs is a no-op

  In detail:

    Rename newTyVarTyVar --> cloneTyVarTyVar

    Add newTyVarTyTyVar that doesn't clone

    Use the non-cloning newTyVarTyVar in

       bindExplicitTKBndrs_Q_Tv

       Rename newFlexiKindedTyVarTyVar

           --> cloneFlexiKindedTyVarTyVar

* Define new utility function and use it

     HsDecls.familyDeclName ::

        FamilyDecl (GhcPass p) -> IdP (GhcPass p)

Updates haddock submodule.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/58ed6c4a0999c0025b1b024bc26171fa6d6773b3">58ed6c4a</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-01T02:29:23-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">rts/M32Alloc: Don't attempt to unmap non-existent pages

The m32 allocator's `pages` list may contain NULLs in the case that the

page was flushed. Some `munmap` implementations (e.g. FreeBSD's) don't

like it if we pass them NULL. Don't do that.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/859db7d639da08622a76a94b85bc9b2b27b8aabc">859db7d6</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2020-02-01T14:18:49+03:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Improve/fix -fcatch-bottoms documentation

Old documentation suggests that -fcatch-bottoms only adds a default

alternative to bottoming case expression, but that's not true. We use a

very simplistic "is exhaustive" check and add default alternatives to

any case expression that does not cover all constructors of the type. In

case of GADTs this simple check assumes all constructors should be

covered, even the ones ruled out by the type of the scrutinee.

Update the documentation to reflect this.

(Originally noticed in #17648)

[ci skip]

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/54dfa94a36a564e5d092aa566d4670c7e008f152">54dfa94a</a></strong>

<div>

<span>by John Ericson</span>

<i>at 2020-02-03T21:14:24-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix docs for FrontendResult

Other variant was removed in ac1a379363618a6f2f17fff65ce9129164b6ef30

but docs were no changed.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/5e63d9c07c0585b85c8fa340d30aeff0130af3f4">5e63d9c0</a></strong>

<div>

<span>by John Ericson</span>

<i>at 2020-02-03T21:15:02-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Refactor HscMain.finish

I found the old control flow a bit hard to follow; I rewrote it to first

decide whether to desugar, and then use that choice when computing

whether to simplify / what sort of interface file to write.

I hope eventually we will always write post-tc interface files, which

will make the logic of this function even simpler, and continue the

thrust of this refactor.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/e580e5b8d08b02c39ca49e387882196931163bcf">e580e5b8</a></strong>

<div>

<span>by Stefan Schulze Frielinghaus</span>

<i>at 2020-02-04T09:29:00-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Do not build StgCRunAsm.S for unregisterised builds

For unregisterised builds StgRun/StgReturn are implemented via a mini

interpreter in StgCRun.c and therefore would collide with the

implementations in StgCRunAsm.S.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/e3b0bd97cb92706b3cee38b2ec13349ca22ecf7d">e3b0bd97</a></strong>

<div>

<span>by Stefan Schulze Frielinghaus</span>

<i>at 2020-02-04T09:29:00-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">fixup! fixup! Do not build StgCRunAsm.S for unregisterised builds

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/eb629fab96fbff43f79190767731501d8642f524">eb629fab</a></strong>

<div>

<span>by John Ericson</span>

<i>at 2020-02-04T09:29:38-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Delete some superfluous helper functions in HscMain

The driver code is some of the nastiest in GHC, and I am worried about

being able to untangle all the tech debt. In `HscMain` we have a number

of helpers which are either not-used or little used. I delete them so we

can reduce cognative load, distilling the essential complexity away from

the cruft.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/c90eca550ccf89c8eb880e4428a43a9d0e8203f0">c90eca55</a></strong>

<div>

<span>by Sebastian Graf</span>

<i>at 2020-02-05T09:21:29-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">PmCheck: Record type constraints arising from existentials in `PmCoreCt`s

In #17703 (a follow-up of !2192), we established that contrary to my

belief, type constraints arising from existentials in code like

```hs

data Ex where Ex :: a -> Ex

f _ | let x = Ex @Int 15 = case x of Ex -> ...

```

are in fact useful.

This commit makes a number of refactorings and improvements to comments,

but fundamentally changes `addCoreCt.core_expr` to record the type

constraint `a ~ Int` in addition to `x ~ Ex @a y` and `y ~ 15`.

Fixes #17703.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/6d3b5d5706a8fcfe724ee0a042d5f7c18880ccba">6d3b5d57</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2020-02-05T09:22:10-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testlib: Extend existing *_opts in extra_*_opts

Previously we'd override the existing {run,hc} opts in

extra_{run,hc}_opts, which caused flakiness in T1969, see #17712.

extra_{run,hc}_opts now extends {run,hc} opts, instead of overriding.

Also we shrank the allocation area for T1969 in order to increase

residency sampling frequency.

Fixes #17712

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/9c89a48dd88b566a23b670ce707208e8c4cf28c1">9c89a48d</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2020-02-05T09:22:52-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Remove CafInfo-related code from STG lambda lift pass

After c846618ae0 we don't have accurate CafInfos for Ids in the current

module and we're free to introduce new CAFFY or non-CAFFY bindings or

change CafInfos of existing binders; so no we no longer need to

maintain CafInfos in Core or STG passes.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/70ddb8bf3066a2e70cffb49b100b640830bec7e1">70ddb8bf</a></strong>

<div>

<span>by Ryan Scott</span>

<i>at 2020-02-05T09:23:30-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add regression test for #17773

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/e8004e5d1e993363a89b0107380604c5bc02be6b">e8004e5d</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-05T13:55:19-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Allow Windows builds to fail again

Due to T7702 and the process issues described in #17777.</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/29b72c00a692b442605b022fcc15c4454ed1c3de">29b72c00</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-06T11:55:41-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">VarSet: Introduce nonDetFoldVarSet

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/c4e6b35daef9336ea6b0922bab595cb02ac742b5">c4e6b35d</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-06T11:55:41-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Move closeOverKinds and friends to TyCoFVs

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/ed2f0e5c38f1dfc9cf049ae63c56117cda99116e">ed2f0e5c</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2020-02-06T11:55:41-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Reform the free variable finders for types

This patch delivers on (much of) #17509.

* Introduces the shallow vs deep free variable distinction

* Introduce TyCoRep.foldType,

    foldType :: Monoid a => TyCoFolder env a

                         -> env -> Type -> a

  and use it in the free variable finders.

* Substitution in TyCoSubst

   * ASSERTs are on for checkValidSubst

   * checkValidSubst uses shallowTyCoVarsOfTypes etc

Quite a few things still to do

* We could use foldType in lots of other places

* We could use mapType for substitution.  (Check that we get

  good code!)

* Some (but not yet all) clients of substitution can now

  save time by using shallowTyCoVarsOfTypes

* All calls to tyCoVarsOfTypes should be inspected; most of

  them should be shallow.  Maybe.

* Currently shallowTyCoVarsOfTypes still returns

  unification variables, but not CoVarHoles.

  Reason: we need to return unification variables

  in some of the calls in TcSimplify, eg when promoting.

* We should do the same thing for tyCoFVsOfTypes, which is

  currently unchanged.

* tyCoFVsOfTypes returns CoVarHoles, because of the

  use in TcSimplify.mkResidualConstraints.  See

  Note [Emitting the residual implication in simplifyInfer]

* #17509 talks about "relevant" variables too.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/01a1f4fbb1bb62ed621421622f6a15d30c91b3c3">01a1f4fb</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2020-02-06T11:55:41-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Use foldTyCo for noFreeVarsOfType

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/0e59afd66788fc704999d1987cd351fa165d3c46">0e59afd6</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2020-02-06T11:55:41-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Simplify closeOverKinds

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/9ca5c88e6370c2cafed6b63fe217f70e87d7fcea">9ca5c88e</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2020-02-06T11:55:41-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Use foldTyCo for coVarsOfType

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/5541b87c498cdd320e9a18d2134e571d713536e1">5541b87c</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2020-02-06T11:55:41-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Use foldTyCo for exactTyCoVarsOfType

This entailed

* Adding a tcf_view field to TyCoFolder

* Moving exactTyCoVarsOtType to TcType.  It properly belongs

  there, since only the typechecker calls this function. But

  it also means that we can "see" and inline tcView.

Metric Decrease:

  T14683

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/7c1228511f2dd4d262c04edb8539174a7de810b2">7c122851</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2020-02-06T11:56:02-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Comments only

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/588acb99bc3cb377ceb76447dd60656b4a11de5a">588acb99</a></strong>

<div>

<span>by Adam Sandberg Eriksson</span>

<i>at 2020-02-08T10:15:38-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">slightly better named cost-centres for simple pattern bindings #17006

```

main = do

  print $ g [1..100] a

  where g xs x   = map (`mod` x) xs

        a :: Int = 324

```

The above program previously attributed the cost of computing 324 to a cost

centre named `(...)`, with this change the cost is attributed to `a` instead.

This change only affects simple pattern bindings (decorated variables: type

signatures, parens, ~ annotations and ! annotations).

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/309f8cfdad9cf81f5ee6003821810ea1205ae1d5">309f8cfd</a></strong>

<div>

<span>by Richard Eisenberg</span>

<i>at 2020-02-08T10:16:33-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Remove unnecessary parentheses

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/7755ffc2920facb7ed74efe379ad825feeaf1024">7755ffc2</a></strong>

<div>

<span>by Richard Eisenberg</span>

<i>at 2020-02-08T10:16:33-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Introduce IsPass; refactor wrappers.

There are two main payloads of this patch:

1. This introduces IsPass, which allows e.g. printing

   code to ask what pass it is running in (Renamed vs

   Typechecked) and thus print extension fields. See

   Note [IsPass] in Hs.Extension

2. This moves the HsWrap constructor into an extension

   field, where it rightly belongs. This is done for

   HsExpr and HsCmd, but not for HsPat, which is left

   as an exercise for the reader.

There is also some refactoring around SyntaxExprs, but this

is really just incidental.

This patch subsumes !1721 (sorry @chreekat).

Along the way, there is a bit of refactoring in GHC.Hs.Extension,

including the removal of NameOrRdrName in favor of NoGhcTc.

This meant that we had no real need for GHC.Hs.PlaceHolder, so

I got rid of it.

Updates haddock submodule.

-------------------------

Metric Decrease:

    haddock.compiler

-------------------------

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/7d452be454857549679b93a0682a3f6fedf5d7c1">7d452be4</a></strong>

<div>

<span>by Dylan Yudaken</span>

<i>at 2020-02-08T10:17:17-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix hs_try_putmvar losing track of running cap

If hs_try_putmvar was called through an unsafe import, it would lose track of the running cap causing a deadlock

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/c2e301aeeae353a64be43e5fa9e7d464797d5648">c2e301ae</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-08T10:17:55-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">compiler: Qualify imports of Data.List

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/aede171a59f9b7b8022548c385a1cb8c4589f905">aede171a</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-08T10:17:55-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Fix -Wcompat-unqualified-imports issues

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/4435a8e0b74337fe5faddb9c46691f0f5bf9e653">4435a8e0</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-08T10:17:55-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Introduce -Wcompat-unqualified-imports

This implements the warning proposed in option (B) of the

Data.List.singleton CLC [discussion][].

This warning, which is included in `-Wcompat` is intended to help users

identify imports of modules that will change incompatibly in future GHC

releases. This currently only includes `Data.List` due to the expected

specialisation and addition of `Data.List.singleton`.

Fixes #17244.

[discussion]: https://groups.google.com/d/msg/haskell-core-libraries/q3zHLmzBa5E/PmlAs_kYAQAJ

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/28b5349a1ae88872e343c58c270fe90329f14d87">28b5349a</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-08T10:17:55-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Bump stm and process submodules

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/7d04b9f298c9cc0ff291e0717826743f488670bb">7d04b9f2</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-08T10:18:31-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">hadrian: Allow override of Cabal configuration in hadrian.settings

Fixes #17612 by adding a `cabal.configure.opts` key for

`hadrian.settings`.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/88bf81aaaf67e9c9bb5b12088ab33accd4a55fb3">88bf81aa</a></strong>

<div>

<span>by Andreas Klebinger</span>

<i>at 2020-02-08T10:19:10-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Optimize unpackCString# to allocate less.

unpackCString# is a recursive function which for each iteration

returns a Cons cell containing the current Char, and a thunk for

unpacking the rest of the string.

In this patch we change from storing addr + offset inside this thunk

to storing only the addr, simply incrementing the address on each

iteration.

This saves one word of allocation per unpacked character.

For a program like "main = print "<largishString>" this amounts

to 2-3% fewer % in bytes allocated.

I also removed the now redundant local unpack definitions.

This removes one call per unpack operation.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/bec76733b818b0489ffea0834ab6b1560207577c">bec76733</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-08T10:19:57-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix GhcThreaded setting

This adopts a patch from NetBSD's packaging fixing the `GhcThreaded`

option of the make build system. In addition we introduce a `ghcThreaded`

option in hadrian's `Flavour` type.

Also fix Hadrian's treatment of the `Use Threaded` entry in `settings`.

Previously it would incorrectly claim `Use Threaded = True` if we were

building the `threaded` runtime way. However, this is inconsistent with

the `make` build system, which defines it to be whether the `ghc`

executable is linked against the threaded runtime.

Fixes #17692.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/545cf1e184a37295245086040750d77a809db16e">545cf1e1</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-08T10:20:37-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">hadrian: Depend upon libray dependencies when configuring packages

This will hopefully fix #17631.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/047d3d75b2861e6dd3c0ba21daf50b012e5a7fd3">047d3d75</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-08T10:21:16-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Add test for #15316

This is the full testcase for T15316.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/768e58664ffd408e5c4e3bea8f00a650cd0c6501">768e5866</a></strong>

<div>

<span>by Julien Debon</span>

<i>at 2020-02-08T10:22:07-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">doc(Data.List): Add some examples to Data.List

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/3900cb83fb10b0209fb5fb1d713bf3aaf3ee42b8">3900cb83</a></strong>

<div>

<span>by Julien Debon</span>

<i>at 2020-02-08T10:22:07-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Apply suggestion to libraries/base/GHC/List.hs</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/bd6667665f160f07550b25e9c80fbc197ae2f642">bd666766</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-08T10:22:45-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">users-guide: Clarify that bundled patsyns were introduced in GHC 8.0

Closes #17094.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/95741ea1f93c6d9bac055eb620c8bcda0ec5c72d">95741ea1</a></strong>

<div>

<span>by Pepe Iborra</span>

<i>at 2020-02-08T10:23:23-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Update to hie-bios 0.3.2 style program cradle

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/fb5c19122ea59d1c6758762cb47e1a451bc3b9bf">fb5c1912</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-02-08T10:24:07-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Remove redundant case

This alternative is redundant and triggers no warning when building with 8.6.5

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/5d83d9489f383cc940ce123d9385c5cad1af517d">5d83d948</a></strong>

<div>

<span>by Matthew Pickering</span>

<i>at 2020-02-08T10:24:43-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add mkHieFileWithSource which doesn't read the source file from disk

cc/ @pepeiborra

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/dfdae56d2ea9b9841e747c36cdff63a826045805">dfdae56d</a></strong>

<div>

<span>by Andreas Klebinger</span>

<i>at 2020-02-08T10:25:20-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Rename ghcAssert to stgAssert in hp2ps/Main.h.

This fixes #17763</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/658f7ac67472a880e19397773edd35a6214ed995">658f7ac6</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-08T10:26:00-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">includes: Avoid using single-line comments in HsFFI.h

While single-line comments are supported by C99, dtrace on SmartOS

apparently doesn't support them yet.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/c95920a6371768276e439811e99f4822571ff4df">c95920a6</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2020-02-08T10:26:42-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Import qualified Prelude in parser

This is in preparation of backwards-incompatible changes in happy.

See https://github.com/simonmar/happy/issues/166

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/b6dc319a2d83a8a8c426d0ad6f46d0b8fad41253">b6dc319a</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2020-02-08T10:27:23-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add regression test for #12760

The bug seems to be fixed in the meantime, make sure it stays fixed.

Closes #12760

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/b3857b62cb74ef50cbe88ab898aae946ca933365">b3857b62</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-08T10:28:03-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">base: Drop out-of-date comment

The comment in GHC.Base claimed that ($) couldn't be used in that module

as it was wired-in. However, this is no longer true; ($) is merely known

key and is defined in Haskell (with a RuntimeRep-polymorphic type) in

GHC.Base.

The one piece of magic that ($) retains is that it a special typing

rule to allow type inference with higher-rank types

(e.g. `runST $ blah`; see Note [Typing rule for ($)] in TcExpr).

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/1183ae94a256b30bfe12ddc5e1c29d6f46abd79d">1183ae94</a></strong>

<div>

<span>by Daniel Gröber</span>

<i>at 2020-02-08T10:29:00-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">rts: Fix Arena blocks accounting for MBlock sized allocations

When requesting more than BLOCKS_PER_MBLOCK blocks allocGroup can return a

different number of blocks than requested. Here we use the number of

requested blocks, however arenaFree will subtract the actual number of

blocks we got from arena_blocks (possibly) resulting in a negative value

and triggering ASSERT(arena_blocks >= 0).

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/97d59db55e3b32861d91783c36d56e5c93f2cd19">97d59db5</a></strong>

<div>

<span>by Daniel Gröber</span>

<i>at 2020-02-08T10:29:48-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">rts: Fix need_prealloc being reset when retainer profiling is on

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/1f630025d457807c4a9da513e4a0823e4643b2a6">1f630025</a></strong>

<div>

<span>by Krzysztof Gogolewski</span>

<i>at 2020-02-09T02:52:27-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add a test for #15712

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/2ac784ab381aae936ee3be7c8685e0203ee4dbf6">2ac784ab</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-09T02:53:05-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">hadrian: Add --test-metrics argument

Allowing the test metric output to be captured to a file, a la

the METRIC_FILE environment variable of the make build system.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/f432d8c64058d3d00d917f0bc65feb1bf9f7aecb">f432d8c6</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-09T02:53:05-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">hadrian: Fix --test-summary argument

This appears to be a cut-and-paste error.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/a906595f060b9c038332f16adaf664eeb1d50f2e">a906595f</a></strong>

<div>

<span>by Arnaud Spiwack</span>

<i>at 2020-02-09T02:53:50-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix an outdated note link

This link appears to have been forgotten in

0dad81ca5fd1f63bf8a3b6ad09787559e8bd05c0 .

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/3ae83da13f03d22a4d810b5d9c0f37f818cb6a00">3ae83da1</a></strong>

<div>

<span>by Alp Mestanogullari</span>

<i>at 2020-02-09T02:54:28-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">hadrian: Windows fixes (bindists, CI)

This commit implements a few Windows-specific fixes which get us from a CI

job that can't even get as far as starting the testsuite driver, to a state

where we can run the entire testssuite (but have test failures to fix).

- Don't forget about a potential extension for the haddock program, when

  preparing the bindist.

- Build the timeout program, used by the testsuite driver on Windows in place

  of the Python script used elsewhere, using the boot compiler. We could

  alternatively build it with the compiler that we're going to test but this

  would be a lot more tedious to write.

- Implement a wrapper-script less installation procedure for Windows, in

  `hadrian/bindist/Makefile.

- Make dependencies a bit more accurate in the aforementioned Makefile.

- Update Windows/Hadrian CI job accordingly.

This patch fixes #17486.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/82f9be8c6aafb70f612e72bef517c76437726c78">82f9be8c</a></strong>

<div>

<span>by Roland Senn</span>

<i>at 2020-02-09T02:55:06-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix #14628: Panic (No skolem Info) in GHCi

This patch implements the [sugggestion from Simon (PJ)](https://gitlab.haskell.org/ghc/ghc/issues/14628#note_146559):

- Make `TcErrors.getSkolemInfo` return a `SkolemInfo` rather than an `Implication`.

- If `getSkolemInfo` gets `RuntimeUnk`s, just return a new data constructor in `SkolemInfo`, called `RuntimeUnkSkol`.

- In `TcErrors.pprSkols` print something sensible for a `RuntimeUnkSkol`.

The `getSkolemInfo` function paniced while formating suggestions to add type annotations (subfunction `suggestAddSig`)

to a *"Couldn't match type ‘x’ with ‘y’"* error message.

The `getSkolemInfo` function didn't find any Implication value and paniced.

With this patch the `getSkolemInfo` function does no longer panic, if it finds `RuntimeUnkSkol`s.

As the panic occured while processing an error message, we don't need to implement any new error message!

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/b2e18e26c34424ab814453a73404429cdf1a4d4a">b2e18e26</a></strong>

<div>

<span>by Andreas Klebinger</span>

<i>at 2020-02-09T02:55:46-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix -ddump-stg-final.

Once again make sure this dumps the STG used for codegen.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/414e2f626321f8eba8b2d5cc38dcc1a528351d83">414e2f62</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-02-09T02:56:26-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Force -fPIC for intree GMP (fix #17799)

Configure intree GMP with `--with-pic` instead of patching it. Moreover

the correct patching was only done for x86_64/darwin (see #17799).

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/f0fd72ee042aa011af7d870febcfead01a424370">f0fd72ee</a></strong>

<div>

<span>by Sebastian Graf</span>

<i>at 2020-02-09T17:22:38-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">8.10 Release notes for improvements to the pattern-match checker [skip ci]

A little late to the game, but better late than never.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/00dc0f7e66a18ab50931325cb6333a2bfd2a6650">00dc0f7e</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2020-02-09T17:23:17-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add regression test for #13142

Closes #13142

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/f3e737bbcae7290856450ff9c8b0c7e31fca0c10">f3e737bb</a></strong>

<div>

<span>by Sebastian Graf</span>

<i>at 2020-02-10T20:04:09-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix long distance info for record updates

For record updates where the `record_expr` is a variable, as in #17783:

```hs

data PartialRec = No

                | Yes { a :: Int, b :: Bool }

update No = No

update r@(Yes {}) = r { b = False }

```

We should make use of long distance info in

`-Wincomplete-record-updates` checking. But the call to `matchWrapper`

in the `RecUpd` case didn't specify a scrutinee expression, which would

correspond to the `record_expr` `r` here. That is fixed now.

Fixes #17783.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/5670881d7779ecd7eee8c969dab66ee343298532">5670881d</a></strong>

<div>

<span>by Tamar Christina</span>

<i>at 2020-02-10T20:05:04-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fs: Fix UNC remapping code.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/375b3c45be0e5de673d03097ebbad442c85c89eb">375b3c45</a></strong>

<div>

<span>by Oleg Grenrus</span>

<i>at 2020-02-11T05:07:30-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add singleton to Data.OldList

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/de32beffde82eec954691703541a2d4081799453">de32beff</a></strong>

<div>

<span>by Richard Eisenberg</span>

<i>at 2020-02-11T05:08:10-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Do not create nested quantified constraints

Previously, we would accidentally make constraints like

forall a. C a => forall b. D b => E a b c as we traversed

superclasses. No longer!

This patch also expands Note [Eagerly expand given superclasses]

to work over quantified constraints; necessary for T16502b.

Close #17202 and #16502.

test cases: typecheck/should_compile/T{17202,16502{,b}}

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/e319570ec141e0f63d82d23b88a40f96fc0b7105">e319570e</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-11T05:08:47-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">rts: Use nanosleep instead of usleep

usleep was removed in POSIX.1-2008.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/b75e7486de73868e6821a9f680e9e13794dff15c">b75e7486</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-11T05:09:24-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">rts: Remove incorrect assertions around MSG_THROWTO messages

Previously we would assert that threads which are sending a

`MSG_THROWTO` message must have their blocking status be blocked on the

message. In the usual case of a thread throwing to another thread this

is guaranteed by `stg_killThreadzh`.  However, `throwToSelf`, used by

the GC to kill threads which ran out of heap, failed to guarantee this.

