[Git][ghc/ghc][wip/T20666] 25 commits: nonmoving: Fix race in marking of blackholes

Simon Peyton Jones (@simonpj) gitlab at gitlab.haskell.org
Tue Jan 3 14:56:19 UTC 2023



Simon Peyton Jones pushed to branch wip/T20666 at Glasgow Haskell Compiler / GHC


Commits:
18d2acd2 by Ben Gamari at 2022-12-23T19:09:30-05:00
nonmoving: Fix race in marking of blackholes

We must use an acquire-fence when marking to ensure that the indirectee
is visible.

- - - - -
11241efa by Ben Gamari at 2022-12-23T19:09:30-05:00
nonmoving: Fix segment list races

- - - - -
602455c9 by Ben Gamari at 2022-12-23T19:09:30-05:00
nonmoving: Use atomic when looking at bd->gen

Since it may have been mutated by a moving GC.

- - - - -
9d63b160 by Ben Gamari at 2022-12-23T19:09:30-05:00
nonmoving: Eliminate race in bump_static_flag

To ensure that we don't race with a mutator entering a new CAF we take
the SM mutex before touching static_flag. The other option here would be
to instead modify newCAF to use a CAS but the present approach is a bit
safer.

- - - - -
26837523 by Ben Gamari at 2022-12-23T19:09:30-05:00
nonmoving: Ensure that mutable fields have acquire barrier

- - - - -
8093264a by Ben Gamari at 2022-12-23T19:09:30-05:00
nonmoving: Fix races in collector status tracking

Mark a number of accesses to do with tracking of the status of the
concurrent collection thread as atomic. No interesting races here,
merely necessary to satisfy TSAN.

- - - - -
387d4fcc by Ben Gamari at 2022-12-23T19:09:30-05:00
nonmoving: Make segment state updates atomic

- - - - -
543cae00 by Ben Gamari at 2022-12-23T19:09:30-05:00
nonmoving: Refactor update remembered set initialization

This avoids a lock inversion between the storage manager mutex and
the stable pointer table mutex by not dropping the SM_MUTEX in
nonmovingCollect. This requires quite a bit of rejiggering but it
does seem like a better strategy.

- - - - -
c9936718 by Ben Gamari at 2022-12-23T19:09:30-05:00
nonmoving: Ensure that we aren't holding locks when closing them

TSAN complains about this sort of thing.

- - - - -
0cd31f7d by Ben Gamari at 2022-12-23T19:09:30-05:00
nonmoving: Make bitmap accesses atomic

This is a benign race on any sensible hard since these are byte
accesses. Nevertheless, atomic accesses are necessary to satisfy
TSAN.

- - - - -
d3fe110a by Ben Gamari at 2022-12-23T19:09:30-05:00
nonmoving: Fix benign race in update remembered set check

Relaxed load is fine here since we will take the lock before looking at
the list.

- - - - -
ab6cf893 by Ben Gamari at 2022-12-23T19:09:30-05:00
nonmoving: Fix race in shortcutting

We must use an acquire load to read the info table pointer since if we
find an indirection we must be certain that we see the indirectee.

- - - - -
36c9f23c by Ben Gamari at 2022-12-23T19:09:30-05:00
nonmoving: Make free list counter accesses atomic

Since these may race with the allocator(s).

- - - - -
aebef31c by doyougnu at 2022-12-23T19:10:09-05:00
add GHC.Utils.Binary.foldGet' and use for Iface

A minor optimization to remove lazy IO and a lazy accumulator

strictify foldGet'

IFace.Binary: use strict foldGet'

remove superfluous bang

- - - - -
5eb357d9 by Ben Gamari at 2022-12-24T00:41:05-05:00
compiler: Ensure that GHC toolchain is first in search path

As noted in #22561, it is important that GHC's toolchain look
first for its own headers and libraries to ensure that the
system's are not found instead. If this happens things can
break in surprising ways (e.g. see #22561).