Noted while debugging #17785.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/aba51b65863db5f2687f29d10ba24ffa49dabd3b">aba51b65</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-02-11T05:10:04-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add arithmetic exception primops (#14664)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/b157399f987c31574968719a52a5b00f2acfb36a">b157399f</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-11T05:10:40-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">configure: Don't assume Gnu linker on Solaris

Compl Yue noticed that the linker was dumping the link map on SmartOS. This is

because Smartos uses the Solaris linker, which uses the `-64` flag, not

`-m64` like Gnu ld, to indicate that it should link for 64-bits. Fix the

configure script to handle the Solaris linker correctly.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/d8d73d779e615ae352934864a347222f175ca277">d8d73d77</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2020-02-11T05:11:18-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Notes only: telescopes

This documentation-only patch fixes #17793

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/58a4ddeff7730d160dd66f19c288f8b5b27679e3">58a4ddef</a></strong>

<div>

<span>by Alp Mestanogullari</span>

<i>at 2020-02-11T05:12:17-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">hadrian: build (and ship) iserv on Windows

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/82023524ff050f26bf00be3432a97f1e537caf41">82023524</a></strong>

<div>

<span>by Matthew Pickering</span>

<i>at 2020-02-11T18:04:17-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">TemplateHaskellQuotes: Allow nested splices

There is no issue with nested splices as they do not require any compile

time code execution. All execution is delayed until the top-level

splice.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/50e24eddbac3d3b34003ad985fe6126aa0f83e0f">50e24edd</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2020-02-11T18:04:57-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Remove Hadrian's copy of (Data.Functor.<&>)

The function was added to base with base-4.11 (GHC 8.4)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/f82a2f90ceda5c2bc74088fa7f6a7c8cb9c9756f">f82a2f90</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-02-12T01:56:46-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Document GMP build [skip ci]

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/da7f74797e8c322006eba385c9cbdce346dd1d43">da7f7479</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-02-12T01:57:27-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Module hierarchy: ByteCode and Runtime (cf #13009)

Update haddock submodule

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/04f51297e3d7bcf7be0dc7db4673cc3e6f9156ea">04f51297</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2020-02-12T01:58:11-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix naming of tests for #12923

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/31fc332191a6298e2643261482e9b43a4b38241b">31fc3321</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2020-02-12T01:58:11-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add regression test for #12926

Closes #12926

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/f0c0ee7d9a942a19361e72553cd08f42cc12b04a">f0c0ee7d</a></strong>

<div>

<span>by Krzysztof Gogolewski</span>

<i>at 2020-02-12T01:58:51-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix order of arguments in specializer (#17801)

See https://gitlab.haskell.org/ghc/ghc/issues/17801#note_253330

No regression test, as it's hard to trigger.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/059c3c9d7c84fc37c69e9f414ff736d47081e72c">059c3c9d</a></strong>

<div>

<span>by Sebastian Graf</span>

<i>at 2020-02-12T11:00:58+01:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Separate CPR analysis from the Demand analyser

The reasons for that can be found in the wiki:

https://gitlab.haskell.org/ghc/ghc/wikis/nested-cpr/split-off-cpr

We now run CPR after demand analysis (except for after the final demand

analysis run just before code gen). CPR got its own dump flags

(`-ddump-cpr-anal`, `-ddump-cpr-signatures`), but not its own flag to

activate/deactivate. It will run with `-fstrictness`/`-fworker-wrapper`.

As explained on the wiki page, this step is necessary for a sane Nested

CPR analysis. And it has quite positive impact on compiler performance:

Metric Decrease:

    T9233

    T9675

    T9961

    T15263

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/f5ffd8d9ec776db708e690c4fdbf671afa8df48f">f5ffd8d9</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-12T17:22:37-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">base: Expose GHC.Unicode.unicodeVersion

This exposes a Data.Version.Version representing the version of the

Unicode database used by `base`. This should clear up some confusion I

have seen in tickets regarding with which Unicode versions a given GHC

can be expected to work.

While in town I also regenerated (but did not update) the Unicode

database with database 12.0.0. Strangely, the file cited in the README

no longer existed. Consequently, I used

https://www.unicode.org/Public/12.0.0/ucd/UnicodeData.txt and was

slightly surprised to find that there were a few changes.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/6c2585e0816cb49487fa6a2be5b8e3f191cf3438">6c2585e0</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-12T17:22:37-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">base: Update Unicode database to 12.1.0

Using `curl https://www.unicode.org/Public/12.1.0/ucd/UnicodeData.txt |

libraries/base/cbits/ubconfc 12.1.0`.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/df08468113ab46832b7ac0a7311b608d1b418c4d">df084681</a></strong>

<div>

<span>by Krzysztof Gogolewski</span>

<i>at 2020-02-12T23:58:52+01:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Always display inferred variables using braces

We now always show "forall {a}. T" for inferred variables,

previously this was controlled by -fprint-explicit-foralls.

This implements part 1 of https://github.com/ghc-proposals/ghc-proposals/pull/179.

Part of GHC ticket #16320.

Furthermore, when printing a levity restriction error, we now display

the HsWrap of the expression. This lets users see the full elaboration with

-fprint-typechecker-elaboration (see also #17670)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/16d643cfe661716c7a6cb3ce04893a732c5a42af">16d643cf</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-02-13T09:16:04-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Remove -ddump-srts flag

This flag is deemed not useful.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/fa28ae95e4c1ff5b70bf3f8dc6e54fb994bbfa51">fa28ae95</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-02-13T09:16:04-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix flag documentation (#17826)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/1bfd825943f6077836361ddc788fb40f2421e283">1bfd8259</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-02-13T09:16:43-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Ensure that Hadrian is built correctly before using it

When Hadrian failed to build, the script would pick a previously built

Hadrian (if available) instead of failing.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/cd6e786a28cc6688b45c8868a3e8a338b081ba01">cd6e786a</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2020-02-14T05:29:56-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add test for #17648

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/9f2c3677b3a08d8ea6c0d1f5fd51e43b7fcfe75b">9f2c3677</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-02-14T05:30:39-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">GMP expects the Target platform as --host parameter

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/aa6086fdf53aef690369771def6cf1c31faa38e8">aa6086fd</a></strong>

<div>

<span>by Oleg Grenrus</span>

<i>at 2020-02-14T05:31:16-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add explicit LANGUAGE Safe to template-haskell

(cherry picked from commit a5e0f376821ca882880b03b07b451aa574e289ec)</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/af6a0c36431639655762440ec8d652796b86fe58">af6a0c36</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-14T05:31:53-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">hadrian: Add execution and target architecture to stage-compilation figure

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/cf739945b8b28ff463dc44925348f20b3c1f22cb">cf739945</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-02-14T05:32:37-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Module hierarchy: HsToCore (cf #13009)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/719db3180e3fb0fa95fcfc90feb51ffbb2a2096d">719db318</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2020-02-14T05:33:16-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">De-duplicate overlapping Notes

Documentation only.  Fixes #17827

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/7550417ac866e562bb015149d8f9a6b8c97b5f84">7550417a</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-02-14T05:33:56-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Hadrian: drop Sphinx flag checking for PDF documentation (#17825)

It seems that Sphinx produces the ghc-flags.txt in

doc/users_guide/_build rather than pdfRoot. We could copy ghc-flags.txt

into pdfRoot (like happens naturally in the HTML case) but the benefit

is pretty small. Let's just only check the HTML case.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/813842f42023457ccb4d93b5906e0f492e4721b2">813842f4</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-14T10:16:36-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">make: Be more selective in building windows-extra-src tarball

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/0725f4bbc7f59282ee5fe41619099957030d85ff">0725f4bb</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-14T10:16:36-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Rework handling of win32 toolchain tarballs

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/565ce7aee5db966070f9d0267f99a13532d5caf0">565ce7ae</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-14T10:16:36-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Consolidate CI logic

This moves nearly all of the CI logic to .gitlab/ci.sh. This improves

things in a number of ways:

 * it's harder for inconsistencies to arise between architectures

 * it's easier to share logic between architectures

 * on Windows, it's easier to ensure that all CI steps are executed from

   within a properly initialized mingw session.

While in town I also add a FreeBSD build job and update the Windows job

to use the gitlab-runner PowerShell executor, since cmd.exe will be

deprecated soon (fixing #17699).

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/9cbace74dadf6fe64267f0b68635402f5c848b72">9cbace74</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-14T10:16:36-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Deduplicate nightly job configuration

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/6e8371442dc6130a03e648a0e1bc0f1bdfdbfe91">6e837144</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-14T10:16:36-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">integer-gmp: Fix unused command-line argument

-L is only needed during linking.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/e5ee07ab565c1bf7a33d3bd4bcd3fb6c7b100d1c">e5ee07ab</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-14T10:16:36-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Don't ask sed to operate in-place on symlinks

Some sed implementations (e.g. FreeBSD) refuse to operate in-place on

symlinks.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/71e5e68fd3794cc63f513d30d467e43761a8ce62">71e5e68f</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-14T10:16:36-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Disable tests that assume name of libstdc++ on FreeBSD

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/7b2da0f429f45ad2d945b4b041e1d58477ce7567">7b2da0f4</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-14T10:16:36-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Mark T6132 as broken on FreeBSD

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/8ef7a15a0c59b4ceff7f6343cabca8497c0eef50">8ef7a15a</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-14T10:16:36-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite/T16930: Don't rely on gnu grep specific --include

In BSD grep this flag only affects directory recursion.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/6060003e4f95101adaef0b2c7f26154c59f2e83a">6060003e</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-14T10:16:36-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Pass -Wno-unused-command-line-arguments during link on FreeBSD

FreeBSD cc throws a warning if we pass -pthread without actually using

any pthread symbols.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/97497baeb70c443030e9e0c59abf4930679b4f82">97497bae</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-14T10:16:36-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">base: Always clamp reads/writes to 2GB in length

Previously we did this only on Darwin due to #17414. However, even on

other platforms >2GB writes are on shaky ground. POSIX explicitly says

that the result is implementation-specified and Linux will write at most

0x7ffff000, even on 64-bit platforms. Moreover, getting the sign

of the syscall result correct is tricky, as demonstrated by the fact

that T17414 currently fails on FreeBSD.

For simplicity we now just uniformly clamp to 0x7ffff000 on all

platforms.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/49be2a3fc5e9f6ca71717cde6889c49dbba816da">49be2a3f</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-14T10:16:36-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">configure: Fix sphinx version test

The check for the "v" prefix is redundant.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/f7f7a556f9b89303da4e1c3475e618653d830cc4">f7f7a556</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-14T10:16:37-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">users-guide: Fix unknown link targets

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/a204102c634743faac8af345ad8bfec6317e7024">a204102c</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-14T10:16:37-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">docs/compare-flags: Don't use python f-strings

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/92e15a379a675f14667d7b80db608c84e58aa0a8">92e15a37</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-14T10:16:37-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Fix various shellcheck warnings

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/459f7c6e29eed46f8a5be8fac02c45450080a67c">459f7c6e</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-14T10:16:37-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">hadrian: Drop empty arguments from target list

Fixes #17748.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/c06df28dd490d0309e0b1885a6121b45f74edf3c">c06df28d</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-14T10:16:37-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">users-guide: Fix "invalid file" failure

I have no idea how this worked previously. Different Python version?

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/3fe8444f7e3904be14956f3fe68583da8b730ab4">3fe8444f</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-14T10:16:59-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Mark T7702 as fragile on Windows

Due to #16799. There was previously an attempt to mark it as broken but

the `opsys` name was incorrect.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/fe02f78172f06fd3e83ba00246d9ed741474e653">fe02f781</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-14T10:16:59-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Assert the opsys names are known

Previously opsys would take any string. This meant it was very easy for

a typo to silently render the predicate ineffective. Fix this by

checking the given operating system name against a list of known values.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/149e2a3a85d92b4afea1643e7e6ed7861d620be2">149e2a3a</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-14T10:16:59-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">compare-flags: Don't rely on encoding flag of subprocess.check_output

Apparently it isn't supported by some slightly older Python versions.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/798d59f6a2b517db50ec2536a8331d610a2b379c">798d59f6</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-14T10:16:59-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">rts: Add more debug output to failed path in onIOComplete

This will help track down #17035.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/e35f3f98b085a6bd925bef1b97f81a3117358e8c">e35f3f98</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-14T10:16:59-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Allow i386 Windows builds to fail again

Due to the resistance of #17736 to resolution.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/261a3cf83b757f6422acafa6e436ea7bb05ff532">261a3cf8</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-14T10:17:00-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Build integer-simple job in the validate flavour

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/b613a961be70e75747816be4d7642b6378a84c3f">b613a961</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-14T10:17:00-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Always use mingw64 python on Windows

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/1bc8c8cd02726c5151afdb02a019aec724be8bec">1bc8c8cd</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-14T10:17:00-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Allow Windows build to fail due to #17777

The fact that `exec` isn't POSIX compliant means that things can break

in arbitrarily bad ways. Sometimes things happen to work correctly but

sadly this isn't always the case.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/ac63020d3c9960d12efef54003dd088f8ec0413b">ac63020d</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-14T10:17:00-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Drop unnecessary GHC_VERSION check

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/6926f369b8740c1c8095eb0786010d9da9c5fcfd">6926f369</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-14T10:17:00-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Bump process submodule

Folds in the second part of Phyx's Windows process exit fixes [1],

hopefully finally resolving issue #17480.

[1] https://github.com/haskell/process/pull/160

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/584eee71f803ebf59f12758755151110a6e8636f">584eee71</a></strong>

<div>

<span>by Tamar Christina</span>

<i>at 2020-02-14T10:17:00-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">SysTools: Use "process job" when spawning processes on Windows

GHC should make calls using process jobs when calling out to GCC and LD.

The reason is these use the exec () family of posix functions.  Window's

process model doesn't allow replacement of processes so this is emulated

by creating a new process and immediately exiting the old one.  Because

of this when using normal Windows wait functions you would return even

without the child process having finished.  In this case if you are

depending on data from the child you will enter a race condition.

The usual fix for this is to use process jobs and wait for the

termination of all children that have ever been spawn by the process you

called. But also waiting for the freeing of all resources.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/ecabfa28518947d752403bc1b59ee716652687da">ecabfa28</a></strong>

<div>

<span>by Tamar Christina</span>

<i>at 2020-02-14T10:17:00-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Revert "compiler: Disable atomic renaming on Windows"

The original reason this was disabled should be fixed by the previous

commit.

This reverts commit 1c1b63d63efe8b0f789aa7d5b87cfac3edd213eb.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/06d60c66d6355e895c7e46275f295b721d63fa92">06d60c66</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-14T10:17:00-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Bump Cabal submodule

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/8cabb384775adacf8326f7dd2cf99291da8dddbb">8cabb384</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-14T10:17:00-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">compare-flags: Fix output

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/8cf646d36b02b8ea1c289cb52781c9171853b514">8cf646d3</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-14T10:17:00-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">users-guide: Document -ddump-srts

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/932307a5486e63335f9ee6dc183ff39f6dfc5857">932307a5</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-14T10:17:00-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">users-guide: Fix broken reference

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/e77818de9cc9b2bc4b19dba4bc479f14078760b7">e77818de</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-15T09:26:55-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Accept performance changes

These manifested in the integer-simple job.

Metric Decrease:

    T12227

    T5549

    T14936

    T4830

    Conversions

    T5237

    T8766

    T4801

    T10359

Metric Increase:

    T12234

    T6048

    T3294

    T14683

    T3064

    T9872b

    T9872c

    T783

    T5837

    T10678

    T14697

    T5631

    T9203

    T13719

    T12707

    T13056

    T9630

    T10547

    T9872d

    T1969

    WWRec

    T10370

    T5321FD

    haddock.Cabal

    T5642

    T9872a

    T15263

    T12425

    MultiLayerModules

    T5205

    T9233

    T13379

    haddock.base

    T9020

    T13035

    T12150

    T9961

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/785008c1235bd77ddb4d13f57f92b249752d8ea5">785008c1</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-15T09:30:13-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Sort test names in expected change output

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/9e851472e59950c9d7ada3224a5bf317c0a9dc1a">9e851472</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2020-02-16T10:38:41+03:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Revert "users-guide: Document -ddump-srts"

This reverts commit 8cf646d36b02b8ea1c289cb52781c9171853b514.

The flag was removed by 16d643cf.

[ci skip]

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/9792c81663d8bef3abfe22cf45405d8d282afe08">9792c816</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-16T09:47:08-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Probe whether symlinks are usable on Windows

Closes #17706.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/ee1e5342f612c8b06ac910cd698558ade7a1a887">ee1e5342</a></strong>

<div>

<span>by Vladislav Zavialov</span>

<i>at 2020-02-16T09:47:44-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix the "unused terminals: 2" warning in Parser.y

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/b4a8ce525fce56eca18358820c0ec35fec8982de">b4a8ce52</a></strong>

<div>

<span>by Roland Senn</span>

<i>at 2020-02-18T20:14:42-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">If a :reload finds syntax errors in the module graph, remove the loaded modules. (Fixes #17549)

The processing in `compiler/main/GhcMake.hs` computes the ModuleGraph. If it finds errors

in the module header or in the import specifications, then the new module graph is

incomplete and should not be used.

The code before #17549 just reported the errors and left the old ModuleGraph in place.

The new code of this MR replaces the old ModuleGraph with an empty one.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/d7029cc09edc052c2f97effe33233c53340fcce0">d7029cc0</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-02-18T20:15:30-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Hadrian: refactor GMP in-tree build support (#17756)

* Hadrian doesn't use integer-gmp/config.mk file anymore to determine if

  building GMP in-tree is required.

  "config.mk" is created by Cabal when the integer-gmp package is

  configured and this file is still untracked by Hadrian. This led to a

  tricky configure "race" because "config.mk" is built by the

  "setup-config" rule, but this rule is also used to find dependencies,

  in particular the "ghc-gmp.h" header, but the creation of this file

  was depending (without being tracked) on "config.mk".

  Now Hadrian only builds in-tree GMP if `--with-intree-gmp` is passed

  to the top-level configure script.

* in-tree GMP isn't built once for all in a fixed stage (Stage1)

  anymore. It is built per stage which is required if we build a

  cross-compiler

* switching between in-tree and external GMP is now supported without

  having to clean the build directory first.

* "wrappers.c" now includes "ghc-gmp.h" instead of "ghc.h". It

  helps ensuring that the build system generates "ghc-gmp.h".

* build in-tree GMP in "<root>/stageN/gmp/gmpbuild" and produce useful

  artefacts (libgmp.a, gmp.h, objs/*.o) in "<root>/stageN/gmp"

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/40d917fb3b99bdb63e7e39d517225d441aaf722d">40d917fb</a></strong>

<div>

<span>by Vladislav Zavialov</span>

<i>at 2020-02-18T20:16:07-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Remove the MonadFail P instance

There were two issues with this instance:

* its existence meant that a pattern match failure in the P monad would

  produce a user-visible parse error, but the error message would not be

  helpful to the user

* due to the MFP migration strategy, we had to use CPP in Lexer.x,

  and that created issues for #17750

Updates haddock submodule.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/5a1ce45d6cf03802caf15548210bb9924468c437">5a1ce45d</a></strong>

<div>

<span>by Joshua Price</span>

<i>at 2020-02-18T20:16:47-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix unboxed tuple size limit (#17837)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/192caf58ca1fc42806166872260d30bdb34dbace">192caf58</a></strong>

<div>

<span>by Vladislav Zavialov</span>

<i>at 2020-02-18T20:17:24-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix testsuite driver output (#17847)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/1500f0898e85316c7c97a2f759d83278a072ab0e">1500f089</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-02-18T20:18:12-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Modules: Llvm (#13009)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/d53e81c02c64e72ccb55bf73f0881c5c80fa50ef">d53e81c0</a></strong>

<div>

<span>by Niklas Hambüchen</span>

<i>at 2020-02-20T10:36:22-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">8.10 Release notes for atomic .o writes [skip ci]

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/19680ee533bb95c0c5c42aca5c81197e4b233979">19680ee5</a></strong>

<div>

<span>by Niklas Hambüchen</span>

<i>at 2020-02-20T10:37:53-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">8.10 Release notes for --disable-delayed-os-memory-return [skip ci]

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/74ad75e87317196c600dfabc61aee1b87d95c214">74ad75e8</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2020-02-20T21:17:57-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Re-implement unsafe coercions in terms of unsafe equality proofs

(Commit message written by Omer, most of the code is written by Simon

and Richard)

See Note [Implementing unsafeCoerce] for how unsafe equality proofs and

the new unsafeCoerce# are implemented.

New notes added:

- [Checking for levity polymorphism] in CoreLint.hs

- [Implementing unsafeCoerce] in base/Unsafe/Coerce.hs

- [Patching magic definitions] in Desugar.hs

- [Wiring in unsafeCoerce#] in Desugar.hs

Only breaking change in this patch is unsafeCoerce# is not exported from

GHC.Exts, instead of GHC.Prim.

Fixes #17443

Fixes #16893

NoFib

-----

--------------------------------------------------------------------------------

        Program           Size    Allocs    Instrs     Reads    Writes

--------------------------------------------------------------------------------

             CS          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

            CSD          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

             FS          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

              S          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

             VS          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

            VSD          -0.1%      0.0%     -0.0%     -0.0%     -0.1%

            VSM          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

           anna          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

           ansi          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

           atom          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

         awards          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

         banner          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

     bernouilli          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

   binary-trees          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

          boyer          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

         boyer2          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

           bspt          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

      cacheprof          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

       calendar          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

       cichelli          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

        circsim          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

       clausify          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

  comp_lab_zift          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

       compress          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

      compress2          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

    constraints          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

   cryptarithm1          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

   cryptarithm2          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

            cse          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

   digits-of-e1          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

   digits-of-e2          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

         dom-lt          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

          eliza          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

          event          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

    exact-reals          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

         exp3_8          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

         expert          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

 fannkuch-redux          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

          fasta          -0.1%      0.0%     -0.5%     -0.3%     -0.4%

            fem          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

            fft          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

           fft2          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

       fibheaps          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

           fish          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

          fluid          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

         fulsom          -0.1%      0.0%     +0.0%     +0.0%     +0.0%

         gamteb          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

            gcd          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

    gen_regexps          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

         genfft          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

             gg          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

           grep          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

         hidden          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

            hpg          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

            ida          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

          infer          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

        integer          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

      integrate          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

   k-nucleotide          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

          kahan          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

        knights          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

         lambda          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

     last-piece          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

           lcss          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

           life          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

           lift          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

         linear          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

      listcompr          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

       listcopy          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

       maillist          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

         mandel          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

        mandel2          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

           mate          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

        minimax          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

        mkhprog          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

     multiplier          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

         n-body          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

       nucleic2          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

           para          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

      paraffins          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

         parser          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

        parstof          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

            pic          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

       pidigits          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

          power          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

         pretty          -0.1%      0.0%     -0.1%     -0.1%     -0.1%

         primes          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

      primetest          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

         prolog          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

         puzzle          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

         queens          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

        reptile          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

reverse-complem          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

        rewrite          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

           rfib          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

            rsa          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

            scc          -0.1%      0.0%     -0.1%     -0.1%     -0.1%

          sched          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

            scs          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

         simple          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

          solid          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

        sorting          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

  spectral-norm          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

         sphere          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

         symalg          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

            tak          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

      transform          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

       treejoin          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

      typecheck          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

        veritas          -0.0%      0.0%     -0.0%     -0.0%     -0.0%

           wang          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

      wave4main          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

   wheel-sieve1          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

   wheel-sieve2          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

           x2n1          -0.1%      0.0%     -0.0%     -0.0%     -0.0%

--------------------------------------------------------------------------------

            Min          -0.1%      0.0%     -0.5%     -0.3%     -0.4%

            Max          -0.0%      0.0%     +0.0%     +0.0%     +0.0%

 Geometric Mean          -0.1%     -0.0%     -0.0%     -0.0%     -0.0%

Test changes

------------

- break006 is marked as broken, see #17833

- The compiler allocates less when building T14683 (an unsafeCoerce#-

  heavy happy-generated code) on 64-platforms. Allocates more on 32-bit

  platforms.

- Rest of the increases are tiny amounts (still enough to pass the

  threshold) in micro-benchmarks. I briefly looked at each one in a

  profiling build: most of the increased allocations seem to be because

  of random changes in the generated code.

Metric Decrease:

    T14683

Metric Increase:

    T12150

    T12234

    T12425

    T13035

    T14683

    T5837

    T6048

Co-Authored-By: Richard Eisenberg <rae@cs.brynmawr.edu>

Co-Authored-By: Ömer Sinan Ağacan <omeragacan@gmail.com>

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/6880d6aa1e6e96579bbff89712efd813489cc828">6880d6aa</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-02-20T21:18:48-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Disentangle DynFlags and SDoc

Remove several uses of `sdocWithDynFlags`. The remaining ones are mostly

CodeGen related (e.g. depend on target platform constants) and will be

fixed separately.

Metric Decrease:

   T12425

   T9961

   WWRec

   T1969

   T14683

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/70a901105044c85c756fd93899387f0215f7b1de">70a90110</a></strong>

<div>

<span>by Julien Debon</span>

<i>at 2020-02-20T21:19:27-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">doc(List): Add examples to GHC.List

* Add examples

* Cleanup documentation

* Clarify merge process and Marge bot

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/c8439fc789ff00cfdd453d2425d2146df69e0729">c8439fc7</a></strong>

<div>

<span>by Peter Trommler</span>

<i>at 2020-02-20T21:20:05-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix testsuite on powerpc64le

Remove expect broken on recomp tests, #11260 was closed

by !2264 and #11323 most likely by !2264 as well.

GHCi scripts tests work on GHCi but not the external interpreter,

adjust test configuration accordingly. Fixes unexpected passes.

Mark test requiring DWARF expect fail on powerpc64[le] for #11261.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/65b7256a88ae2bd878da5d026e4183cba6f6eedf">65b7256a</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2020-02-20T21:20:45-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Use concatMap(M) instead of `concat . map` and the monadic variant

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/8b76d4574d703ee66f346775d408220fdddb8155">8b76d457</a></strong>

<div>

<span>by Roland Senn</span>

<i>at 2020-02-20T21:21:28-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix #17832: Weird handling of exports named main in 8.10-rc1

Switching from `lookupGlobalOccRn_maybe` to `lookupInfoOccRn`

to check whether a `main` function is in scope. Unfortunately

`lookupGlobalOccRn_maybe` complains if there are multiple `main`

functions in scope.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/466e1ad5d54cb2e8a3b6f16904b873cad882a736">466e1ad5</a></strong>

<div>

<span>by Krzysztof Gogolewski</span>

<i>at 2020-02-20T21:22:11-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Use TTG for HsSplicedT constructor

The constructor HsSplicedT occurs only in the GhcTc pass.

This enforces this fact statically via TTG.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/4e622fcabb30ab7f3836bfc5739747b5608bef43">4e622fca</a></strong>

<div>

<span>by Alexis King</span>

<i>at 2020-02-20T21:22:49-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Normalize types when dropping absent arguments from workers

fixes #17852

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/a533e54722ec2ec22ef598e07dcfe6c9e84d2602">a533e547</a></strong>

<div>

<span>by Adam Sandberg Eriksson</span>

<i>at 2020-02-20T21:23:31-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Mention users guide and release notes in merge request template</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/05251b175b3745ed96c9f672115c78c527b63b9e">05251b17</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-20T21:24:08-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Fix typo in BIN_DIST_PREP_TAR_COMP variable name</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/f44c7e6723498c5dde0cd78e4af26142a14d98f4">f44c7e67</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-20T21:24:46-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Avoid duplicating ~/.cabal contents with every build

Previously our attempt to cache the cabal store would

`cp cabal-cache ~/.cabal`. However, if the latter already

 existed this meant that we would end up with

~/.cabal/cabal-cache. Not only would this not help caching

but it would exponentially grow the size of ~/.cabal. Not

good!</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/c5ec996583373025488c090fb2c89f7bda38c1cb">c5ec9965</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-20T21:56:13-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">GHC.Hs.Extension: Use Type instead of *

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/89cb4cc4cc76f834b0bcc53fb551db706ef143b7">89cb4cc4</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-20T21:56:13-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Use Type instead of * in GHC

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/04eb0d6c4de23a4cfe3953e7496f5bc4b5b1d53c">04eb0d6c</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-20T21:56:13-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Enable -Wstar-is-type in -Wall

As noted in [proposal 0143][proposal] this is supposed to happen in

8.12.

Also fix an incorrect claim in the users guide that -Wstar-is-type is

enabled by default.

[proposal]: https://github.com/ghc-proposals/ghc-proposals/blob/master/proposals/0143-remove-star-kind.rst

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/6de966f1369740d00193e64ec82d11e934256616">6de966f1</a></strong>

<div>

<span>by Andreas Klebinger</span>

<i>at 2020-02-20T21:56:15-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix #17724 by having occAnal preserve used bindings.

It sometimes happened that occAnal would remove bindings

as dead code by relying on bindings to be in dependency

order. The fix was contributed by SPJ.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/abd7f962a9a9207c07cb864921941750818f61a4">abd7f962</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-20T21:56:15-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">users-guide: Mention dependency on `exceptions` in release notes

Fixes #17845.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/581753790da8d094ce05b3fc5f5360f5fbf6a7da">58175379</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-02-20T21:56:20-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Hadrian: minor GMP refactoring

Somehow I forgot to totally remove `gmpContext` in

d7029cc09edc052c2f97effe33233c53340fcce0. This patch fixes it and adds

some additional comments.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/33fa8d9433d82b03f3209698bf27420a32302321">33fa8d94</a></strong>

<div>

<span>by Ryan Scott</span>

<i>at 2020-02-20T21:56:21-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Generalize liftData to work over any Quote (#17857)

The Overloaded Quotations proposal generalized the type of `lift`

to work over any `Quote`, but not the type of `liftData`, leading

to #17857. Thankfully, generalizing `liftData` is extremely

straightforward.

Fixes #17857.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/3cea67955a5ae7d3de7aa9dcfdabaf6ee3681ffc">3cea6795</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-02-20T21:56:23-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Make: fix sdist target (#17848)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/e2cce997320652031e0c6cc62db5e35deff43a79">e2cce997</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-02-20T21:56:23-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Hadrian: fix source-dist target (#17849)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/0a4c89b208c59ddf79c55ee446fcad5a012bb1bc">0a4c89b2</a></strong>

<div>

<span>by Matthew Pickering</span>

<i>at 2020-02-21T20:44:45-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Special case `mkTyConApp liftedTypeKind []`

We really need to make sure that these are shared because otherwise GHC

will allocate thousands of identical `TyConApp` nodes.

See #17292

-------------------------

Metric Decrease:

    haddock.Cabal

    T14683

-------------------------

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/0482f58ab0490b2394ad60946dde3214a0ca1810">0482f58a</a></strong>

<div>

<span>by Matthew Pickering</span>

<i>at 2020-02-21T20:45:21-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">TH: wrapGenSyns, don't split the element type too much

The invariant which allowed the pervious method of splitting the type of

the body to find the type of the elements didn't work in the new

overloaded quotation world as the type can be something like

`WriterT () m a` rather than `Q a` like before.

Fixes #17839

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/be7068a6130f394dcefbcb5d09c2944deca2270d">be7068a6</a></strong>

<div>

<span>by Vladislav Zavialov</span>

<i>at 2020-02-21T20:45:59-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Parser API annotations: RealSrcLoc

During parsing, GHC collects lexical information about AST nodes and

stores it in a map. It is needed to faithfully restore original source

code, e.g. compare these expressions:

        a =  b

        a  = b

The position of the equality sign is not recorded in the AST, so it must

be stored elsewhere.

This system is described in Note [Api annotations].

Before this patch, the mapping was represented by:

        Map (SrcSpan, AnnKeywordId) SrcSpan

After this patch, the mapping is represented by:

        Map (RealSrcSpan, AnnKeywordId) RealSrcSpan

The motivation behind this change is to avoid using the Ord SrcSpan

instance (required by Map here), as it interferes with #17632 (see the

discussion there).

SrcSpan is isomorphic to  Either String RealSrcSpan,  but we shouldn't

use those strings as Map keys. Those strings are intended as hints to

the user, e.g. "<interactive>" or "<compiler-generated code>", so they

are not a valid way to identify nodes in the source code.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/240f5bf6f53515535be5bf3ef7632aa69ae21e3e">240f5bf6</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-02-21T20:46:40-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Modules: Driver (#13009)

submodule updates: nofib, haddock

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/9d09411122b9b534b96e988b6d3f6d7eb04b8f66">9d094111</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-02-21T20:47:19-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Hadrian: `docs` rule needs `configure` (#17840)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/1674353a1ab40fde183f22bef9eb019d8150dc5a">1674353a</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-23T17:31:19-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">fs: Port fixes from ghc-jailbreak repository

* Override rename, unlink, and remove

* Factor out wchar conversion

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/853210f20699b54dee4f85a15b70b3f02a1b68b7">853210f2</a></strong>

<div>

<span>by Adam Sandberg Ericsson</span>

<i>at 2020-02-23T17:32:03-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">show gcc linker options in configure summary

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/2831544aaab15ab17d66e7af42f890d92d47407a">2831544a</a></strong>

<div>

<span>by Adam Sandberg Ericsson</span>

<i>at 2020-02-23T17:32:44-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">hadrian: docs depend on stage1 ghc

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/1d9df9e00b021724d570a12d8c6d7870bdc054ca">1d9df9e0</a></strong>

<div>

<span>by Adam Sandberg Ericsson</span>

<i>at 2020-02-23T17:33:23-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">ci: after 5ce63d52fed the linux bindist for doc-tarball has changed name

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/26e8fff33b1b46ba559538b4519730c60c0d45e5">26e8fff3</a></strong>

<div>

<span>by Vladislav Zavialov</span>

<i>at 2020-02-24T02:05:30-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Remove Ord SrcLoc, Ord SrcSpan

Before this patch, GHC relied on Ord SrcSpan to identify source elements, by

using SrcSpan as Map keys:

        blackList :: Map SrcSpan ()      -- compiler/GHC/HsToCore/Coverage.hs

        instanceMap :: Map SrcSpan Name  -- compiler/GHC/HsToCore/Docs.hs

Firstly, this design is not valid in presence of UnhelpfulSpan, as it

distinguishes between  UnhelpfulSpan "X"  and  UnhelpfulSpan "Y", but those

strings are messages for the user, unfit to serve as identifiers for source

elements.

Secondly, this design made it hard to extend SrcSpan with additional data.

Recall that the definition of SrcSpan is:

        data SrcSpan =

            RealSrcSpan !RealSrcSpan

          | UnhelpfulSpan !FastString

Say we want to extend the RealSrcSpan constructor with additional information:

        data SrcSpan =

            RealSrcSpan !RealSrcSpan !AdditionalInformation

          | UnhelpfulSpan !FastString

        getAdditionalInformation :: SrcSpan -> AdditionalInformation

        getAdditionalInformation (RealSrcSpan _ a) = a

Now, in order for  Map SrcSpan  to keep working correctly, we must *ignore* additional

information when comparing SrcSpan values:

        instance Ord SrcSpan where

          compare (RealSrcSpan r1 _) (RealSrcSpan r2 _) = compare r1 r2

          ...

However, this would violate an important law:

        a == b  therefore  f a == f b

Ignoring  AdditionalInformation  in comparisons would mean that with

f=getAdditionalInformation, the law above does not hold.

A more robust design is to avoid  Ord SrcSpan  altogether, which is what this patch implements.

The mappings are changed to use RealSrcSpan instead:

        blackList :: Set RealSrcSpan         -- compiler/GHC/HsToCore/Coverage.hs

        instanceMap :: Map RealSrcSpan Name  -- compiler/GHC/HsToCore/Docs.hs

All SrcSpan comparisons are now done with explicit comparison strategies:

        SrcLoc.leftmost_smallest

        SrcLoc.leftmost_largest

        SrcLoc.rightmost_smallest

These strategies are not subject to the law mentioned above and can easily

discard both the string stored in  UnhelpfulSpan  and  AdditionalInformation.

Updates haddock submodule.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/5aa6c18866cf4071246891da3b0f7651b5d28d0f">5aa6c188</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-24T02:06:09-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">users-guide: Shuffle text

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/e3f1741356491332613aacf66c96991682b84603">e3f17413</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-24T02:06:09-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">users-guide: Drop old release notes

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/84dd96105faff71cb17d1c1d069d724d014847dc">84dd9610</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-24T02:06:09-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Bump directory submodule to 1.3.6.0

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/e295a02440db8b1d96cebad22c5ee48774774681">e295a024</a></strong>

<div>

<span>by Stefan Pavikevik</span>

<i>at 2020-02-24T20:53:44-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">check for safe arguments, raising error when invalid (fix #17720)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/354e2787be08fb6d973de1a39e58080ff8e107f8">354e2787</a></strong>

<div>

<span>by Krzysztof Gogolewski</span>

<i>at 2020-02-24T20:54:35-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Comments, small refactor

* Remove outdated Note [HsForAllTy tyvar binders] and [Context quantification].

  Since the wildcard refactor 1e041b7382, HsForAllTy no longer has an flag

  controlling explicity. The field `hsq_implicit` is gone too.

  The current situation is covered by Note [HsType binders] which is already

  linked from LHsQTyVars.

* Small refactor in CoreLint, extracting common code to a function

* Remove "not so sure about WpFun" in TcEvidence, per Richard's comment

  https://gitlab.haskell.org/ghc/ghc/merge_requests/852#note_223226

* Use mkIfThenElse in Foreign/Call, as it does exactly what we need.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/1b1067d14b656bbbfa7c47f156ec2700c9751549">1b1067d1</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-02-24T20:55:25-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Modules: CmmToAsm (#13009)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/621468f680a111c20366a9445a3b225cbc34ffbf">621468f6</a></strong>

<div>

<span>by Alexis King</span>

<i>at 2020-02-26T15:08:09-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Treat coercions as arguments for floating and inlining

This reverts commit 8924224ecfa065ebc67b96a90d01cf9d2edd0e77

and fixes #17787.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/def486c90ef6f37d81d0d9c6df754431ff94c6a4">def486c9</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-26T15:08:47-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">hadrian: Allow libnuma library path to be specified

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/ed03d4e76ead6384e38bb742fee73905d92d1545">ed03d4e7</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-26T15:08:47-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">hadrian: Refactor gmp arguments

Move the gmp configuration to its own binding.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/09b8838422d5dde32af7b09c8adcb4dff4cd00fb">09b88384</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-26T15:08:47-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">hadrian: Tell Cabal about integer-gmp library location

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/161e08c5d85d7ee7f731db31481cd818480f07a7">161e08c5</a></strong>

<div>

<span>by Krzysztof Gogolewski</span>

<i>at 2020-02-26T15:09:30-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Remove dead code

* FailablePattern can no longer be created since ab51bee40c82

  Therefore, Opt_WarnMissingMonadFailInstances has no effect anymore.

* XWrap is no longer used, it was moved to an extension field

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/e0d09db377bcb3d4caef2abe9db0cac36811d740">e0d09db3</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-26T15:10:09-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Use 8.8.3 to bootstrap on Windows

This should fix #17861.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/972bcf3a1971691ec809452322404ce8f65ad721">972bcf3a</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-26T15:10:09-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Fix symlink test

Needs to `write` bytes, not str.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/273e60dee183707315415799ae3ed08079ab9ee9">273e60de</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-26T15:10:09-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Add shell subcommand for debugging within CI environment

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/43b13ed33f50274383a8dce11cbd7c77b8e851d5">43b13ed3</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-26T15:10:09-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Fix colors on Darwin

Darwin sh doesn't support \e.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/217546a798e1eb5a931a677771bbd54bb0e4b5ee">217546a7</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-26T15:10:09-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Flush stdout buffers in InitEventLogging

Otherwise we are sensitive to libc's buffering strategy.

Similar to the issue fixed in 543dfaab166c81f46ac4af76918ce32190aaab22.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/c7d4fa55d19f7ac72f3071f9d19e03443c9b538c">c7d4fa55</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-26T15:10:09-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Add run_hadrian subcommand

I've ruined two trees already by failing to pass --flavour to hadrian.

Let's factor this out so it can be reused during troubleshooting.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/7dc54873c0768ad3234c40300ae20e4f1e847bdd">7dc54873</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-26T15:10:09-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Allow tests to be marked as broken on the command line

This allows us to work-around distribution-specific breakage easily.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/25e2458ee9e8007687c165fe6b27adf91f405ed0">25e2458e</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-26T15:10:09-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">hadrian: Add --broken-test flag

This exposes the flag of the same name supported by the testsuite

driver.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/557699967f9438d3bf2936cbc6edce55b11622e0">55769996</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-26T15:10:09-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Mark some tests as broken on Alpine

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/9ee7f87dbea786259d12fb9177fd2759d4fad3bb">9ee7f87d</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-26T15:10:09-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">SysTools: Don't use process jobs if they are broken

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/bfaa39618a25ae642221a478889b9667281b52e8">bfaa3961</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-26T15:10:09-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Bump hsc2hs submodule

Fixes name of C compiler.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/b2b49a0aad353201678970c76d8305a5dcb1bfab">b2b49a0a</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-26T15:10:09-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Make hasMetricsFile RHS more descriptive

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/817f93eac4d13f680e8e3e7a25eb403b1864f82e">817f93ea</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-02-26T15:10:58-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Modules: Core (#13009)

Update haddock submodule

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/74311e10bbb6ced9cd3093c8949f2535a715d8f6">74311e10</a></strong>

<div>

<span>by Sebastian Graf</span>

<i>at 2020-02-27T16:22:45-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">PmCheck: Implement Long-distance information with Covered sets

Consider

```hs

data T = A | B | C

f :: T -> Int

f A = 1

f x = case x of

  A -> 2

  B -> 3

  C -> 4

```

Clearly, the RHS returning 2 is redundant. But we don't currently see

that, because our approximation to the covered set of the inner case

expression just picks up the positive information from surrounding

pattern matches. It lacks the context sensivity that `x` can't be `A`

anymore!

Therefore, we adopt the conceptually and practically superior approach

of reusing the covered set of a particular GRHS from an outer pattern

match. In this case, we begin checking the `case` expression with the

covered set of `f`s second clause, which encodes the information that

`x` can't be `A` anymore. After this MR, we will successfully warn about

the RHS returning 2 being redundant.

Perhaps surprisingly, this was a great simplification to the code of

both the coverage checker and the desugarer.

Found a redundant case alternative in `unix` submodule, so we have to

bump it with a fix.

Metric Decrease:

    T12227

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/59c023ba5ccb10fff62810591f20608bd73c97af">59c023ba</a></strong>

<div>

<span>by Adam Sandberg Ericsson</span>

<i>at 2020-02-27T16:23:25-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">configure: correctly generate LIBRARY_template_haskell_VERSION

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/9be82389a702c09d03975e7a0c17e5866dc17ffc">9be82389</a></strong>

<div>

<span>by Krzysztof Gogolewski</span>

<i>at 2020-02-28T02:35:35-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">boot: Remove remote origin check

Previously, we used relative paths in submodules. When cloning from

GitHub, they had to be manually tweaked.

Since a76b233d we use absolute paths, so this workaround can be removed.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/f4b6b594fae6b257e997ade3fae67b3378e54a22">f4b6b594</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-28T02:36:12-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">nonmoving: Fix marking in compact regions

Previously we were tracing the object we were asked to mark, even if it

lives in a compact region. However, there is no need to do this; we need

only to mark the region itself as live.

I have seen a segfault due to this due to the concurrent mark seeing a

an object in the process of being compacted by the mutator.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/f97d1fb614ff7638d13eb7c552d2a72fce4c613a">f97d1fb6</a></strong>

<div>

<span>by Alp Mestanogullari</span>

<i>at 2020-02-28T02:36:59-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">base: use an explicit import list in System.Environment.ExecutablePath

This was making -Werror builds fail on Windows (at least with Hadrian).

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/66f5d6d634698ce054a9b9ce0c53ce9682f6f042">66f5d6d6</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2020-02-28T22:03:23-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Improve error handling for VTA + deferred type errors

This fixes #17792

See Note [VTA for out-of-scope functions] in TcExpr

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/37f126033f1e5bf0331143f005ef90ba6e2e02cd">37f12603</a></strong>

<div>

<span>by Ilias Tsitsimpis</span>

<i>at 2020-02-28T22:04:04-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">llvm-targets: Add arm-unknown-linux-gnueabi

Add arm-unknown-linux-gnueabi, which is used by Debian's ARM EABI port

(armel), as an LLVM target.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/327b29e1a05d9f1ea04465c9b23aed92473dd453">327b29e1</a></strong>

<div>

<span>by Vladislav Zavialov</span>

<i>at 2020-02-29T05:06:31-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Monotonic locations (#17632)

When GHC is parsing a file generated by a tool, e.g. by the C preprocessor, the

tool may insert #line pragmas to adjust the locations reported to the user.

As the result, the locations recorded in RealSrcLoc are not monotonic. Elements

that appear later in the StringBuffer are not guaranteed to have a higher

line/column number.

In fact, there are no guarantees whatsoever, as #line pragmas can arbitrarily

modify locations. This lack of guarantees makes ideas such as #17544

infeasible.

This patch adds an additional bit of information to every SrcLoc:

        newtype BufPos = BufPos { bufPos :: Int }

A BufPos represents the location in the StringBuffer, unaffected by any

pragmas.

Updates haddock submodule.

Metric Increase:

    haddock.Cabal

    haddock.base

    haddock.compiler

    MultiLayerModules

    Naperian

    parsing001

    T12150

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/99d2de860ff82e3f626c8b3d47d355ab16d18fd1">99d2de86</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-02-29T05:07:10-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">plugins: Ensure that loadInterface plugins can see annotations

loadInterface replaces the `mi_decls`, `mi_insts`, `mi_fam_insts`,

`mi_rules`, `mi_anns` fields of ModIface with `undefined` before

inserting the interface into the EPS. However, we still want to give

loadInterface plugins access to these fields. Consequently, we want to

pass the unmodified `ModIface` the plugin.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/a999ee96a22650b59855609f84e2808b4133e9e0">a999ee96</a></strong>

<div>

<span>by Xavier Denis</span>

<i>at 2020-02-29T05:07:50-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Rename ghci.sh and build.sh to ghci and build respectively

Convert hadrian buildscripts to unsuffixed, dashed form

final cleanups

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/b5fb58fd1a4a24b9273d9d2de65b6347e1654e98">b5fb58fd</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2020-02-29T05:08:36-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Document and refactor a few things around bitmap scavenging

- Added a few comments in StgPAP

- Added a few comments and assertions in scavenge_small_bitmap and

  walk_large_bitmap

- Did tiny refactor in GHC.Data.Bitmap: added some comments, deleted

  dead code, used PlatformWordSize type.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/18757cab04c5c5c48eaceea19469d4811c5d0371">18757cab</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-02-29T05:09:25-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Refactor runtime interpreter code

In #14335 we want to be able to use both the internal interpreter (for

the plugins) and the external interpreter (for TH and GHCi) at the same

time.

This patch performs some preliminary refactoring: the `hsc_interp` field

of HscEnv replaces `hsc_iserv` and is now used to indicate which

interpreter (internal, external) to use to execute TH and GHCi.

Opt_ExternalInterpreter flag and iserv options in DynFlags are now

queried only when we set the session DynFlags. It should help making GHC

multi-target in the future by selecting an interpreter according to the

selected target.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/b86a639504e1de79c861a2c11b897089cc63ae5f">b86a6395</a></strong>

<div>

<span>by Adam Sandberg Ericsson</span>

<i>at 2020-02-29T05:10:06-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">docs: correct relative links to haddocks from users guide (fixes #17866)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/0f55df7f197b7e30d7c6e39d80deb36b7dc86fb5">0f55df7f</a></strong>

<div>

<span>by Adam Sandberg Ericsson</span>

<i>at 2020-02-29T05:10:06-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">docs: correct link to th haddocks from users guide

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/252e51179bd179f52e25dea762ef7b95bc56ce0a">252e5117</a></strong>

<div>

<span>by Jean-Baptiste Mazon</span>

<i>at 2020-02-29T05:10:46-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">rts: enforce POSIX numeric locale for heap profiles

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/34c7d23074f47c720b5722ca14d78a34213eabb6">34c7d230</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-02-29T05:11:27-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix Hadrian's ``--configure`` (fix #17883)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/04d30137771a6cf8a18fda1ced25f78d0b2eb204">04d30137</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2020-02-29T05:12:06-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Simplify IfaceIdInfo type

IfaceIdInfo type is confusing: there's practically no difference between

`NoInfo` and `HasInfo []`. The comments say NoInfo is used when

-fomit-interface-pragmas is enabled, but we don't need to distinguish

`NoInfo` from `HasInfo []` in when reading the interface so the

distinction is not important.

This patch simplifies the type by removing NoInfo. When we have no info

we use an empty list.

With this change we no longer read the info list lazily when reading an

IfaceInfoItem, but when reading an IfaceId the ifIdInfo field is

read lazily, so I doubt this is going to be a problem.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/3979485bd97771373214c44d14b7830ba447ad23">3979485b</a></strong>

<div>

<span>by Roland Senn</span>

<i>at 2020-02-29T17:36:59+01:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Show breakpoint locations of breakpoints which were ignored during :force (#2950)

GHCi is split up into 2 major parts: The user-interface (UI)

and the byte-code interpreter. With `-fexternal-interpreter`

they even run in different processes. Communication between

the UI and the Interpreter (called `iserv`) is done using

messages over a pipe. This is called `Remote GHCI` and

explained in the Note [Remote GHCi] in `compiler/ghci/GHCi.hs`.

To process a `:force` command the UI sends a `Seq` message

to the `iserv` process. Then `iserv` does the effective

evaluation of the value. When during this process a breakpoint

is hit, the `iserv` process has no additional information to

enhance the `Ignoring breakpoint` output with the breakpoint

location.

To be able to print additional breakpoint information,

there are 2 possible implementation choices:

1. Store the needed information in the `iserv` process.

2. Print the `Ignoring breakpoint` from the UI process.

For option 1 we need to store the breakpoint info redundantely

in 2 places and this is bad. Therfore option 2 was implemented

in this MR:

- The user enters a `force` command

- The UI sends  a `Seq` message to the `iserv` process.

- If processing of the `Seq` message hits a breakpoint,

  the `iserv` process returns control to the UI process.

- The UI looks up the source location of the breakpoint,

  and prints the enhanced `Ignoring breakpoint` output.

- The UI sends a `ResumeSeq` message to the `iserv` process,

  to continue forcing.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/3cf7303bf39bca734bd87f0f42872f1a4025fa34">3cf7303b</a></strong>

<div>

<span>by Krzysztof Gogolewski</span>

<i>at 2020-03-02T01:18:33-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Remove dead code

* The names in PrelName and THNames are no longer used

  since TH merged types and kinds, Typeable is kind-polymorphic,

  .net support was removed

* unqualQuasiQuote no longer used since 6f8ff0bbad3b9fa3

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/dbea7e9d4d4a8812f3dbeed720fb6aa97e16f896">dbea7e9d</a></strong>

<div>

<span>by Ilias Tsitsimpis</span>

<i>at 2020-03-02T01:19:12-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Do not define hs_atomic{read,write}64() on non-64bit

Do not define hs_atomicread64() and hs_atomicwrite64() on machines where

WORD_SIZE_IN_BITS is less than 64, just like we do with the rest of the atomic

functions which work on 64-bit values.

Without this, compilation fails on MIPSel and PowerPC with the following error:

/usr/bin/ld: /<<PKGBUILDDIR>>/libraries/ghc-prim/dist-install/build/libHSghc-prim-0.5.3_p.a(atomic.p_o): in function `hs_atomicread64':

atomic.c:(.text.hs_atomicread64+0x8): undefined reference to `__sync_add_and_fetch_8'

/usr/bin/ld: /<<PKGBUILDDIR>>/libraries/ghc-prim/dist-install/build/libHSghc-prim-0.5.3_p.a(atomic.p_o): in function `hs_atomicwrite64':

atomic.c:(.text.hs_atomicwrite64+0x38): undefined reference to `__sync_bool_compare_and_swap_8'

Fixes #17886.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/7c0c76fb1a11532076a0342655a9c889a2a12768">7c0c76fb</a></strong>

<div>

<span>by Roland Senn</span>

<i>at 2020-03-02T17:13:55-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Set `ImpredicativeTypes` during :print command. (#14828)

If ImpredicativeTypes is not enabled, then `:print <term>` will fail if the

type of <term> has nested `forall`s or `=>`s.

This is because the GHCi debugger's internals will attempt to unify a

metavariable with the type of <term> and then display the result, but if the

type has nested `forall`s or `=>`s, then unification will fail.

As a result, `:print` will bail out and the unhelpful result will be

`<term> = (_t1::t1)` (where `t1` is a metavariable).

Beware: <term> can have nested `forall`s even if its definition doesn't use

RankNTypes! Here is an example from #14828:

  class Functor f where

    fmap :: (a -> b) -> f a -> f b

Somewhat surprisingly, `:print fmap` considers the type of fmap to have

nested foralls. This is because the GHCi debugger sees the type

`fmap :: forall f. Functor f => forall a b. (a -> b) -> f a -> f b`.

We could envision deeply instantiating this type to get the type

`forall f a b. Functor f => (a -> b) -> f a -> f b`,

but this trick wouldn't work for higher-rank types.

Instead, we adopt a simpler fix: enable `ImpredicativeTypes` when using

`:print` and friends in the GHCi debugger. This is allows metavariables

to unify with types that have nested (or higher-rank) `forall`s/`=>`s,

which makes `:print fmap` display as

`fmap = (_t1::forall a b. Functor f => (a -> b) -> f a -> f b)`, as expected.

Although ImpredicativeTypes is a somewhat unpredictable from a type inference

perspective, there is no danger in using it in the GHCi debugger, since all

of the terms that the GHCi debugger deals with have already been typechecked.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/2a2f51d79f145e015cc089d97cf71c19dd27bee4">2a2f51d7</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-03-02T17:14:38-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Use configure script to detect that we should use in-tree GMP on Windows

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/8c663c2c2ff0d79b2c7cfadb0ce961339f1e7e4c">8c663c2c</a></strong>

<div>

<span>by Andreas Klebinger</span>

<i>at 2020-03-04T16:12:14+01:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Be explicit about how stack usage of mvar primops are covered.

This fixes #17893

[skip-ci]

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/cedd6f3041de6abe64dfa3257bec7730a9dced9f">cedd6f30</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-03-05T14:53:12-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">rts: Add getCurrentThreadCPUTime helper

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/ace618cd2294989e783bd453cee88e0e1c0dad77">ace618cd</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-03-05T14:53:12-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">nonmoving-gc: Track time usage of nonmoving marking

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/022b5ad5775ecf6652e663cddb2773d7230385f2">022b5ad5</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-03-05T14:53:12-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Stats: Add sync pauses to +RTS -S output

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/067632342cf2f063b0f23c255740e2717e5e14c7">06763234</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-03-05T14:53:12-05:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">rts: Report nonmoving collector statistics in machine-readable output

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/70d2b9956d1ecc9d40d1e2d4920983af00ea846d">70d2b995</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-03-09T06:10:52-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">nonmoving: Fix collection of sparks

Previously sparks living in the non-moving heap would be promptly GC'd

by the minor collector since pruneSparkQueue uses the BF_EVACUATED flag,

which non-moving heap blocks do not have set.

Fix this by implementing proper support in pruneSparkQueue for

determining reachability in the non-moving heap. The story is told in

Note [Spark management in the nonmoving heap].

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/9668781a36941e7552fcec38f6d4e1d5ec3ef6d1">9668781a</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-03-09T06:11:30-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Disable Sphinx documentation in Alpine build</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/8eb2c2639494bed3326163df52fa13de941cd047">8eb2c263</a></strong>

<div>

<span>by Jean-Baptiste Mazon</span>

<i>at 2020-03-09T16:33:37-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix Windows breakage by not touching locales on Windows

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/b8dab057f0730a324b7cb7b74a11152f6b96a889">b8dab057</a></strong>

<div>

<span>by Jean-Baptiste Mazon</span>

<i>at 2020-03-09T16:33:37-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">rts: ensure C numerics in heap profiles using Windows locales if needed

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/7d95260f0b9a91186eb6a99d6235151a22b2c5ae">7d95260f</a></strong>

<div>

<span>by Jean-Baptiste Mazon</span>

<i>at 2020-03-09T16:33:37-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">rts: refactor and comment profile locales

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/5b62781336aff0007aa5fde261c3028a82543a09">5b627813</a></strong>

<div>

<span>by Ryan Scott</span>

<i>at 2020-03-09T16:34:14-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Use InstanceSigs in GND/DerivingVia-generated code (#17899)

Aside from making the generated code easier to read when

`-ddump-deriv` is enabled, this makes the error message in `T15073`

substantially simpler (see the updated `T15073` expected stderr).

Fixes #17899.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/70b5077820a0859d22edd29885b62eef3c363058">70b50778</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-03-10T02:05:42-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">SysTools: Ensure that error parser can handle absolute paths on Windows

This fixes #17786, where the error parser fails to correctly handle the

drive name in absolute Windows paths.

Unfortunately I couldn't find a satisfactory way to test this.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/85b861d822251486fdf900ee951c275ebd8dfcb3">85b861d8</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-03-10T02:05:42-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Add test for #17786

This isn't pretty but it's perhaps better than nothing.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/ee2c50cbeead0d865a02a963f3f3c72e298398d7">ee2c50cb</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-03-10T02:06:33-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Hadrian: track missing configure results

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/ca8f51d475a69583a228f118e6b9dac98ba483d3">ca8f51d4</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2020-03-10T02:07:22-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add regression test for T17904

Closes #17904

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/5fa9cb82223de1c1c2684aa6917bf85a2e3c6469">5fa9cb82</a></strong>

<div>

<span>by Richard Eisenberg</span>

<i>at 2020-03-10T12:29:46-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">anyRewritableTyVar now looks in RuntimeReps

Previously, anyRewritableTyVar looked only at the arg and res

of `arg -> res`, but their RuntimeReps are also subject to

rewriting. Easy to fix.

Test case: typecheck/should_compile/T17024

Fixes #17024.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/5ba01d83a2078acce789a9d5f111efa4d4ffc616">5ba01d83</a></strong>

<div>

<span>by Ben Price</span>

<i>at 2020-03-10T12:30:27-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Clarify a Lint message

When developing a plugin I had a shadowing problem, where I generated

code

  app = \f{v r7B} x{v r7B} -> f{v r7B} x{v r7B}

This is obviously wrong, since the occurrence of `f` to the right of the

arrow refers to the `x` binder (they share a Unique). However, it is

rather confusing when Lint reports

  Mismatch in type between binder and occurrence

  Var: x{v rB7}

since it is printing the binder, rather than the occurrence.

It is rather easy to read this as claiming there is something wrong with

the `x` occurrence!

We change the report to explicitly print both the binder and the

occurrence variables.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/7b2c827b7b68f0ade7f4ae66e7033fdb84d75a5f">7b2c827b</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2020-03-10T12:31:15-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Comments only

Clarify code added in #17852 and MR !2724

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/3300eeacbbf7a3d1f961f809be5d236c48827b28">3300eeac</a></strong>

<div>

<span>by Krzysztof Gogolewski</span>

<i>at 2020-03-10T12:31:54-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Misc cleanup

- Remove Note [Existentials in shift_con_pat].

  The function shift_con_pat has been removed 15 years ago in 23f40f0e9be6d4.

- Remove kcLookupTcTyCon - it's the same as tcLookupTcTyCon

- Remove ASSERT in tyConAppArgN. It's already done by getNth,

  and it's the only reason getNth exists.

- Remove unused function nextRole

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/abf5736bcad2d740b5854e2e4a9b3547b9b06639">abf5736b</a></strong>

<div>

<span>by Krzysztof Gogolewski</span>

<i>at 2020-03-10T18:05:01+01:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Typos in comments [skip ci]

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/bb586f894532baf1bcb822afd0df7f9fea198671">bb586f89</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-03-11T00:14:59-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">rts: Prefer darwin-specific getCurrentThreadCPUTime

macOS Catalina now supports a non-POSIX-compliant version of clock_gettime

which cannot use the clock_gettime codepath.

Fixes #17906.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/20800b9a9e88a8784a3ee8720544f504aba7b4f7">20800b9a</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-03-11T08:17:19-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Split GHC.Iface.Utils module

* GHC.Iface.Recomp: recompilation avoidance stuff

* GHC.Iface.Make: mkIface*

Moved `writeIfaceFile` into GHC.Iface.Load alongside `readIface` and

renamed it `writeIface` for consistency.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/1daa20298880487b2a351c88f7ade840eb4632c6">1daa2029</a></strong>

<div>

<span>by Greg Steuck</span>

<i>at 2020-03-11T08:17:56-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fixed a minor typo in codegen.rst</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/0bc23338b02ffd624247ace409ab690b2b4a6186">0bc23338</a></strong>

<div>

<span>by Ryan Scott</span>

<i>at 2020-03-11T08:18:32-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Re-quantify when generalising over rewrite rule types

Previously, `tcRules` would check for naughty quantification

candidates (see `Note [Naughty quantification candidates]` in

`TcMType`) when generalising over the type of a rewrite rule. This

caused sensible-looking rewrite rules (like those in #17710) to be

rejected. A more permissing (and easier-to-implement) approach is to

do what is described in `Note [Generalising in tcTyFamInstEqnGuts]`

in `TcTyClsDecls`: just re-quantify all the type variable binders,

regardless of the order in which the user specified them. After all,

the notion of type variable specificity has no real meaning in

rewrite rules, since one cannot "visibly apply" a rewrite rule.

I have written up this wisdom in

`Note [Re-quantify type variables in rules]` in `TcRules`.

As a result of this patch, compiling the `ExplicitForAllRules1` test

case now generates one fewer warning than it used to. As far as I can

tell, this is benign, since the thing that the disappearing warning

talked about was also mentioned in an entirely separate warning.

Fixes #17710.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/336eac7ef049ba9feab5a948120f0f732e26f773">336eac7e</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-03-11T08:19:08-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Mark ghci056 and ghcilink004 as fragile in unreg

As noted in #17018.

Also fix fragile declaration of T13786, which only runs in the normal

way.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/c61b9b02925acf248cde1ec0f67730c1a6f6c6e5">c61b9b02</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2020-03-11T08:19:44-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Deepen call stack for isIn

I see quite a few warnings like:

  WARNING: file compiler/utils/Util.hs, line 593

    Over-long elem in unionLists

But the call stack is uninformative.   Better to add HasDebugCallStack

to isIn.  Ditto isn'tIn.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/3aa9b35fcc417ab39d8da633482fe64dc9f898b1">3aa9b35f</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2020-03-11T08:20:27-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Zero any slop after compaction in compacting GC

In copying GC, with the relevant debug flags enabled, we release the old

blocks after a GC, and the block allocator zeroes the space before

releasing a block. This effectively zeros the old heap.

In compacting GC we reuse the blocks and previously we didn't zero the

unused space in a compacting generation after compaction. With this

patch we zero the slop between the free pointer and the end of the block

when we're done with compaction and when switching to a new block

(because the current block doesn't have enough space for the next object

we're shifting).

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/8e6febcee4b91a88a5027baac4bee5a8847fe79b">8e6febce</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-03-11T20:33:37-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Refactor GHC.Driver.Session (Ways and Flags)

* extract flags and ways into their own modules (with some renaming)

* remove one SOURCE import of GHC.Driver.Session from GHC.Driver.Phases

* when GHC uses dynamic linking (WayDyn), `interpWays` was only

  reporting WayDyn even if the host was profiled (WayProf).  Now it

  returns both as expected (might fix #16803).

* `mkBuildTag :: [Way] -> String` wasn't reporting a canonical tag for

  differently ordered lists. Now we sort and nub the list to fix this.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/bc41e47123b205a45385a3aa69de97ce22686423">bc41e471</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-03-11T20:33:37-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Refactor interpreterDynamic and interpreterProfiled

* `interpreterDynamic` and `interpreterProfiled` now take `Interp`

  parameters instead of DynFlags

* slight refactoring of `ExternalInterp` so that we can read the iserv

  configuration (which is pure) without reading an MVar.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/a6989971379c26d8c288551d536149675e009e34">a6989971</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-03-11T20:33:37-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Use a Set to represent Ways

Should make `member` queries faster and avoid messing up with missing

`nubSort`.

Metric Increase:

    hie002

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/cb93a1a4405b448e83cad973f93dab3f7f050736">cb93a1a4</a></strong>

<div>

<span>by Ryan Scott</span>

<i>at 2020-03-11T20:34:14-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Make DeriveFunctor-generated code require fewer beta reductions

Issue #17880 demonstrates that `DeriveFunctor`-generated code is

surprisingly fragile when rank-_n_ types are involved. The culprit is

that `$fmap` (the algorithm used to generate `fmap` implementations)

was too keen on applying arguments with rank-_n_ types to lambdas,

which fail to typecheck more often than not.

In this patch, I change `$fmap` (both the specification and the

implementation) to produce code that avoids creating as many lambdas,

avoiding problems when rank-_n_ field types arise.

See the comments titled "Functor instances" in `TcGenFunctor` for a

more detailed description. Not only does this fix #17880, but it also

ensures that the code that `DeriveFunctor` generates will continue

to work after simplified subsumption is implemented (see #17775).

What is truly amazing is that #17880 is actually a regression

(introduced in GHC 7.6.3) caused by commit

49ca2a37bef18aa57235ff1dbbf1cc0434979b1e, the fix #7436. Prior to

that commit, the version of `$fmap` that was used was almost

identical to the one used in this patch! Why did that commit change

`$fmap` then? It was to avoid severe performance issues that would

arise for recursive `fmap` implementations, such as in the example

below:

```hs

data List a = Nil | Cons a (List a) deriving Functor

-- ===>

instance Functor List where

  fmap f Nil = Nil

  fmap f (Cons x xs) = Cons (f x) (fmap (\y -> f y) xs)

```

The fact that `\y -> f y` was eta expanded caused significant

performance overheads. Commit

49ca2a37bef18aa57235ff1dbbf1cc0434979b1e fixed this performance

issue, but it went too far. As a result, this patch partially

reverts 49ca2a37bef18aa57235ff1dbbf1cc0434979b1e.

To ensure that the performance issues pre-#7436 do not resurface,

I have taken some precautionary measures:

* I have added a special case to `$fmap` for situations where the

  last type variable in an application of some type occurs directly.

  If this special case fires, we avoid creating a lambda expression.

  This ensures that we generate

  `fmap f (Cons x xs) = Cons (f x) (fmap f xs)` in the derived

  `Functor List` instance above. For more details, see

  `Note [Avoid unnecessary eta expansion in derived fmap implementations]`

  in `TcGenFunctor`.

* I have added a `T7436b` test case to ensure that the performance

  of this derived `Functor List`-style code does not regress.

When implementing this, I discovered that `$replace`, the algorithm

which generates implementations of `(<$)`, has a special case that is

very similar to the `$fmap` special case described above. `$replace`

marked this special case with a custom `Replacer` data type, which

was a bit overkill. In order to use the same machinery for both

`Functor` methods, I ripped out `Replacer` and instead implemented

a simple way to detect the special case. See the updated commentary

in `Note [Deriving <$]` for more details.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/1f9db3e79bd0d70e5a1491174d540717f3bce2bf">1f9db3e7</a></strong>

<div>

<span>by Kirill Elagin</span>

<i>at 2020-03-12T09:45:51-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">pretty-printer: Properly parenthesise LastStmt

After ApplicatveDo strips the last `return` during renaming, the pretty

printer has to restore it. However, if the return was followed by `$`,

the dollar was stripped too and not restored.

For example, the last stamement in:

```

  foo = do

    x <- ...

    ...

    return $ f x

```

would be printed as:

```

    return f x

```

This commit preserved the dolar, so it becomes:

```

    return $ f x

```

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/5cb93af73499f9cee4a17427629840feb26171e5">5cb93af7</a></strong>

<div>

<span>by Kirill Elagin</span>

<i>at 2020-03-12T09:45:51-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">pretty-printer: Do not print ApplicativeDo join

* Do not print `join` in ApplictiveStmt, unless ppr-debug

* Print parens around multiple parallel binds

When ApplicativeDo is enabled, the renamer analyses the statements of a

`do` block and in certain cases marks them as needing to be rewritten

using `join`.

For example, if you have:

```

foo = do

  a <- e1

  b <- e2

  doSomething a b

```

it will be desugared into:

```

foo = join (doSomething <$> e1 <*> e2)

```

After renaming but before desugaring the expression is stored

essentially as:

```

foo = do

  [will need join] (a <- e1 | b <- e2)

  [no return] doSomething a b

```

Before this change, the pretty printer would print a call to `join`,

even though it is not needed at this stage at all. The expression will be

actually rewritten into one using join only at desugaring, at which

point a literal call to join will be inserted.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/3a259092a02e84d6b45da6b232cfc022898451a0">3a259092</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2020-03-12T09:46:29-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Expose compulsory unfoldings always

The unsafeCoerce# patch requires that unsafeCoerce# has

a compulsory unfolding that is always available.  So we have

to be careful to expose compulsory unfoldings unconditionally

and consistently.

We didn't get this quite right: #17871.  This patch fixes

it.  No real surprises here.

See Note [Always expose compulsory unfoldings] in GHC.Iface.Tidy

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/6a65b8c2f5b4bc7abdb0ca3b5876df694acb8194">6a65b8c2</a></strong>

<div>

<span>by Alp Mestanogullari</span>

<i>at 2020-03-13T02:29:20-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">hadrian: improve dependency tracking for the check-* programs

The code in Rules.Register responsible for finding all the build artifacts

that Cabal installs when registering a library (static/shared libs, .hi files,

...) was looking in the wrong place. This patch fixes that logic and makes sure

we gather all those artifacts in a list to declare that the rule for a given

`.conf` file, our proxy for "Hadrian, please install this package in the package

db for this stage", also produces those artifacts under the said package

database.

We also were completely missing some logic to declare that the check-* programs

have dependencies besides their source code, at least when testing an in-tree

compiler.

Finally, this patch also removes redundant packages from 'testsuitePackages',

since they should already be covered by the stage<N>Packages lists from

Settings.Default.

With this patch, after a complete build and freezing stage 1, a change to

`compiler/parser/Parser.y` results in rebuilding the ghc lib, reinstalling it,

and rebuilding the few programs that depend on it, _including_ `check-ppr` and

`check-api-annotations` (therefore fixing #17273).

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/44fad4a925c06fa2b14611ea08acea9210ee4e00">44fad4a9</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-03-13T02:30:22-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Rename isDllName

I wanted to fix the dangling comment in `isDllName` ("This is the cause

of #", #8696 is already mentioned earlier). I took the opportunity to

change the function name to better reflect what it does.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/2f292db8f10134919c420b71b8a5eccd379212ad">2f292db8</a></strong>

<div>

<span>by Paavo</span>

<i>at 2020-03-13T02:31:03-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Update documentation for closureSize

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/f124ff0dfccced755ee97ecac027119269996f8f">f124ff0d</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-03-13T02:31:40-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Rework triggering of release builds

Use a push option instead of tagging.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/7f25557a6f240943ebe3eb3b3c7178e76a3e93e1">7f25557a</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-03-13T10:38:09-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Distinguish integer-simple test envs

Previously two integer-simple jobs declared the same test environment. One (the nightly job) was built in the perf way, the other in the validate way. Consequently they had appreciably different performance characteristics, causing in the nightly job to spuriously fail with performance changes.</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/c12a2ec5fe4e7f94d565c0e6398d1d79854db146">c12a2ec5</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2020-03-14T05:25:30-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix Lint

Ticket #17590 pointed out a bug in the way the linter dealt with

type lets, exposed by the new uniqAway story.

The fix is described in Note [Linting type lets]. I ended up

putting the in-scope Ids in a different env field, le_ids,

rather than (as before) sneaking them into the TCvSubst.

Surprisingly tiresome, but done.

Metric Decrease:

    hie002

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/b989845e378081f0932a7fec528e68daeeaa14fb">b989845e</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-03-14T05:26:11-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Hadrian: fix absolute buildroot support (#17822)

Shake's "**" wildcard doesn't match absolute root. We must use "//" instead.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/4f117135673b36816c343bc11efcbb8396160c75">4f117135</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-03-14T05:26:49-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Make: refactor GMP rules

Document and use simpler rules for the ghc-gmp.h header.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/7432b327a3c43c4e2e7c777b41a8e17899b8d0d1">7432b327</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-03-14T05:27:28-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Use correct option name (-opti) (fix #17314)

s/pgmo/opti

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/8f7dd5710b80906ea7a3e15b7bb56a883a49fed8">8f7dd571</a></strong>

<div>

<span>by Judah Jacobson</span>

<i>at 2020-03-14T05:28:07-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Allow overriding LD_STAGE0 and AR_STAGE0 in the configure script.

Previously it was possible to override the stage0 C compiler via `CC_STAGE0`,

but you couldn't override `ld` or `ar` in stage0.  This change allows overriding them

by setting `LD_STAGE0` or `AR_STAGE0`, respectively.

Our team uses this feature internally to take more control of our GHC build

and make it run more hermetically.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/7c3e39a9a7ccb3b6c2953b0397a0d315dc0ec7d5">7c3e39a9</a></strong>

<div>

<span>by Judah Jacobson</span>

<i>at 2020-03-14T05:28:07-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Use AC_ARG_VAR for LD_STAGE0 and AR_STAGE0.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/20d4d676964382b313b9e44062e45a7c38621999">20d4d676</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-03-14T05:28:43-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">nonmoving: Don't traverse filled segment list in pause

The non-moving collector would previously walk the entire filled segment

list during the preparatory pause. However, this is far more work than

is strictly necessary. We can rather get away with merely collecting the

allocators' filled segment list heads and process the lists themselves

during the concurrent phase. This can significantly reduce the maximum

gen1 GC pause time in programs with high rates of long-lived allocations.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/fdfa2d0121ca8cc22479dd17a74afa58fc2b39f5">fdfa2d01</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-03-14T05:29:18-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">nonmoving: Remove redundant bitmap clearing

nonmovingSweep already clears the bitmap in the sweep loop. There is no

reason to do so a second time.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/2f8c77673f1faf0d8fed6df2bdd1ca15d696a010">2f8c7767</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2020-03-14T05:29:55-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Simple refactor of cheapEqExpr

No change in functionality.  Just seems tidier (and signficantly more

efficient) to deal with ticks directly than to call stripTicksTopE.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/88f7a76208b0fcf41ca0e16d18a71523f0601ee5">88f7a762</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2020-03-14T05:29:55-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Improve CSE.combineAlts

This patch improves the way that CSE combines identical

alternatives.  See #17901.

I'm still not happy about the duplication between CSE.combineAlts

and GHC.Core.Utils.combineIdenticalAlts; see the Notes with those

functions.  But this patch is a step forward.

Metric Decrease:

    T12425

    T5642

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/8b95ddd3f20a67acf5251347d80f9cab191bdfc4">8b95ddd3</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-03-14T05:30:31-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Add integer-simple release build for Windows

Closes #16144.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/e3c374cc5bd7eb49649b9f507f9f7740697e3f70">e3c374cc</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2020-03-14T05:31:07-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Wrap an implication around class-sig kind errors

Ticket #17841 showed that we can get a kind error

in a class signature, but lack an enclosing implication

that binds its skolems.

This patch

* Adds the wrapping implication: the new call to

  checkTvConstraints in tcClassDecl1

* Simplifies the API to checkTvConstraints, which

  was not otherwise called at all.

* Simplifies TcErrors.report_unsolved by *not*

  initialising the TidyEnv from the typechecker lexical

  envt.  It's enough to do so from the free vars of the

  unsolved constraints; and we get silly renamings if

  we add variables twice: once from the lexical scope

  and once from the implication constraint.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/73133a3b601b76c46098fb8ad3c76de5fe04c9b2">73133a3b</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2020-03-14T05:31:07-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Refactoring in TcSMonad

This patch is just refactoring: no change in

behaviour.

I removed the rather complicated

    checkConstraintsTcS

    checkTvConstraintsTcS

in favour of simpler functions

    emitImplicationTcS

    emitTvImplicationTcS

    pushLevelNoWorkList

The last of these is a little strange, but overall

it's much better I think.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/93c88c266eacd80a7f2a1754778167390c287b18">93c88c26</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-03-14T05:31:42-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">base: Make `open` calls interruptible

As noted in #17912, `open` system calls were `safe` rather than

`interruptible`. Consequently, the program could not be interrupted with

SIGINT if stuck in a slow open operation. Fix this by marking

`c_safe_open` as interruptible.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/bee4cdad4ce68a5bbe6af493d99f0197a34eef5c">bee4cdad</a></strong>

<div>

<span>by Vladislav Zavialov</span>

<i>at 2020-03-14T05:32:18-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Remove second tcLookupTcTyCon in tcDataDefn

Before this patch, tcDataDefn used to call tcLookupTcTyCon twice in a row:

        1. in bindTyClTyVars itself

        2. in the continuation passed to it

Now bindTyClTyVars passes the TcTyCon to the continuation, making

the second lookup unnecessary.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/3f116d35a5b028ae2bb22da5063900d2f3d0b476">3f116d35</a></strong>

<div>

<span>by Cale Gibbard</span>

<i>at 2020-03-14T19:34:42-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Enable stage1 build of haddock

The submodule has already been bumped to contain the fix.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/49e9d7395a13584f15798bce1601b8d3a6633f11">49e9d739</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2020-03-14T19:35:24-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">rts: Fix printClosure when printing fwd ptrs

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/1de3ab4a147eeb0b34b24a3c0e91f174e6e5cb79">1de3ab4a</a></strong>

<div>

<span>by Krzysztof Gogolewski</span>

<i>at 2020-03-14T19:36:04-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Remove unused field var_inline (#17915)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/d30aeb4b38381758025bc1002eb2135ad6bc58b8">d30aeb4b</a></strong>

<div>

<span>by Krzysztof Gogolewski</span>

<i>at 2020-03-15T03:57:41-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Document restriction on SCC pragma syntax

Currently, the names of cost centres must be quoted or

be lowercase identifiers.

Fixes #17916.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/b4774598e6bd060b1b4230b11b734ca40022980d">b4774598</a></strong>

<div>

<span>by Brian Foley</span>

<i>at 2020-03-15T03:58:18-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Remove some dead code

>From the notes.ghc.drop list found using weeder in #17713

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/dd6ffe6be742cf3ec98406704fef53ad86cc1560">dd6ffe6b</a></strong>

<div>

<span>by Viktor Dukhovni</span>

<i>at 2020-03-15T03:58:55-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Note platform-specific Foreign.C.Types in context

Also fix the markup in the general note at the top of the module.  Haddock

(usability trade-off), does not support multi-line emphasised text.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/2e82465fff5851f00449131fdc8bacd3ca95f90f">2e82465f</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-03-15T10:57:10-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Refactor CmmToAsm (disentangle DynFlags)

This patch disentangles a bit more DynFlags from the native code

generator (CmmToAsm).

In more details:

- add a new NCGConfig datatype in GHC.CmmToAsm.Config which contains the

  configuration of a native code generation session

- explicitly pass NCGConfig/Platform arguments when necessary

- as a consequence `sdocWithPlatform` is gone and there are only a few

  `sdocWithDynFlags` left

- remove the use of `unsafeGlobalDynFlags` from GHC.CmmToAsm.CFG

- remove `sdocDebugLevel` (now we pass the debug level via NCGConfig)

There are still some places where DynFlags is used, especially because

of pretty-printing (CLabel), because of Cmm helpers (such as

`cmmExprType`) and because of `Outputable` instance for the

instructions. These are left for future refactoring as this patch is

already big.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/c35c545d3f32f092c52052349f741739a844ec0f">c35c545d</a></strong>

<div>

<span>by Judah Jacobson</span>

<i>at 2020-03-15T10:57:48-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add a -no-haddock flag.

This flag undoes the effect of a previous "-haddock" flag.  Having both flags makes it easier

for build systems to enable Haddock parsing in a set of global flags, but then disable it locally for

specific targets (e.g., third-party packages whose comments don't pass the validation in the latest GHC).

I added the flag to expected-undocumented-flags.txt since `-haddock` was alreadyin that list.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/cfcc3c9a1f2e4e33bed4c40767f8e7971e331c15">cfcc3c9a</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2020-03-15T10:58:27-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix global_link of TSOs for threads reachable via dead weaks

Fixes #17785

Here's how the problem occurs:

- In generation 0 we have a TSO that is finished (i.e. it has no more

  work to do or it is killed).

- The TSO only becomes reachable after collectDeadWeakPtrs().

- After collectDeadWeakPtrs() we switch to WeakDone phase where we don't

  move TSOs to different lists anymore (like the next gen's thread list

  or the resurrected_threads list).

- So the TSO will never be moved to a generation's thread list, but it

  will be promoted to generation 1.

- Generation 1 collected via mark-compact, and because the TSO is

  reachable it is marked, and its `global_link` field, which is bogus at

  this point (because the TSO is not in a list), will be threaded.

- Chaos ensues.

In other words, when these conditions hold:

- A TSO is reachable only after collectDeadWeakPtrs()

- It's finished (what_next is ThreadComplete or ThreadKilled)

- It's retained by mark-compact collector (moving collector doesn't

  evacuate the global_list field)

We end up doing random mutations on the heap because the TSO's

global_list field is not valid, but it still looks like a heap pointer

so we thread it during compacting GC.

The fix is simple: when we traverse old_threads lists to resurrect

unreachable threads the threads that won't be resurrected currently

stays on the old_threads lists. Those threads will never be visited

again by MarkWeak so we now reset the global_list fields. This way

compacting GC does not thread pointers to nowhere.

Testing

-------

The reproducer in #17785 is quite large and hard to build, because of

the dependencies, so I'm not adding a regression test.

In my testing the reproducer would take a less than 5 seconds to run,

and once in every ~5 runs would fail with a segfault or an assertion

error. In other cases it also fails with a test failure. Because the

tests never fail with the bug fix, assuming the code is correct, this

also means that this bug can sometimes lead to incorrect runtime

results.

After the fix I was able to run the reproducer repeatedly for about an

hour, with no runtime crashes or test failures.

To run the reproducer clone the git repo:

    $ git clone https://github.com/osa1/streamly --branch ghc-segfault

Then clone primitive and atomic-primops from their git repos and point

to the clones in cabal.project.local. The project should then be

buildable using GHC HEAD. Run the executable `properties` with `+RTS -c

-DZ`.

In addition to the reproducer above I run the test suite using:

    $ make slowtest EXTRA_HC_OPTS="-debug -with-rtsopts=-DS \

        -with-rtsopts=-c +RTS -c -RTS" SKIPWAY='nonmoving nonmoving_thr'

This enables compacting GC always in both GHC when building the test

programs and when running the test programs, and also enables sanity

checking when running the test programs. These set of flags are not

compatible for all tests so there are some failures, but I got the same

set of failures with this patch compared to GHC HEAD.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/818b3c38e7548f4720815f76969238d82c9650f7">818b3c38</a></strong>

<div>

<span>by Lysxia</span>

<i>at 2020-03-16T23:52:42-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">base: add strict IO functions: readFile', getContents', hGetContents'

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/18a346a4b5a02b8c62e8eedb91b35c2d8e754b96">18a346a4</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-03-16T23:53:24-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Modules: Core (#13009)

Update submodule: haddock

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/92327e3afd9d2650c9cc9610297d40c2712da085">92327e3a</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2020-03-16T23:54:04-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Update sanity checking for TSOs:

- Remove an invalid assumption about GC checking what_next field. The GC

  doesn't care about what_next at all, if a TSO is reachable then all

  its pointers are followed (other than global_tso, which is only

  followed by compacting GC).

- Remove checkSTACK in checkTSO: TSO stacks will be visited in

  checkHeapChain, or checkLargeObjects etc.

- Add an assertion in checkTSO to check that the global_link field is

  sane.

- Did some refactor to remove forward decls in checkGlobalTSOList and

  added braces around single-statement if statements.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/e1aa40525ddac370766907a49682976e2ea78422">e1aa4052</a></strong>

<div>

<span>by PHO</span>

<i>at 2020-03-17T07:36:09-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Don't use non-portable operator "==" in configure.ac

The test operator "==" is a Bash extension and produces a wrong result

if /bin/sh is not Bash.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/89f034ddeb4ad7d93c877d5bdbcf2cca7a44f79c">89f034dd</a></strong>

<div>

<span>by Maximilian Tagher</span>

<i>at 2020-03-17T07:36:48-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Document the units of -ddump-timings

Right now, in the output of -ddump-timings to a file, you can't tell what the units are:

```

CodeGen [TemplateTestImports]: alloc=22454880 time=14.597

```

I believe bytes/milliseconds are the correct units, but confirmation would be appreciated. I'm basing it off of this snippet from `withTiming'`:

```

when (verbosity dflags >= 2 && prtimings == PrintTimings)

  $ liftIO $ logInfo dflags (defaultUserStyle dflags)

      (text "!!!" <+> what <> colon <+> text "finished in"

       <+> doublePrec 2 time

       <+> text "milliseconds"

       <> comma

       <+> text "allocated"

       <+> doublePrec 3 (realToFrac alloc / 1024 / 1024)

       <+> text "megabytes")

```

which implies time is in milliseconds, and allocations in bytes (which divided by 1024 would be KB, and again would be MB)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/beffa14771ebd6ba24b20337f29045364621c5fa">beffa147</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2020-03-17T07:37:25-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Implement mapTyCo like foldTyCo

This patch makes mapType use the successful idiom described

in TyCoRep

   Note [Specialising foldType]

I have not yet changed any functions to use mapType, though there

may be some suitable candidates.

This patch should be a no-op in terms of functionality but,

because it inlines the mapper itself, I'm hoping that there may

be some modest perf improvements.

Metric Decrease:

    T5631

    T5642

    T3064

    T9020

    T14683

    hie002

    haddock.Cabal

    haddock.base

    haddock.compiler

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/5800ebfeb2fe3e3ed985cdf08a66defea73db71d">5800ebfe</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2020-03-17T07:38:08-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Don't update ModDetails with CafInfos when opts are disabled

This is consistent with the interface file behavior where we omit

HsNoCafRefs annotations with -fomit-interface-pragmas (implied by -O0).

ModDetails and ModIface are just different representations of the same

thing, so they really need to be in sync. This patch does the right

thing and does not need too much explanation, but here's an example of a

problem not doing this causes in !2842:

    -- MyInteger.hs

    module MyInteger

      ( MyInteger (MyInteger)

      , ToMyInteger (toMyInteger)

      ) where

    newtype MyInteger = MyInteger Integer

    class ToMyInteger a where

      toMyInteger :: a -> MyInteger

    instance ToMyInteger Integer where

      toMyInteger = MyInteger {- . succ -}

    -- Main.hs

    module Main

      ( main

      ) where

    import MyInteger (MyInteger (MyInteger), toMyInteger)

    main :: IO ()

    main = do

      let (MyInteger i) = (id . toMyInteger) (41 :: Integer)

      print i

If I build this with -O0, without this fix, we generate a ModDetails with

accurate LFInfo for toMyInteger (MyInteger.$fToMyIntegerInteger) which says that

it's a LFReEntrant with arity 1. This means in the use site (Main) we tag the

value:

    R3 = MyInteger.$fToMyIntegerInteger_closure + 1;

    R2 = GHC.Base.id_closure;

    R1 = GHC.Base.._closure;

    Sp = Sp - 16;

    call stg_ap_ppp_fast(R4, R3, R2, R1) args: 24, res: 0, upd: 24;

Now we change the definition by uncommenting the `succ` part and it becomes a thunk:

    MyInteger.$fToMyIntegerInteger [InlPrag=INLINE (sat-args=0)]

      :: MyInteger.ToMyInteger GHC.Integer.Type.Integer

    [GblId[DFunId(nt)]] =

        {} \u [] $ctoMyInteger_rEA;

and its LFInfo is now LFThunk. This change in LFInfo makes a difference in the

use site: we can no longer tag it.

But becuase the interface fingerprint does not change (because ModIface does not

change) we don't rebuild Main and tag the thunk.

(1.2% increase in allocations when building T12545 on armv7 because we

generate more code without CafInfos)

Metric Increase:

    T12545

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/5b632dad2e1d373606fe29f7eee0daf15641560f">5b632dad</a></strong>

<div>

<span>by Paavo</span>

<i>at 2020-03-17T07:38:48-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add example for Data.Semigroup.diff

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/4d85d68b004a3d577c0e27d9e6fdddd118e8a9f3">4d85d68b</a></strong>

<div>

<span>by Paavo</span>

<i>at 2020-03-17T07:38:48-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Clean up

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/75168d07c9c30289709423fc184bbab8dcad0f4e">75168d07</a></strong>

<div>

<span>by Paavo</span>

<i>at 2020-03-17T07:38:48-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Make example collapsible

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/53ff2cd0c49735e8f709ac8a5ceab68483eb89df">53ff2cd0</a></strong>

<div>

<span>by Richard Eisenberg</span>

<i>at 2020-03-17T13:46:57+00:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix #17021 by checking more return kinds

All the details are in new Note [Datatype return kinds] in

TcTyClsDecls.

Test case: typecheck/should_fail/T17021{,b}

           typecheck/should_compile/T17021a

Updates haddock submodule

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/528df8ecb4e2f9c78b1ae4ab7ff8230644e9b643">528df8ec</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-03-18T10:06:43-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Modules: Core operations (#13009)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/4e8a71c1138b587dfbab8a1823b3f7fa6f0166bd">4e8a71c1</a></strong>

<div>

<span>by Richard Eisenberg</span>

<i>at 2020-03-18T10:07:19-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add release note about fix to #16502.

We thought we needed to update the manual, but the fix for #16502

actually brings the implementation in line with the manual. So we

just alert users of how to update their code.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/5cbf9934c59f7726781cc4cccf4748a5c09c4997">5cbf9934</a></strong>

<div>

<span>by Andreas Klebinger</span>

<i>at 2020-03-19T00:39:27-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Update "GHC differences to the FFI Chapter" in user guide.

The old entry had a heavy focus on how things had been. Which is

not what I generally look for in a user guide.

I also added a small section on behaviour of nested safe ffi calls.

[skip-ci]

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/b03fd3bcd4ff14aed2942275c3b0db5392dc913c">b03fd3bc</a></strong>

<div>

<span>by Sebastian Graf</span>

<i>at 2020-03-19T00:40:06-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">PmCheck: Use ConLikeSet to model negative info

In #17911, Simon recognised many warnings stemming from over-long list

unions while coverage checking Cabal's `LicenseId` module.

This patch introduces a new `PmAltConSet` type which uses a `UniqDSet`

instead of an association list for `ConLike`s. For `PmLit`s, it will

still use an assocation list, though, because a similar map data

structure would entail a lot of busy work.

Fixes #17911.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/64f207566931469648e791df4f0f0384d45cddd0">64f20756</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-03-19T12:16:49-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Refactoring: use Platform instead of DynFlags when possible

Metric Decrease:

    ManyConstructors

    T12707

    T13035

    T1969

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/cb1785d9f839e34a3a4892f354f0c51cc6553c0e">cb1785d9</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2020-03-19T12:16:54-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">FastString: fix eager reading of string ptr in hashStr

This read causes NULL dereferencing when len is 0.

Fixes #17909

In the reproducer in #17909 this bug is triggered as follows:

- SimplOpt.dealWithStringLiteral is called with a single-char string

  ("=" in #17909)

- tailFS gets called on the FastString of the single-char string.

- tailFS checks the length of the string, which is 1, and calls

  mkFastStringByteString on the tail of the ByteString, which is an

  empty ByteString as the original ByteString has only one char.

- ByteString's unsafeUseAsCStringLen returns (NULL, 0) for the empty

  ByteString, which is passed to mkFastStringWith.

- mkFastStringWith gets hash of the NULL pointer via hashStr, which

  fails on empty strings because of this bug.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/73a7383ebc17f495d7acd04007c8c56b46532cb6">73a7383e</a></strong>

<div>

<span>by Richard Eisenberg</span>

<i>at 2020-03-20T20:42:56-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Simplify treatment of heterogeneous equality

Previously, if we had a [W] (a :: k1) ~ (rhs :: k2), we would

spit out a [D] k1 ~ k2 and part the W as irreducible, hoping for

a unification. But we needn't do this. Instead, we now spit out

a [W] co :: k2 ~ k1 and then use co to cast the rhs of the original

Wanted. This means that we retain the connection between the

spat-out constraint and the original.

The problem with this new approach is that we cannot use the

casted equality for substitution; it's too like wanteds-rewriting-

wanteds. So, we forbid CTyEqCans that mention coercion holes.

All the details are in Note [Equalities with incompatible kinds]

in TcCanonical.

There are a few knock-on effects, documented where they occur.

While debugging an error in this patch, Simon and I ran into

infelicities in how patterns and matches are printed; we made

small improvements.

This patch includes mitigations for #17828, which causes spurious

pattern-match warnings. When #17828 is fixed, these lines should

be removed.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/faa36e5b3674a7b2cfc6b931eec27b3558fad33b">faa36e5b</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-03-20T20:43:41-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Hadrian: ignore in-tree GMP objects with ``--lint``

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/9a96ff6b1c19c9d0af9c9a39fb2c086f311c7239">9a96ff6b</a></strong>

<div>

<span>by Richard Eisenberg</span>

<i>at 2020-03-20T20:44:17-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Update core spec to reflect changes to Core.

Key changes:

 * Adds a new rule for forall-coercions over coercion variables, which

was implemented but conspicuously missing from the spec.

 * Adds treatment for FunCo.

 * Adds treatment for ForAllTy over coercion variables.

 * Improves commentary (including restoring a Note lost in

03d4852658e1b7407abb4da84b1b03bfa6f6db3b) in the source.

No changes to running code.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/7e0451c6ce78b08b5f7564aa03a9a86051b35163">7e0451c6</a></strong>

<div>

<span>by Sergej Jaskiewicz</span>

<i>at 2020-03-20T20:44:55-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix event message in withTiming'

This typo caused generating 'end' events without the corresponding 'begin' events.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/1542a626a3649e5fc1becec1eff2715a6108d144">1542a626</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-03-22T22:37:47-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">fs.h: Add missing declarations on Windows

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/3bcf2ccdd0bdd3cf176943ddb946e4fcef2ac7c1">3bcf2ccd</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-03-22T22:37:47-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Bump process submodule

Avoids redundant case alternative warning.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/3b363ef9405c52369b63fc167ded8a533c6a5079">3b363ef9</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-03-22T22:37:47-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Normalize slashes in ghc-api annotations output

Enable `normalise_slashes` on `annotations`, `listcomps`, and

`parseTree` to fix Windows failures.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/25fc94295c68620c6928751aaadc46b11a185f0c">25fc9429</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-03-22T22:37:47-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Update expected output on Windows

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/7f58ec6d94f49cf5713757710c6b80a682ee066b">7f58ec6d</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-03-22T22:37:47-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Fix TOP of T17786

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/aadcd909563d5ddf7b2fc469ee3c09c301b1a813">aadcd909</a></strong>

<div>

<span>by GHC GitLab CI</span>

<i>at 2020-03-22T22:37:47-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Update expected output on Windows

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/dc1eb10dae03f3fe5f11de845c6772c229684c28">dc1eb10d</a></strong>

<div>

<span>by GHC GitLab CI</span>

<i>at 2020-03-22T22:37:47-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">hadrian: Fix executable extension passed to testsuite driver

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/58f62e2caa19e13082807da1d49f94e9cf869322">58f62e2c</a></strong>

<div>

<span>by GHC GitLab CI</span>

<i>at 2020-03-22T22:37:47-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Require that Windows-hadrian job passes

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/8dd2415dd2f423688f33a8fb267857e3ff39754d">8dd2415d</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-03-22T22:37:47-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">hadrian: Eliminate redundant .exe from GHC path

Previously we were invoking:

    bash -c

    "c:/GitLabRunner/builds/eEQrxK4p/0/ghc/ghc/toolchain/bin/ghc.exe.exe

    testsuite/mk/ghc-config.hs -o _build/test/bin/ghc-config.exe"

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/373621f61f812accdeba39c745eb7d015430ff8e">373621f6</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-03-22T22:37:47-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Bump hsc2hs submodule

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/abc02b4036c2d8efe50b720d8c8103c4f1b8899a">abc02b40</a></strong>

<div>

<span>by Hécate</span>

<i>at 2020-03-22T22:38:33-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Annotate the non-total function in Data.Foldable as such

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/19f125578247dfe8036b5793cb3f6b684474f9c7">19f12557</a></strong>

<div>

<span>by Josef Svenningsson</span>

<i>at 2020-03-23T14:05:33-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix ApplicativeDo regression #17835

A previous fix for #15344 made sure that monadic 'fail' is used properly

when translating ApplicativeDo. However, it didn't properly account

for when a 'fail' will be inserted which resulted in some programs

failing with a type error.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/2643ba465cd2a133b6f495f34fc59cd1a6d23525">2643ba46</a></strong>

<div>

<span>by Paavo</span>

<i>at 2020-03-24T08:31:32-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add example and doc for Arg (Fixes #17153)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/703221f408b023a1b3433938572e7b5c24b4af60">703221f4</a></strong>

<div>

<span>by Roland Senn</span>

<i>at 2020-03-25T14:45:04-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Use export list of Main module in function TcRnDriver.hs:check_main (Fix #16453)

- Provide the export list of the `Main` module as parameter to the

  `compiler/typecheck/TcRnDriver.hs:check_main` function.

- Instead of `lookupOccRn_maybe` call the function `lookupInfoOccRn`.

  It returns the list `mains_all` of all the main functions in scope.

- Select from this list `mains_all` all `main` functions that are in

  the export list of the `Main` module.

- If this new list contains exactly one single `main` function, then

  typechecking continues.

- Otherwise issue an appropriate error message.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/3e27205a66b06a4501d87eb31e285eadbc693eb7">3e27205a</a></strong>

<div>

<span>by Sebastian Graf</span>

<i>at 2020-03-25T14:45:40-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Remove -fkill-absence and -fkill-one-shot flags

They seem to be a benchmarking vestige of the Cardinality paper and

probably shouldn't have been merged to HEAD in the first place.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/262e42aa34c4d5705c8d011907c351497dd4e862">262e42aa</a></strong>

<div>

<span>by Peter Trommler</span>

<i>at 2020-03-25T22:41:39-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Do not panic on linker errors

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/0de03cd78729dc58a846c64b645e71057ec5d24e">0de03cd7</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-03-25T22:42:02-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">DynFlags refactoring III

Use Platform instead of DynFlags when possible:

* `tARGET_MIN_INT` et al. replaced with `platformMinInt` et al.

* no more DynFlags in PreRules: added a new `RuleOpts` datatype

* don't use `wORD_SIZE` in the compiler

* make `wordAlignment` use `Platform`

* make `dOUBLE_SIZE` a constant

Metric Decrease:

    T13035

    T1969

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/7a04920b23376b2759b5049820400ddd40f7b89e">7a04920b</a></strong>

<div>

<span>by Tristan Cacqueray</span>

<i>at 2020-03-25T22:42:06-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Base: fix a typo in liftA doc

This change removes an extra '|' that should not be rendered in

the liftA documentation.

Tracking: #17929

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/1c5a15f7d257d0a2ded4850e38b6174965a7735b">1c5a15f7</a></strong>

<div>

<span>by Tristan Cacqueray</span>

<i>at 2020-03-25T22:42:06-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Base: add Control.Applicative optional example

This change adds an optional example.

Tracking: #17929

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/6d172e63f3dd3590b0a57371efb8f924f1fcdf05">6d172e63</a></strong>

<div>

<span>by Tristan Cacqueray</span>

<i>at 2020-03-25T22:42:06-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Base: add markup around Except

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/eb2162c81964b3a1057c21c73574cc2b542d863a">eb2162c8</a></strong>

<div>

<span>by John Ericson</span>

<i>at 2020-03-26T12:37:08-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Remove unused `ghciTablesNextToCode` from compiler proper

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/f51efc4bf5f20d1df3d11a2bb4c147f416367ae4">f51efc4b</a></strong>

<div>

<span>by Joachim Breitner</span>

<i>at 2020-03-26T12:37:09-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Prepare to use run-time tablesNextToCode in compiler exclusively

Factor out CPP as much as possible to prepare for runtime

determinattion.

Progress towards #15548

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/1c446220250dcada51d4bb33a0cc7d8ce572e8b6">1c446220</a></strong>

<div>

<span>by Joachim Breitner</span>

<i>at 2020-03-26T12:37:09-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Use run-time tablesNextToCode in compiler exclusively (#15548)

Summary:

 - There is no more use of the TABLES_NEXT_TO_CODE CPP macro in

   `compiler/`. GHCI_TABLES_NEXT_TO_CODE is also removed entirely.

   The field within `PlatformMisc` within `DynFlags` is used instead.

 - The field is still not exposed as a CLI flag. We might consider some

   way to ensure the right RTS / libraries are used before doing that.

Original reviewers:

Original subscribers: TerrorJack, rwbarton, carter

Original Differential Revision: https://phabricator.haskell.org/D5082

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/1941ef4f050c0dfcb68229641fcbbde3a10f1072">1941ef4f</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-03-29T17:28:51-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Modules: Types (#13009)

Update Haddock submodule

Metric Increase:

   haddock.compiler

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/1c7c6f1afc8e7f7ba5d256780bc9d5bb5f3e7601">1c7c6f1a</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-03-29T17:28:51-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Remove GHC.Types.Unique.Map module

This module isn't used anywhere in GHC.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/f1a6c73d01912b389e012a0af81a5c2002e82636">f1a6c73d</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-03-29T17:28:51-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Merge GHC.Types.CostCentre.Init into GHC.Driver.CodeOutput

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/54250f2d8de910b094070c1b48f086030df634b1">54250f2d</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2020-03-29T17:29:30-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Demand analysis: simplify the demand for a RHS

Ticket #17932 showed that we were using a stupid demand for the RHS

of a let-binding, when the result is a product.  This was the result

of a "fix" in 2013, which (happily) turns out to no longer be

necessary.

So I just deleted the code, which simplifies the demand analyser,

and fixes #17932. That in turn uncovered that the anticipation

of worker/wrapper in CPR analysis was inaccurate, hence the logic

that decides whether to unbox an argument in WW was extracted into

a function `wantToUnbox`, now consulted by CPR analysis.

I tried nofib, and got 0.0% perf changes.

All this came up when messing about with !2873 (ticket #17917),

but is idependent of it.

Unfortunately, this patch regresses #4267 and realised that it is now

blocked on #16335.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/03060b2f963436dd22582f3a1c218b7a74d302ec">03060b2f</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-03-29T17:30:05-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Fix T17786 on Windows

Fixes line ending normalization issue.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/1f7995baaaf4dcba9b3abaddcd571ac66675ab19">1f7995ba</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-03-29T17:30:05-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Fix T17786

Fix missing quoting and expected exit code.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/ef9c608eba417c59fe45c9edd6a946c59f50b5d2">ef9c608e</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-03-29T17:30:05-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Mark T12971 as broken on Windows

Due to #17945.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/e54500c12de051cb9695728d27c812e5160593ee">e54500c1</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-03-29T17:30:47-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Store ComponentId details

As far as GHC is concerned, installed package components ("units") are

identified by an opaque ComponentId string provided by Cabal. But we

don't want to display it to users (as it contains a hash) so GHC queries

the database to retrieve some infos about the original source package

(name, version, component name).

This patch caches these infos in the ComponentId itself so that we don't

need to provide DynFlags (which contains installed package informations)

to print a ComponentId.

In the future we want GHC to support several independent package states

(e.g. for plugins and for target code), hence we need to avoid

implicitly querying a single global package state.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/7e7cb714173652165b8372c3450f4ccd6a377497">7e7cb714</a></strong>

<div>

<span>by Marius Bakke</span>

<i>at 2020-03-29T17:31:27-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Remove test that dlopens a PIE object.

glibc 2.30 disallowed dlopening PIE objects, so just remove the test.

Fixes #17952.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/6c8f80d8876965539571d37d3c7f0fee43c300d2">6c8f80d8</a></strong>

<div>

<span>by Andreas Klebinger</span>

<i>at 2020-03-29T17:32:04-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Correct haddocks for testBit in Data.Bits

It conflated the nth bit with the bit at offset n.

Now we instead give the definition in terms of `bit and `.&.`

on top of clearer phrasing.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/c916f190b455d9c43459f81c7ee06d06ab3db957">c916f190</a></strong>

<div>

<span>by Andreas Klebinger</span>

<i>at 2020-03-29T17:32:04-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Apply suggestion to libraries/base/Data/Bits.hs</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/64bf7f51064dad9c63728ac8bccdb9cf00bdb420">64bf7f51</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-03-29T17:32:41-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Add FreeBSD release job

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/a0d8e92e9c9b67426aa139d6bc46363d8940f992">a0d8e92e</a></strong>

<div>

<span>by Ryan Scott</span>

<i>at 2020-03-29T17:33:20-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Run checkNewDataCon before constraint-solving newtype constructors

Within `checkValidDataCon`, we used to run `checkValidType` on the

argument types of a newtype constructor before running

`checkNewDataCon`, which ensures that the user does not attempt

non-sensical things such as newtypes with multiple arguments or

constraints. This works out in most situations, but this falls over

on a corner case revealed in #17955:

```hs

newtype T = Coercible () T => T ()

```

`checkValidType`, among other things, peforms an ambiguity check on

the context of a data constructor, and that it turn invokes the

constraint solver. It turns out that there is a special case in the

constraint solver for representational equalities (read: `Coercible`

constraints) that causes newtypes to be unwrapped (see

`Note [Unwrap newtypes first]` in `TcCanonical`). This special case

does not know how to cope with an ill formed newtype like `T`, so

it ends up panicking.

The solution is surprisingly simple: just invoke `checkNewDataCon`

before `checkValidType` to ensure that the illicit newtype

constructor context is detected before the constraint solver can

run amok with it.

Fixes #17955.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/45eb9d8cad254440eaea25676d6788ca13baa2fb">45eb9d8c</a></strong>

<div>

<span>by Krzysztof Gogolewski</span>

<i>at 2020-03-29T17:33:59-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Minor cleanup

- Simplify mkBuildExpr, the function newTyVars was called

  only on a one-element list.

- TTG: use noExtCon in more places. This is more future-proof.

- In zonkExpr, panic instead of printing a warning.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/f024b6e385bd1448968b7bf20de05f655c815bae">f024b6e3</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-03-30T12:48:39+02:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Expect T4267 to pass

Since 54250f2d8de910b094070c1b48f086030df634b1 we expected T4267 to

fail, but it passes on CI.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/57b888c0e90be7189285a6b078c30b26d0923809">57b888c0</a></strong>

<div>

<span>by Ryan Scott</span>

<i>at 2020-03-31T10:54:20-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Require GHC 8.8 as the minimum compiler for bootstrapping

This allows us to remove several bits of CPP that are either always

true or no longer reachable. As an added bonus, we no longer need to

worry about importing `Control.Monad.Fail.fail` qualified to avoid

clashing with `Control.Monad.fail`, since the latter is now the same

as the former.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/33f095511a8fce4c945bbcd4feb3910c854dcb61">33f09551</a></strong>

<div>

<span>by Ryan Scott</span>

<i>at 2020-03-31T10:54:57-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add regression test for #17963

The panic in #17963 happened to be fixed by commit

e3c374cc5bd7eb49649b9f507f9f7740697e3f70. This patch adds a

regression test to ensure that it remains fixed.

Fixes #17963.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/09a36e80ecaefcfb60eccda98bd06461d0aeca70">09a36e80</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2020-03-31T10:55:37-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Simplify stderrSupportsAnsiColors

The combinator andM is used only once, and the code is shorter and

simpler if you inline it.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/95bccdd034ce4dd2d1bc36db9f1ba5e172550249">95bccdd0</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-03-31T10:56:19-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">base: Ensure that encoding global variables aren't inlined

As noted in #17970, these (e.g. `getFileSystemEncoding` and

`setFileSystemEncoding`) previously had unfoldings, which would

break their global-ness.

While not strictly necessary, I also add a NOINLINE on

`initLocaleEncoding` since it is used in `System.IO`, ensuring that we

only system's query the locale encoding once.

Fixes #17970.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/982aaa837aed564ae9b418cda8e97d4facff8fb8">982aaa83</a></strong>

<div>

<span>by Andreas Klebinger</span>

<i>at 2020-03-31T10:56:55-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Update hadrian index revision.

Required in order to build hadrian using ghc-8.10

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/4b9c586472bf99425f7bbcf346472d7c54f05028">4b9c5864</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-03-31T10:57:32-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">integer-gmp: Bump version and add changelog entry

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/9b39f2e6f63ae50cedd96eaf49146de8ed00fbc8">9b39f2e6</a></strong>

<div>

<span>by Ryan Scott</span>

<i>at 2020-04-01T01:20:00-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Clean up "Eta reduction for data families" Notes

Before, there were two distinct Notes named

"Eta reduction for data families". This renames one of them to

"Implementing eta reduction for data families" to disambiguate the

two and fixes references in other parts of the codebase to ensure

that they are pointing to the right place.

Fixes #17313.

[ci skip]

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/7627eab5dd882eb6f1567e3ae95c6c770830a5eb">7627eab5</a></strong>

<div>

<span>by Ryan Scott</span>

<i>at 2020-04-01T01:20:38-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix the changelog/@since information for hGetContents'/getContents'/readFile'

Fixes #17979.

[ci skip]

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/0002db1bf436cbd32f97b659a52b1eee4e8b21db">0002db1b</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-04-01T01:21:27-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Kill wORDS_BIGENDIAN and replace it with platformByteOrder (#17957)

Metric Decrease:

    T13035

    T1969

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/7b21717907a741b56513f5e1fa1ebceecf971613">7b217179</a></strong>

<div>

<span>by Sebastian Graf</span>

<i>at 2020-04-01T15:03:24-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">PmCheck: Adjust recursion depth for inhabitation test

In #17977, we ran into the reduction depth limit of the typechecker.

That was only a symptom of a much broader issue: The recursion depth

of the coverage checker for trying to instantiate strict fields in the

`nonVoid` test was far too high (100, the `defaultMaxTcBound`).

As a result, we were performing quite poorly on `T17977`.

Short of a proper termination analysis to prove emptyness of a type,

we just arbitrarily default to a much lower recursion limit of 3.

Fixes #17977.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/3c09f636a459f50119bfbb5bf798b9a9e19eb464">3c09f636</a></strong>

<div>

<span>by Andreas Klebinger</span>

<i>at 2020-04-01T15:03:59-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Make hadrian pass on the no-colour setting to GHC.

Fixes #17983.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/b943b25d0786da64031ac63ddf9b4574182057bb">b943b25d</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2020-04-02T01:45:58-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Re-engineer the binder-swap transformation

The binder-swap transformation is implemented by the occurrence

analyser -- see Note [Binder swap] in OccurAnal. However it had

a very nasty corner in it, for the case where the case scrutinee

was a GlobalId.  This led to trouble and hacks, and ultimately

to #16296.

This patch re-engineers how the occurrence analyser implements

the binder-swap, by actually carrying out a substitution rather

than by adding a let-binding.  It's all described in

Note [The binder-swap substitution].

I did a few other things along the way

* Fix a bug in StgCse, which could allow a loop breaker to be CSE'd

  away.  See Note [Care with loop breakers] in StgCse.  I think it can

  only show up if occurrence analyser sets up bad loop breakers, but

  still.

* Better commenting in SimplUtils.prepareAlts

* A little refactoring in CoreUnfold; nothing significant

  e.g. rename CoreUnfold.mkTopUnfolding to mkFinalUnfolding

* Renamed CoreSyn.isFragileUnfolding to hasCoreUnfolding

* Move mkRuleInfo to CoreFVs

We observed respectively 4.6% and 5.9% allocation decreases for the following

tests:

Metric Decrease:

    T9961

    haddock.base

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/42d68364f66846969edf029f878875c10cdfe0b2">42d68364</a></strong>

<div>

<span>by Sebastian Graf</span>

<i>at 2020-04-02T01:46:34-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Preserve precise exceptions in strictness analysis

Fix #13380 and #17676 by

1. Changing `raiseIO#` to have `topDiv` instead of `botDiv`

2. Give it special treatment in `Simplifier.Util.mkArgInfo`, treating it

   as if it still had `botDiv`, to recover dead code elimination.

This is the first commit of the plan outlined in

https://gitlab.haskell.org/ghc/ghc/-/merge_requests/2525#note_260886.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/0a88dd11594b8d8fd20500d026e657a5f99dfdd2">0a88dd11</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2020-04-02T01:47:25-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix a pointer format string in RTS

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/5beac042b995b055a66bc16be536d9e920f6864d">5beac042</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2020-04-02T01:48:05-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Remove unused closure stg_IND_direct

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/88f38b03025386f0f1e8f5861eed67d80495168a">88f38b03</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-04-02T01:48:42-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Session: Memoize stderrSupportsAnsiColors

Not only is this a reasonable efficiency measure but it avoids making

reentrant calls into ncurses, which is not thread-safe. See #17922.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/27740f24cb70fc14b00c1212c06642a144a6117d">27740f24</a></strong>

<div>

<span>by Ryan Scott</span>

<i>at 2020-04-02T01:49:21-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Make Hadrian build with Cabal-3.2

GHC 8.10 ships with `Cabal-3.2.0.0`, so it would be convenient to

make Hadrian supporting building against 3.2.* instead of having to

rebuild the entirety of `Cabal-3.0.0.0`. There is one API change in

`Cabal-3.2.*` that affects Hadrian: the `synopsis` and `description`

functions now return `ShortText` instead of `String`. Since Hadrian

manipulates these `String`s in various places, I found that the

simplest fix was to use CPP to convert `ShortText` to `String`s

where appropriate.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/4980200255dabf59ae537f10c55d19ef1a00bbdd">49802002</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-04-02T01:50:00-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Update Stack resolver for hadrian/build-stack

Broken by 57b888c0e90be7189285a6b078c30b26d0923809

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/30a63e79c65b023497af4fe2347149382c71829d">30a63e79</a></strong>

<div>

<span>by Ryan Scott</span>

<i>at 2020-04-02T01:50:36-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix two ASSERT buglets in reifyDataCon

Two `ASSERT`s in `reifyDataCon` were always using `arg_tys`, but

`arg_tys` is not meaningful for GADT constructors. In fact, it's

worse than non-meaningful, since using `arg_tys` when reifying a

GADT constructor can lead to failed `ASSERT`ions, as #17305

demonstrates.

This patch applies the simplest possible fix to the immediate

problem. The `ASSERT`s now use `r_arg_tys` instead of `arg_tys`, as

the former makes sure to give something meaningful for GADT

constructors. This makes the panic go away at the very least. There

is still an underlying issue with the way the internals of

`reifyDataCon` work, as described in

https://gitlab.haskell.org/ghc/ghc/issues/17305#note_227023, but we

leave that as future work, since fixing the underlying issue is

much trickier (see

https://gitlab.haskell.org/ghc/ghc/issues/17305#note_227087).

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/ef7576c40f8de391ed8b1c81c38156202e6d17cf">ef7576c4</a></strong>

<div>

<span>by Zubin Duggal</span>

<i>at 2020-04-03T06:24:56-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add outputable instances for the types in GHC.Iface.Ext.Types, add -ddump-hie

flag to dump pretty printed contents of the .hie file

Metric Increase:

   hie002

Because of the regression on i386:

compile_time/bytes allocated increased from i386-linux-deb9 baseline @ HEAD~10:

    Expected    hie002 (normal) compile_time/bytes allocated: 583014888.0 +/-10%

    Lower bound hie002 (normal) compile_time/bytes allocated:   524713399

    Upper bound hie002 (normal) compile_time/bytes allocated:   641316377

    Actual      hie002 (normal) compile_time/bytes allocated:   877986292

    Deviation   hie002 (normal) compile_time/bytes allocated:        50.6 %

*** unexpected stat test failure for hie002(normal)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/9462452a4843a2c42fe055a0a7e274d5164d1dc0">9462452a</a></strong>

<div>

<span>by Andreas Klebinger</span>

<i>at 2020-04-03T06:25:33-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Improve and refactor StgToCmm codegen for DataCons.

We now differentiate three cases of constructor bindings:

1)Bindings which we can "replace" with a reference to

  an existing closure. Reference the replacement closure

  when accessing the binding.

2)Bindings which we can "replace" as above. But we still

  generate a closure which will be referenced by modules

  importing this binding.

3)For any other binding generate a closure. Then reference

  it.

Before this patch 1) did only apply to local bindings and we

didn't do 2) at all.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/a214d2142c1bafa71fe52cb3823351ff9322d336">a214d214</a></strong>

<div>

<span>by Moritz Bruder</span>

<i>at 2020-04-03T06:26:11-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add singleton to NonEmpty in libraries/base

This adds a definition to construct a singleton non-empty list

(Data.List.NonEmpty) according to issue #17851.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/f7597aa0c028ced898ac97e344754dd961b70c57">f7597aa0</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-04-03T06:26:54-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Testsuite: measure compiler stats for T16190

We were mistakenly measuring program stats

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/a485c3c4049fff09e989bfd7d2ba47035c92a69b">a485c3c4</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-04-03T06:26:54-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Move blob handling into StgToCmm

Move handling of big literal strings from CmmToAsm to StgToCmm. It

avoids the use of `sdocWithDynFlags` (cf #10143). We might need to move

this handling even higher in the pipeline in the future (cf #17960):

this patch will make it easier.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/cc2918a0407e1581e824ebd90a1fcbb0637d5744">cc2918a0</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-04-03T06:26:54-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Refactor CmmStatics

In !2959 we noticed that there was some redundant code (in GHC.Cmm.Utils

and GHC.Cmm.StgToCmm.Utils) used to deal with `CmmStatics` datatype

(before SRT generation) and `RawCmmStatics` datatype (after SRT

generation).

This patch removes this redundant code by using a single GADT for

(Raw)CmmStatics.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/9e60273db47364fc08aeb5a389caf67559e0d353">9e60273d</a></strong>

<div>

<span>by Maxim Koltsov</span>

<i>at 2020-04-03T06:27:32-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix haddock formatting in Control.Monad.ST.Lazy.Imp.hs

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/1b7e8a94cb3334fc0e513dec2db323f32c3a0713">1b7e8a94</a></strong>

<div>

<span>by Andreas Klebinger</span>

<i>at 2020-04-03T06:28:08-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Turn newlines into spaces for hadrian/ghci.

The newlines break the command on windows.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/4291bddaea3148908c55f235ee8978e1d9aa6f20">4291bdda</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2020-04-03T06:28:44-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Major improvements to the specialiser

This patch is joint work of Alexis King and Simon PJ.  It does some

significant refactoring of the type-class specialiser.  Main highlights:

* We can specialise functions with types like

     f :: Eq a => a -> Ord b => b => blah

  where the classes aren't all at the front (#16473).  Here we can

  correctly specialise 'f' based on a call like

     f @Int @Bool dEqInt x dOrdBool

  This change really happened in an earlier patch

     commit 2d0cf6252957b8980d89481ecd0b79891da4b14b

     Author: Sandy Maguire <sandy@sandymaguire.me>

     Date:   Thu May 16 12:12:10 2019 -0400

  work that this new patch builds directly on that work, and refactors

  it a bit.

* We can specialise functions with implicit parameters (#17930)

     g :: (?foo :: Bool, Show a) => a -> String

  Previously we could not, but now they behave just like a non-class

  argument as in 'f' above.

* We can specialise under-saturated calls, where some (but not all of

  the dictionary arguments are provided (#17966).  For example, we can

  specialise the above 'f' based on a call

     map (f @Int dEqInt) xs

  even though we don't (and can't) give Ord dictionary.

  This may sound exotic, but #17966 is a program from the wild, and

  showed significant perf loss for functions like f, if you need

  saturation of all dictionaries.

* We fix a buglet in which a floated dictionary had a bogus demand

  (#17810), by using zapIdDemandInfo in the NonRec case of specBind.

* A tiny side benefit: we can drop dead arguments to specialised

  functions; see Note [Drop dead args from specialisations]

* Fixed a bug in deciding what dictionaries are "interesting"; see

  Note [Keep the old dictionaries interesting]

This is all achieved by by building on Sandy Macguire's work in

defining SpecArg, which mkCallUDs uses to describe the arguments of

the call. Main changes:

* Main work is in specHeader, which marched down the [InBndr] from the

  function definition and the [SpecArg] from the call site, together.

* specCalls no longer has an arity check; the entire mechanism now

  handles unders-saturated calls fine.

* mkCallUDs decides on an argument-by-argument basis whether to

  specialise a particular dictionary argument; this is new.

  See mk_spec_arg in mkCallUDs.

It looks as if there are many more lines of code, but I think that

all the extra lines are comments!

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/40a85563a46c682eaab5fdf970f7c46afca78cb3">40a85563</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2020-04-03T18:26:19+03:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Revert accidental change in 9462452

[ci skip]

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/bd75e5da0f1f05f107325733b570bf28b379d2f2">bd75e5da</a></strong>

<div>

<span>by Ryan Scott</span>

<i>at 2020-04-04T07:07:58-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Enable ImpredicativeTypes internally when typechecking selector bindings

This is necessary for certain record selectors with higher-rank

types, such as the examples in #18005. See

`Note [Impredicative record selectors]` in `TcTyDecls`.

Fixes #18005.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/dcfe29c8520244764146c7a5f336be1f9700db6c">dcfe29c8</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2020-04-06T13:16:08-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Don't override proc CafInfos in ticky builds

Fixes #17947

When we have a ticky label for a proc, IdLabels for the ticky counter

and proc entry share the same Name. This caused overriding proc CafInfos

with the ticky CafInfos (i.e. NoCafRefs) during SRT analysis.

We now ignore the ticky labels when building SRTMaps. This makes sense

because:

- When building the current module they don't need to be in SRTMaps as

  they're initialized as non-CAFFY (see mkRednCountsLabel), so they

  don't take part in the dependency analysis and they're never added to

  SRTs.

  (Reminder: a "dependency" in the SRT analysis is a CAFFY dependency,

  non-CAFFY uses are not considered as dependencies for the algorithm)

- They don't appear in the interfaces as they're not exported, so it

  doesn't matter for cross-module concerns whether they're in the SRTMap

  or not.

See also the new Note [Ticky labels in SRT analysis].

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/cec2c71fe91c88649628c6e83416533b816b86a5">cec2c71f</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2020-04-06T13:16:44-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix an tricky specialiser loop

Issue #17151 was a very tricky example of a bug in which the

specialiser accidentally constructs a recurive dictionary,

so that everything turns into bottom.

I have fixed variants of this bug at least twice before:

see Note [Avoiding loops].  It was a bit of a struggle

to isolate the problem, greatly aided by the work that

Alexey Kuleshevich did in distilling a test case.

Once I'd understood the problem, it was not difficult to fix,

though it did lead me a bit of refactoring in specImports.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/e850d14ffbeea39ad386b1e888cd97375758d6d6">e850d14f</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2020-04-06T13:16:44-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Refactoring only

This refactors DictBinds into a data type rather than a pair.

No change in behaviour, just better code

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/f38e8d61f066c3064c600c352eebcd87f28d989a">f38e8d61</a></strong>

<div>

<span>by Daniel Gröber</span>

<i>at 2020-04-07T02:00:05-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">rts: ProfHeap: Fix memory leak when not compiled with profiling

If we're doing heap profiling on an unprofiled executable we keep

allocating new space in initEra via nextEra on each profiler run but we

don't have a corresponding freeEra call.

We do free the last era in endHeapProfiling but previous eras will have

been overwritten by initEra and will never get free()ed.

Metric Decrease:

    space_leak_001

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/bcd668593e1c9e00c5d9c72960b4833dd526cb9a">bcd66859</a></strong>

<div>

<span>by Sebastian Graf</span>

<i>at 2020-04-07T02:00:41-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Re-export GHC.Magic.noinline from base

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/3d2991f8b4c1b686323b2c9452ce845a60b8d94c">3d2991f8</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-04-07T18:36:09-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">simplifier: Kill off ufKeenessFactor

We used to have another factor, ufKeenessFactor, which would scale the

discounts before they were subtracted from the size. This was justified

with the following comment:

  -- We multiple the raw discounts (args_discount and result_discount)

  -- ty opt_UnfoldingKeenessFactor because the former have to do with

  --  *size* whereas the discounts imply that there's some extra

  --  *efficiency* to be gained (e.g. beta reductions, case reductions)

  -- by inlining.

However, this is highly suspect since it means that we subtract a

*scaled* size from an absolute size, resulting in crazy (e.g. negative)

scores in some cases (#15304). We consequently killed off

ufKeenessFactor and bumped up the ufUseThreshold to compensate.

Adjustment of unfolding use threshold

=====================================

Since this removes a discount from our inlining heuristic, I revisited our

default choice of -funfolding-use-threshold to minimize the change in

overall inlining behavior. Specifically, I measured runtime allocations

and executable size of nofib and the testsuite performance tests built

using compilers (and core libraries) built with several values of

-funfolding-use-threshold.

This comes as a result of a quantitative comparison of testsuite

performance and code size as a function of ufUseThreshold, comparing

GHC trees using values of 50, 60, 70, 80, 90, and 100. The test set

consisted of nofib and the testsuite performance tests.

A full summary of these measurements are found in the description of

!2608

Comparing executable sizes (relative to the base commit) across all

nofib tests, we see that sizes are similar to the baseline:

            gmean      min      max   median

thresh

50         -6.36%   -7.04%   -4.82%   -6.46%

60         -5.04%   -5.97%   -3.83%   -5.11%

70         -2.90%   -3.84%   -2.31%   -2.92%

80         -0.75%   -2.16%   -0.42%   -0.73%

90         +0.24%   -0.41%   +0.55%   +0.26%

100        +1.36%   +0.80%   +1.64%   +1.37%

baseline   +0.00%   +0.00%   +0.00%   +0.00%

Likewise, looking at runtime allocations we see that 80 gives slightly

better optimisation than the baseline:

            gmean      min      max   median

thresh

50         +0.16%   -0.16%   +4.43%   +0.00%

60         +0.09%   -0.00%   +3.10%   +0.00%

70         +0.04%   -0.09%   +2.29%   +0.00%

80         +0.02%   -1.17%   +2.29%   +0.00%

90         -0.02%   -2.59%   +1.86%   +0.00%

100        +0.00%   -2.59%   +7.51%   -0.00%

baseline   +0.00%   +0.00%   +0.00%   +0.00%

Finally, I had to add a NOINLINE in T4306 to ensure that `upd` is

worker-wrappered as the test expects. This makes me wonder whether the

inlining heuristic is now too liberal as `upd` is quite a large

function. The same measure was taken in T12600.

             Wall clock time compiling Cabal with -O0

thresh       50     60     70     80     90      100    baseline

build-Cabal  93.88  89.58  92.59  90.09  100.26  94.81  89.13

Also, this change happens to avoid the spurious test output in

`plugin-recomp-change` and `plugin-recomp-change-prof` (see #17308).

Metric Decrease:

    hie002

    T12234

    T13035

    T13719

    T14683

    T4801

    T5631

    T5642

    T9020

    T9872d

    T9961

Metric Increase:

    T12150

    T12425

    T13701

    T14697

    T15426

    T1969

    T3064

    T5837

    T6048

    T9203

    T9872a

    T9872b

    T9872c

    T9872d

    haddock.Cabal

    haddock.base

    haddock.compiler

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/255418da5d264fb2758bc70925adb2094f34adc3">255418da</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-04-07T18:36:49-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Modules: type-checker (#13009)

Update Haddock submodule

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/04b6cf947ea065a210a216cc91f918cc1660d430">04b6cf94</a></strong>

<div>

<span>by Ryan Scott</span>

<i>at 2020-04-07T19:43:20-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Make NoExtCon fields strict

This changes every unused TTG extension constructor to be strict in

its field so that the pattern-match coverage checker is smart enough

any such constructors are unreachable in pattern matches. This lets

us remove nearly every use of `noExtCon` in the GHC API. The only

ones we cannot remove are ones underneath uses of `ghcPass`, but that

is only because GHC 8.8's and 8.10's coverage checkers weren't smart

enough to perform this kind of reasoning. GHC HEAD's coverage

checker, on the other hand, _is_ smart enough, so we guard these uses

of `noExtCon` with CPP for now.

Bumps the `haddock` submodule.

Fixes #17992.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/7802fa17a9a1a0f02fbf95170c13d7a9711a681e">7802fa17</a></strong>

<div>

<span>by Ryan Scott</span>

<i>at 2020-04-08T16:43:44-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Handle promoted data constructors in typeToLHsType correctly

Instead of using `nlHsTyVar`, which hardcodes `NotPromoted`, have

`typeToLHsType` pick between `Promoted` and `NotPromoted` by checking

if a type constructor is promoted using `isPromotedDataCon`.

Fixes #18020.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/ce481361fc95405cfadcd8f930629381e80e7f84">ce481361</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-04-09T16:17:21-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">hadrian: Use --export-dynamic when linking iserv

As noticed in #17962, the make build system currently does this (see

3ce0e0ba) but the change was never ported to Hadrian.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/fa66f143a61f2285618c611a27c23815ca588299">fa66f143</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-04-09T16:17:21-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">iserv: Don't pass --export-dynamic on FreeBSD

This is definitely a hack but it's probably the best we can do for now.

Hadrian does the right thing here by passing --export-dynamic only to

the linker.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/390751768104cd3d2cb57e2037062916476ebd10">39075176</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2020-04-09T16:18:00-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix CNF handling in compacting GC

Fixes #17937

Previously compacting GC simply ignored CNFs. This is mostly fine as

most (see "What about small compacts?" below) CNF objects don't have

outgoing pointers, and are "large" (allocated in large blocks) and large

objects are not moved or compacted.

However if we do GC *during* sharing-preserving compaction then the CNF

will have a hash table mapping objects that have been moved to the CNF

to their location in the CNF, to be able to preserve sharing.

This case is handled in the copying collector, in `scavenge_compact`,

where we evacuate hash table entries and then rehash the table.

Compacting GC ignored this case.

We now visit CNFs in all generations when threading pointers to the

compacted heap and thread hash table keys. A visited CNF is added to the

list `nfdata_chain`. After compaction is done, we re-visit the CNFs in

that list and rehash the tables.

The overhead is minimal: the list is static in `Compact.c`, and link

field is added to `StgCompactNFData` closure. Programs that don't use

CNFs should not be affected.

To test this CNF tests are now also run in a new way 'compacting_gc',

which just passes `-c` to the RTS, enabling compacting GC for the oldest

generation. Before this patch the result would be:

    Unexpected failures:

       compact_gc.run          compact_gc [bad exit code (139)] (compacting_gc)

       compact_huge_array.run  compact_huge_array [bad exit code (1)] (compacting_gc)

With this patch all tests pass. I can also pass `-c -DS` without any

failures.

What about small compacts? Small CNFs are still not handled by the

compacting GC. However so far I'm unable to write a test that triggers a

runtime panic ("update_fwd: unknown/strange object") by allocating a

small CNF in a compated heap. It's possible that I'm missing something

and it's not possible to have a small CNF.

NoFib Results:

--------------------------------------------------------------------------------

        Program           Size    Allocs    Instrs     Reads    Writes

--------------------------------------------------------------------------------

             CS          +0.1%      0.0%      0.0%     +0.0%     +0.0%

            CSD          +0.1%      0.0%      0.0%      0.0%      0.0%

             FS          +0.1%      0.0%      0.0%      0.0%      0.0%

              S          +0.1%      0.0%      0.0%      0.0%      0.0%

             VS          +0.1%      0.0%      0.0%      0.0%      0.0%

            VSD          +0.1%      0.0%     +0.0%     +0.0%     -0.0%

            VSM          +0.1%      0.0%     +0.0%     -0.0%      0.0%

           anna          +0.0%      0.0%     -0.0%     -0.0%     -0.0%

           ansi          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

           atom          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

         awards          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

         banner          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

     bernouilli          +0.1%      0.0%      0.0%     -0.0%     +0.0%

   binary-trees          +0.1%      0.0%     -0.0%     -0.0%      0.0%

          boyer          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

         boyer2          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

           bspt          +0.1%      0.0%     -0.0%     -0.0%     -0.0%

      cacheprof          +0.1%      0.0%     -0.0%     -0.0%     -0.0%

       calendar          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

       cichelli          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

        circsim          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

       clausify          +0.1%      0.0%     -0.0%     +0.0%     +0.0%

  comp_lab_zift          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

       compress          +0.1%      0.0%     +0.0%     +0.0%      0.0%

      compress2          +0.1%      0.0%     -0.0%      0.0%      0.0%

    constraints          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

   cryptarithm1          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

   cryptarithm2          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

            cse          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

   digits-of-e1          +0.1%      0.0%     +0.0%     -0.0%     -0.0%

   digits-of-e2          +0.1%      0.0%     -0.0%     -0.0%     -0.0%

         dom-lt          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

          eliza          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

          event          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

    exact-reals          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

         exp3_8          +0.1%      0.0%     +0.0%     -0.0%      0.0%

         expert          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

 fannkuch-redux          +0.1%      0.0%     -0.0%      0.0%      0.0%

          fasta          +0.1%      0.0%     -0.0%     +0.0%     +0.0%

            fem          +0.1%      0.0%     -0.0%     +0.0%      0.0%

            fft          +0.1%      0.0%     -0.0%     +0.0%     +0.0%

           fft2          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

       fibheaps          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

           fish          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

          fluid          +0.0%      0.0%     +0.0%     +0.0%     +0.0%

         fulsom          +0.1%      0.0%     -0.0%     +0.0%      0.0%

         gamteb          +0.1%      0.0%     +0.0%     +0.0%      0.0%

            gcd          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

    gen_regexps          +0.1%      0.0%     -0.0%     +0.0%      0.0%

         genfft          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

             gg          +0.1%      0.0%      0.0%     +0.0%     +0.0%

           grep          +0.1%      0.0%     -0.0%     +0.0%     +0.0%

         hidden          +0.1%      0.0%     +0.0%     -0.0%      0.0%

            hpg          +0.1%      0.0%     -0.0%     -0.0%     -0.0%

            ida          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

          infer          +0.1%      0.0%     +0.0%      0.0%     -0.0%

        integer          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

      integrate          +0.1%      0.0%     -0.0%     -0.0%     -0.0%

   k-nucleotide          +0.1%      0.0%     +0.0%     +0.0%      0.0%

          kahan          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

        knights          +0.1%      0.0%     -0.0%     -0.0%     -0.0%

         lambda          +0.1%      0.0%     +0.0%     +0.0%     -0.0%

     last-piece          +0.1%      0.0%     +0.0%      0.0%      0.0%

           lcss          +0.1%      0.0%     +0.0%     +0.0%      0.0%

           life          +0.1%      0.0%     -0.0%     +0.0%     +0.0%

           lift          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

         linear          +0.1%      0.0%     -0.0%     +0.0%      0.0%

      listcompr          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

       listcopy          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

       maillist          +0.1%      0.0%     +0.0%     -0.0%     -0.0%

         mandel          +0.1%      0.0%     +0.0%     +0.0%      0.0%

        mandel2          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

           mate          +0.1%      0.0%     +0.0%      0.0%     +0.0%

        minimax          +0.1%      0.0%     -0.0%      0.0%     -0.0%

        mkhprog          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

     multiplier          +0.1%      0.0%     +0.0%      0.0%      0.0%

         n-body          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

       nucleic2          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

           para          +0.1%      0.0%      0.0%     +0.0%     +0.0%

      paraffins          +0.1%      0.0%     +0.0%     -0.0%      0.0%

         parser          +0.1%      0.0%     -0.0%     -0.0%     -0.0%

        parstof          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

            pic          +0.1%      0.0%     -0.0%     -0.0%      0.0%

       pidigits          +0.1%      0.0%     +0.0%     -0.0%     -0.0%

          power          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

         pretty          +0.1%      0.0%     -0.0%     -0.0%     -0.1%

         primes          +0.1%      0.0%     -0.0%     -0.0%     -0.0%

      primetest          +0.1%      0.0%     -0.0%     -0.0%     -0.0%

         prolog          +0.1%      0.0%     -0.0%     -0.0%     -0.0%

         puzzle          +0.1%      0.0%     -0.0%     -0.0%     -0.0%

         queens          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

        reptile          +0.1%      0.0%     -0.0%     -0.0%     +0.0%

reverse-complem          +0.1%      0.0%     +0.0%      0.0%     -0.0%

        rewrite          +0.1%      0.0%     -0.0%     -0.0%     -0.0%

           rfib          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

            rsa          +0.1%      0.0%     -0.0%     +0.0%     -0.0%

            scc          +0.1%      0.0%     -0.0%     -0.0%     -0.1%

          sched          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

            scs          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

         simple          +0.1%      0.0%     -0.0%     -0.0%     -0.0%

          solid          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

        sorting          +0.1%      0.0%     -0.0%     -0.0%     -0.0%

  spectral-norm          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

         sphere          +0.1%      0.0%     -0.0%     -0.0%     -0.0%

         symalg          +0.1%      0.0%     -0.0%     -0.0%     -0.0%

            tak          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

      transform          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

       treejoin          +0.1%      0.0%     +0.0%     -0.0%     -0.0%

      typecheck          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

        veritas          +0.0%      0.0%     +0.0%     +0.0%     +0.0%

           wang          +0.1%      0.0%      0.0%     +0.0%     +0.0%

      wave4main          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

   wheel-sieve1          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

   wheel-sieve2          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

           x2n1          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

--------------------------------------------------------------------------------

            Min          +0.0%      0.0%     -0.0%     -0.0%     -0.1%

            Max          +0.1%      0.0%     +0.0%     +0.0%     +0.0%

 Geometric Mean          +0.1%     -0.0%     -0.0%     -0.0%     -0.0%

Bumping numbers of nonsensical perf tests:

Metric Increase:

    T12150

    T12234

    T12425

    T13035

    T5837

    T6048

It's simply not possible for this patch to increase allocations, and

I've wasted enough time on these test in the past (see #17686). I think

these tests should not be perf tests, but for now I'll bump the numbers.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/dce50062e35d3246b63fba9357dea6313c23c780">dce50062</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-04-09T16:18:44-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Rts: show errno on failure (#18033)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/045139f40089f288866c1c59c7379be82ecdaf34">045139f4</a></strong>

<div>

<span>by Hécate</span>

<i>at 2020-04-09T23:10:44-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add an example to liftIO and explain its purpose

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/101fab6ee6cee72b9ffce40e45ebf39466d1c01a">101fab6e</a></strong>

<div>

<span>by Sebastian Graf</span>

<i>at 2020-04-09T23:11:21-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Special case `isConstraintKindCon` on `AlgTyCon`

Previously, the `tyConUnique` record selector would unfold into a huge

case expression that would be inlined in all call sites, such as the

`INLINE`-annotated `coreView`, see #18026. `constraintKindTyConKey` only

occurs as the `Unique` of an `AlgTyCon` anyway, so we can make the code

a lot more compact, but have to move it to GHC.Core.TyCon.

Metric Decrease:

    T12150

    T12234

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/f5212dfc10414212e42247c2f2dcc45252f7e1d2">f5212dfc</a></strong>

<div>

<span>by Sebastian Graf</span>

<i>at 2020-04-09T23:11:57-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">DmdAnal: No need to attach a StrictSig to DataCon workers

In GHC.Types.Id.Make we were giving a strictness signature to every data

constructor wrapper Id that we weren't looking at in demand analysis

anyway. We used to use its CPR info, but that has its own CPR signature

now.

`Note [Data-con worker strictness]` then felt very out of place, so I

moved it to GHC.Core.DataCon.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/75a185dc2a648ab1f592d401daa5efcacb451c83">75a185dc</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-04-09T23:12:37-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Hadrian: fix --summary

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/723062edf6191084a99787d3f235183cf6b7d051">723062ed</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2020-04-10T09:18:14+03:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Move no_lint to the top level, tweak hie002

- We don't want to benchmark linting so disable lints in hie002 perf

  test

- Move no_lint to the top-level to be able to use it in tests other than

  those in `testsuite/tests/perf/compiler`.

- Filter out -dstg-lint in no_lint.

- hie002 allocation numbers on 32-bit are unstable, so skip it on 32-bit

Metric Decrease:

    hie002

    ManyConstructors

    T12150

    T12234

    T13035

    T1969

    T4801

    T9233

    T9961

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/bcafaa82a0223afd5d103e052ab9a097a676e5ea">bcafaa82</a></strong>

<div>

<span>by Peter Trommler</span>

<i>at 2020-04-10T19:29:33-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Testsuite: mark T11531 fragile

The test depends on a link editor allowing undefined symbols in an ELF

shared object. This is the standard but it seems some distributions

patch their link editor. See the report by @hsyl20 in #11531.

Fixes #11531

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/0889f5eecfea8af6a9d74d48d9d86ff3aea331d6">0889f5ee</a></strong>

<div>

<span>by Takenobu Tani</span>

<i>at 2020-04-12T11:44:52+09:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">testsuite: Fix comment for a language extension

[skip ci]

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/cd4f92b5f4251f1a37d1e08ee97d99f2ccb41f26">cd4f92b5</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2020-04-12T11:20:58-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Significant refactor of Lint

This refactoring of Lint was triggered by #17923, which is

fixed by this patch.

The main change is this.  Instead of

   lintType :: Type -> LintM LintedKind

we now have

   lintType :: Type -> LintM LintedType

Previously, all of typeKind was effectively duplicate in lintType.

Moreover, since we have an ambient substitution, we still had to

apply the substition here and there, sometimes more than once. It

was all very tricky, in the end, and made my head hurt.

Now, lintType returns a fully linted type, with all substitutions

performed on it.  This is much simpler.

The same thing is needed for Coercions.  Instead of

  lintCoercion :: OutCoercion

               -> LintM (LintedKind, LintedKind,

                         LintedType, LintedType, Role)

we now have

  lintCoercion :: Coercion -> LintM LintedCoercion

Much simpler!  The code is shorter and less bug-prone.

There are a lot of knock on effects.  But life is now better.

Metric Decrease:

    T1969

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/0efaf301fec9ed9ea827392cbe03de3335e995c7">0efaf301</a></strong>

<div>

<span>by Josh Meredith</span>

<i>at 2020-04-12T11:21:34-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Implement extensible interface files

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/54ca66a7d30d7f7cfbf3753ebe547f5a20d76b96">54ca66a7</a></strong>

<div>

<span>by Ryan Scott</span>

<i>at 2020-04-12T11:22:10-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Use conLikeUserTyVarBinders to quantify field selector types

This patch:

1. Writes up a specification for how the types of top-level field

   selectors should be determined in a new section of the GHC User's

   Guide, and

2. Makes GHC actually implement that specification by using

   `conLikeUserTyVarBinders` in `mkOneRecordSelector` to preserve the

   order and specificity of type variables written by the user.

Fixes #18023.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/35799dda07813e4c510237290a631d4d11fb92d2">35799dda</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-04-12T11:22:50-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">hadrian: Don't --export-dynamic on Darwin

When fixing #17962 I neglected to consider that --export-dynamic is only

supported on ELF platforms.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/e8029816fda7602a8163c4d2703ff02982a3e48c">e8029816</a></strong>

<div>

<span>by Alexis King</span>

<i>at 2020-04-12T11:23:27-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add an INLINE pragma to Control.Category.>>>

This fixes #18013 by adding INLINE pragmas to both Control.Category.>>>

and GHC.Desugar.>>>. The functional change in this patch is tiny (just

two lines of pragmas!), but an accompanying Note explains in gory

detail what’s going on.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/0da186c1b5a47e08e91c1c674d46c040c83932fc">0da186c1</a></strong>

<div>

<span>by Krzysztof Gogolewski</span>

<i>at 2020-04-14T07:55:20-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Change zipWith to zipWithEqual in a few places

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/074c1ccd3f8c3fcab117e336316173e8e869230a">074c1ccd</a></strong>

<div>

<span>by Andreas Klebinger</span>

<i>at 2020-04-14T07:55:55-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Small change to the windows ticker.

We already have a function to go from time to ms so use it.

Also expand on the state of timer resolution.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/b69cc8842aa7e2df52b92a9c9ad3b9d8dcf624ab">b69cc884</a></strong>

<div>

<span>by Alp Mestanogullari</span>

<i>at 2020-04-14T07:56:38-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">hadrian: get rid of unnecessary levels of nesting in source-dist

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/d0c3b0696f1ca809ebd83b5fc2c0b911cde38e77">d0c3b069</a></strong>

<div>

<span>by Julien Debon</span>

<i>at 2020-04-14T07:57:16-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">doc (Foldable): Add examples to Data.Foldable

See #17929

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/5b08e0c06e038448a63aa9bd7f163b23d824ba4b">5b08e0c0</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-04-14T23:28:20-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">StgCRun: Enable unwinding only on Linux

It's broken on macOS due and SmartOS due to assembler differences

(#15207) so let's be conservative in enabling it. Also, refactor things

to make the intent clearer.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/27cc2e7b1c1268e59c9d16b4530f27c0d40e9464">27cc2e7b</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-04-14T23:28:57-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">rts: Don't mark evacuate_large as inline

This function has two callsites and is quite large. GCC consequently

decides not to inline and warns instead. Given the situation, I can't

blame it. Let's just remove the inline specifier.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/9853fc5e3556e733b56976b0a2fce9e82130a9ef">9853fc5e</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-04-14T23:29:48-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">base: Enable large file support for OFD locking impl.

Not only is this a good idea in general but this should also avoid

issue #17950 by ensuring that off_t is 64-bits.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/7b41f21bbfa9e266ba6654b08c3f9fec549c8bca">7b41f21b</a></strong>

<div>

<span>by Matthew Pickering</span>

<i>at 2020-04-14T23:30:24-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Hadrian: Make -i paths absolute

The primary reason for this change is that ghcide does not work with

relative paths. It also matches what cabal and stack do, they always

pass absolute paths.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/41230e2601703df0233860be3f7d53f3a01bdbe5">41230e26</a></strong>

<div>

<span>by Daniel Gröber</span>

<i>at 2020-04-14T23:31:01-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Zero out pinned block alignment slop when profiling

The heap profiler currently cannot traverse pinned blocks because of

alignment slop. This used to just be a minor annoyance as the whole block

is accounted into a special cost center rather than the respective object's

CCS, cf. #7275. However for the new root profiler we would like to be able

to visit _every_ closure on the heap. We need to do this so we can get rid

of the current 'flip' bit hack in the heap traversal code.

Since info pointers are always non-zero we can in principle skip all the

slop in the profiler if we can rely on it being zeroed. This assumption

caused problems in the past though, commit a586b33f8e ("rts: Correct

handling of LARGE ARR_WORDS in LDV profiler"), part of !1118, tried to use

the same trick for BF_LARGE objects but neglected to take into account that

shrink*Array# functions don't ensure that slop is zeroed when not

compiling with profiling.

Later, commit 0c114c6599 ("Handle large ARR_WORDS in heap census (fix

as we will only be assuming slop is zeroed when profiling is on.

This commit also reduces the ammount of slop we introduce in the first

place by calculating the needed alignment before doing the allocation for

small objects where we know the next available address. For large objects

we don't know how much alignment we'll have to do yet since those details

are hidden behind the allocateMightFail function so there we continue to

allocate the maximum additional words we'll need to do the alignment.

So we don't have to duplicate all this logic in the cmm code we pull it

into the RTS allocatePinned function instead.

Metric Decrease:

    T7257

    haddock.Cabal

    haddock.base

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/15fa9bd6dd2d0b8d1fcd7135c85ea0d60853340d">15fa9bd6</a></strong>

<div>

<span>by Daniel Gröber</span>

<i>at 2020-04-14T23:31:01-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">rts: Expand and add more notes regarding slop

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/caf3f444bcc29f75145834207da00d938c08c2d3">caf3f444</a></strong>

<div>

<span>by Daniel Gröber</span>

<i>at 2020-04-14T23:31:01-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">rts: allocatePinned: Fix confusion about word/byte units

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/c3c0f662df06500a11970fd391d0a88e081a5296">c3c0f662</a></strong>

<div>

<span>by Daniel Gröber</span>

<i>at 2020-04-14T23:31:01-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">rts: Underline some Notes as is conventional

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/e149dea9bb89b77d34f50075946d6b4751a974f0">e149dea9</a></strong>

<div>

<span>by Daniel Gröber</span>

<i>at 2020-04-14T23:31:38-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">rts: Fix nomenclature in OVERWRITING_CLOSURE macros

The additional commentary introduced by commit 8916e64e5437 ("Implement

shrinkSmallMutableArray# and resizeSmallMutableArray#.") unfortunately got

this wrong. We set 'prim' to true in overwritingClosureOfs because we

_don't_ want to call LDV_recordDead().

The reason is because of this "inherently used" distinction made in the LDV

profiler so I rename the variable to be more appropriate.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/1dd3d18c2afd9e6009cd53295d26f8b31ca58fec">1dd3d18c</a></strong>

<div>

<span>by Daniel Gröber</span>

<i>at 2020-04-14T23:31:38-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Remove call to LDV_RECORD_CREATE for array resizing

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/19de2fb090a25ab0d640d0cd5aef09f35e7455a0">19de2fb0</a></strong>

<div>

<span>by Daniel Gröber</span>

<i>at 2020-04-14T23:31:38-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">rts: Assert LDV_recordDead is not called for inherently used closures

The comments make it clear LDV_recordDead should not be called for

inhererently used closures, so add an assertion to codify this fact.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/0b934e30417a767063625494ecf135c9d6006f71">0b934e30</a></strong>

<div>

<span>by Ryan Scott</span>

<i>at 2020-04-14T23:32:14-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Bump template-haskell version to 2.17.0.0

This requires bumping the `exceptions` and `text` submodules to bring

in commits that bump their respective upper version bounds on

`template-haskell`.

Fixes #17645. Fixes #17696.

Note that the new `text` commit includes a fair number of additions

to the Haddocks in that library. As a result, Haddock has to do more

work during the `haddock.Cabal` test case, increasing the number of

allocations it requires. Therefore,

-------------------------

Metric Increase:

    haddock.Cabal

-------------------------

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/22cc8e513fcfa89a4391f075534d903596a05895">22cc8e51</a></strong>

<div>

<span>by Ryan Scott</span>

<i>at 2020-04-15T17:48:47-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix #18052 by using pprPrefixOcc in more places

This fixes several small oversights in the choice of pretty-printing

function to use. Fixes #18052.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/ec77b2f16a78b13f54794c954953d8878dea9db2">ec77b2f1</a></strong>

<div>

<span>by Daniel Gröber</span>

<i>at 2020-04-15T17:49:24-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">rts: ProfHeap: Fix wrong time in last heap profile sample

We've had this longstanding issue in the heap profiler, where the time of

the last sample in the profile is sometimes way off causing the rendered

graph to be quite useless for long runs.

It seems to me the problem is that we use mut_user_time() for the last

sample as opposed to getRTSStats(), which we use when calling heapProfile()

in GC.c.

The former is equivalent to getProcessCPUTime() but the latter does

some additional stuff:

    getProcessCPUTime() - end_init_cpu - stats.gc_cpu_ns -

    stats.nonmoving_gc_cpu_ns

So to fix this just use getRTSStats() in both places.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/85fc32f03a6df92ec8d4ec9accca3c11b31a1596">85fc32f0</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-04-17T12:45:25-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Hadrian: fix dyn_o/dyn_hi rule (#17534)

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/bfde3b76ac7f5a72eca012fe34ac1340a5ce2011">bfde3b76</a></strong>

<div>

<span>by Ryan Scott</span>

<i>at 2020-04-17T12:46:02-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Fix #18065 by fixing an InstCo oversight in Core Lint

There was a small thinko in Core Lint's treatment of `InstCo`

coercions that ultimately led to #18065. The fix: add an apostrophe.

That's it!

Fixes #18065.

Co-authored-by: Simon Peyton Jones <simonpj@microsoft.com>

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/a05348ebaa11d563ab2e33325055317ff3cb8afc">a05348eb</a></strong>

<div>

<span>by Cale Gibbard</span>

<i>at 2020-04-17T13:08:47-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Change the fail operator argument of BindStmt to be a Maybe

Don't use noSyntaxExpr for it. There is no good way to defensively case

on that, nor is it clear one ought to do so.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/79e27144db7011f6d01a2f5ed15fd110d579bb8e">79e27144</a></strong>

<div>

<span>by John Ericson</span>

<i>at 2020-04-17T13:08:47-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Use trees that grow for rebindable operators for `<-` binds

Also add more documentation.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/18bc16ed78dfa1fe9498c5ac1ca38e9f84267872">18bc16ed</a></strong>

<div>

<span>by Cale Gibbard</span>

<i>at 2020-04-17T13:08:47-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Use FailOperator in more places, define a couple datatypes (XBindStmtRn and XBindStmtTc) to help clarify the meaning of XBindStmt in the renamer and typechecker

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/84cc8394d075cea236faa0bcd9ef0a84de89ee8c">84cc8394</a></strong>

<div>

<span>by Simon Peyton Jones</span>

<i>at 2020-04-18T13:20:29-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add a missing zonk in tcHsPartialType

I omitted a vital zonk when refactoring tcHsPartialType in

   commit 48fb3482f8cbc8a4b37161021e846105f980eed4

   Author: Simon Peyton Jones <simonpj@microsoft.com>

   Date:   Wed Jun 5 08:55:17 2019 +0100

   Fix typechecking of partial type signatures

This patch fixes it and adds commentary to explain why.

Fixes #18008

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/2ee96ac1aa2cbff4e70bc45988930d64599029f2">2ee96ac1</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-04-18T13:21:05-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Bump FreeBSD bootstrap compiler to 8.10.1

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/434312e5fdde91d6ad7b6d199bbb560a72ab6b89">434312e5</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-04-18T13:21:05-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Enable FreeBSD job for so-labelled MRs

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/ddffb2278a22b7c0e3a9fce11ada67a471243409">ddffb227</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-04-18T13:21:05-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">gitlab-ci: Use rules syntax for conditional jobs

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/e2586828a30202a28dcacac9ab7fc83c43da9e02">e2586828</a></strong>

<div>

<span>by Ben Gamari</span>

<i>at 2020-04-18T13:21:05-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Bump hsc2hs submodule

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/15ab6cd548f284732a7f89d78c2b89b1bfc4ea1d">15ab6cd5</a></strong>

<div>

<span>by Ömer Sinan Ağacan</span>

<i>at 2020-04-18T13:21:44-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Improve prepForeignCall error reporting

Show parameters and description of the error code when ffi_prep_cif

fails.

This may be helpful for debugging #17018.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/3ca52151881451ce5b3a7740d003e811b586140d">3ca52151</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-04-18T20:04:14+02:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">GHC.Core.Opt renaming

* GHC.Core.Op                  => GHC.Core.Opt

* GHC.Core.Opt.Simplify.Driver => GHC.Core.Opt.Driver

* GHC.Core.Opt.Tidy            => GHC.Core.Tidy

* GHC.Core.Opt.WorkWrap.Lib    => GHC.Core.Opt.WorkWrap.Utils

As discussed in:

 * https://mail.haskell.org/pipermail/ghc-devs/2020-April/018758.html

 * https://gitlab.haskell.org/ghc/ghc/issues/13009#note_264650

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/15312bbb53f247c9ed2c5cf75100a9f44c1c7227">15312bbb</a></strong>

<div>

<span>by Sylvain Henry</span>

<i>at 2020-04-18T20:04:46+02:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Modules (#13009)

* SysTools

* Parser

* GHC.Builtin

* GHC.Iface.Recomp

* Settings

Update Haddock submodule

Metric Decrease:

    Naperian

    parsing001

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/eaed0a3289e4c24ff1a70c6fc4b7f8bae6cd2dd3">eaed0a32</a></strong>

<div>

<span>by Alexis King</span>

<i>at 2020-04-19T03:16:44-04:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">Add missing addInScope call for letrec binders in OccurAnal

This fixes #18044, where a shadowed variable was incorrectly substituted

by the binder swap on the RHS of a floated-in letrec. This can only

happen when the uniques line up *just* right, so writing a regression

test would be very difficult, but at least the fix is small and

straightforward.

</pre>

</li>

<li>

<strong><a href="https://gitlab.haskell.org/ghc/ghc/-/commit/d0dd6f0e7301f15d0f3e0fcc83b2cb143c29a6a2">d0dd6f0e</a></strong>

<div>

<span>by Matthew Pickering</span>

<i>at 2020-04-19T16:01:09+01:00</i>

</div>

<pre class="commit-message" style="white-space: pre-wrap; margin: 0;">rts: Implement ghc-debug API

There are four components to this patch which make it possible to

implement `ghc-debug`.

1. Add four new functions to the RtsAPI.

  * rts_pause and rts_unpause allow an external process to completely

  pause and unpause the RTS.

  * rts_listThreads and rts_listMiscRoots are used to find the current

  roots of the garbage collector.

These changes also mean that `Task.h` is exposed to the user.

2. Generalise the `ghc-heap` API so that raw `Word`s can be returned

rather than actual objects. This is necessary when trying to decode

closures on an external process because the pointers in such closures

are correct for the internal rather than external process. If you used

the previous API then you would get a segfault as the garbage collector

would try to traverse into these nonsensical branches.

```

-- before

getClosureData :: a -> IO Closure

-- after

getClosureDataX :: (forall c . c -> IO (Ptr StgInfoTable, [Word], [b]))

                              -> a -> IO (GenClosure b)

```

For the normal case `b` is instantiated to `Box`, which contains a

pointer to a heap object.

```

data Box = Box a

-- GenClosure Box

```

For `ghc-debug` we instead just take the word of the address as we have

to explicitly interpret it on the external process.

```

GenClosure Word

```

3. Support for decoding `TSO` and `STACK` closures is partially

implemented. There is still quite a bit of work to do to finish both but

these at least allow us to make some more progress.

4. findPtr is generalised to take a callback argument. This means that

its result can be communicated to the debugger rather than just printing

out the result. The debugger has a function which invokes `findPtr` and

passes a callback which sends the result over a socket.

Co-authored-by: Ben Gamari <ben@smart-cactus.org>

</pre>

</li>

</ul>

<h4>19 changed files:</h4>

<ul>

<li class="file-stats">

<a href="#03615ac1622815b67e9829de1c43ac90fa6d8068">

.ghcid

</a>

</li>

<li class="file-stats">

<a href="#587d266bb27a4dc3022bbed44dfa19849df3044c">

.gitlab-ci.yml

</a>

</li>

<li class="file-stats">

<a href="#157f7634c25bc6366cb7c9c9edb48e819dce38db">

<span class="new-file">

+

.gitlab/ci.sh

</span>

</a>

</li>

<li class="file-stats">

<a href="#0b4c5784bb7e87b68e9cfeb0ef799d273c982a59">

<span class="deleted-file">

−

.gitlab/darwin-init.sh

</span>

</a>

</li>

<li class="file-stats">

<a href="#5b24e4ac8e89486e4e26b14d758dd7b7af21440b">

<span class="new-file">

+

.gitlab/linters/check-changelogs.sh

</span>

</a>

</li>

<li class="file-stats">

<a href="#412267f7d5ed866deae5ac2ca598ff8092b826b6">

.gitlab/linters/check-cpp.py

</a>

</li>

<li class="file-stats">

<a href="#b25b374c2459de3e2aaed42163f492c8b8614385">

.gitlab/merge_request_templates/merge-request.md

</a>

</li>

<li class="file-stats">

<a href="#1726cdd2367b3b0246025a33d56ad19b52d3e5d4">

<span class="deleted-file">

−

.gitlab/prepare-system.sh

</span>

</a>

</li>

<li class="file-stats">

<a href="#f01d7155b7d732755fc5205809746325be866225">

<span class="deleted-file">

−

.gitlab/push-test-metrics.sh

</span>

</a>

</li>

<li class="file-stats">

<a href="#f656107cc03a27946448887125380c358788500e">

<span class="new-file">

+

.gitlab/test-metrics.sh

</span>

</a>

</li>

<li class="file-stats">

<a href="#0641ef536c197a46fddf169b39613b0ff5ad92b3">

<span class="deleted-file">

−

.gitlab/win32-init.sh

</span>

</a>

</li>

<li class="file-stats">

<a href="#7445606fbf8f3683cd42bdc54b05d7a0bc2dfc44">

.gitmodules

</a>

</li>

<li class="file-stats">

<a href="#abe895765c4ce08c5dfbc95e5e3c5db67ff92ded">

CODEOWNERS

</a>

</li>

<li class="file-stats">

<a href="#0c39c6a7c54e5d668d208665d6e5ccecdd3bb246">

HACKING.md

</a>

</li>

<li class="file-stats">

<a href="#9ab3868b23ed5d5a6e12ef902049902556fa4009">

aclocal.m4

</a>

</li>

<li class="file-stats">

<a href="#5c73b0c6f476ded38de389f894770f06f4d02b2f">

boot

</a>

</li>

<li class="file-stats">

<a href="#d0d96a6d03668aeab20ebe05e2c4ccb798c7e64c">

<span class="new-file">

+

compiler/GHC.hs

</span>

</a>

</li>

<li class="file-stats">

<a href="#0887cf39c5cdf9cf8d6758f410d7dab3023c0d77">

<span class="new-file">

+

compiler/GHC/Builtin/Names.hs

</span>

</a>

</li>

<li class="file-stats">

<a href="#57f535740df5928fb0193abc3b28ca771369d8fd">

<span class="new-file">

+

compiler/GHC/Builtin/Names.hs-boot

</span>

</a>

</li>

</ul>

<h5>The diff was not included because it is too large.</h5>

</div>

<div class="footer" style="margin-top: 10px;">

<p style="font-size: small; color: #777;">

—

<br>

<a href="https://gitlab.haskell.org/ghc/ghc/-/compare/c5f8d4741503a043ddfe5fefeac5e984a0f3062f...d0dd6f0e7301f15d0f3e0fcc83b2cb143c29a6a2">View it on GitLab</a>.

<br>

You're receiving this email because of your account on gitlab.haskell.org.

If you'd like to receive fewer emails, you can

adjust your notification settings.

</p>

</div>

</body>

</html>