- - - - -
cbaebfb9 by Matthew Pickering at 2022-12-24T00:41:40-05:00
head.hackage: Use slow-validate bindist for linting jobs

This enables the SLOW_VALIDATE env var for the linting head.hackage
jobs, namely the jobs enabled manually, by the label or on the nightly
build now use the deb10-numa-slow-validate bindist which has assertions
enabled.

See #22623 for a ticket which was found by using this configuration
already!

The head.hackage jobs triggered by upstream CI are now thusly:

hackage-lint: Can be triggered on any MR, normal validate pipeline or nightly build.
              Runs head.hackage with -dlint and a slow-validate bindist

hackage-label-lint: Trigged on MRs with "user-facing" label, runs the slow-validate
                    head.hackage build with -dlint.

nightly-hackage-lint: Runs automatically on nightly pipelines with slow-validate + dlint config.

nightly-hackage-perf: Runs automaticaly on nightly pipelines with release build and eventlogging enabled.

release-hackage-lint: Runs automatically on release pipelines with -dlint on a release bindist.

- - - - -
f4850f36 by Matthew Pickering at 2022-12-24T00:41:40-05:00
ci: Don't run abi-test-nightly on release jobs

The test is not configured to get the correct dependencies for the
release pipelines (and indeed stops the release pipeline being run at
all)

- - - - -
c264b06b by Matthew Pickering at 2022-12-24T00:41:40-05:00
ci: Run head.hackage jobs on upstream-testing branch rather than master

This change allows less priviledged users to trigger head.hackage jobs
because less permissions are needed to trigger jobs on the
upstream-testing branch, which is not protected.

There is a CI job which updates upstream-testing each hour to the state
of the master branch so it should always be relatively up-to-date.

- - - - -
63b97430 by Ben Gamari at 2022-12-24T00:42:16-05:00
llvmGen: Fix relaxed ordering

Previously I used LLVM's `unordered` ordering for the C11 `relaxed`
ordering. However, this is wrong and should rather use the LLVM
`monotonic` ordering.

Fixes #22640
- - - - -
f42ba88f by Ben Gamari at 2022-12-24T00:42:16-05:00
gitlab-ci: Introduce aarch64-linux-llvm job

This nightly job will ensure that we don't break the LLVM backend on
AArch64/Linux by bootstrapping GHC.

This would have caught #22640.

- - - - -
6d62f6bf by Matthew Pickering at 2022-12-24T00:42:51-05:00
Store RdrName rather than OccName in Holes

In #20472 it was pointed out that you couldn't defer out of scope but
the implementation collapsed a RdrName into an OccName to stuff it into
a Hole. This leads to the error message for a deferred qualified name
dropping the qualification which affects the quality of the error
message.

This commit adds a bit more structure to a hole, so a hole can replace a
RdrName without losing information about what that RdrName was. This is
important when printing error messages.

I also added a test which checks the Template Haskell deferral of out of
scope qualified names works properly.

Fixes #22130

- - - - -
3c3060e4 by Richard Eisenberg at 2022-12-24T17:34:19+00:00
Drop support for kind constraints.

This implements proposal 547 and closes ticket #22298.
See the proposal and ticket for motivation.

Compiler perf improves a bit

Metrics: compile_time/bytes allocated
-------------------------------------
  CoOpt_Singletons(normal)   -2.4% GOOD
            T12545(normal)   +1.0%
            T13035(normal)  -13.5% GOOD
            T18478(normal)   +0.9%
            T9872d(normal)   -2.2% GOOD

                 geo. mean   -0.2%
                 minimum    -13.5%
                 maximum     +1.0%

Metric Decrease:
    CoOpt_Singletons
    T13035
    T9872d

- - - - -
6d7d4393 by Ben Gamari at 2022-12-24T21:09:56-05:00
hadrian: Ensure that linker scripts are used when merging objects

In #22527 @rui314 inadvertantly pointed out a glaring bug in Hadrian's
implementation of the object merging rules: unlike the old `make` build
system we utterly failed to pass the needed linker scripts. Fix this.

- - - - -
a5bd0eb8 by Bodigrim at 2022-12-24T21:10:34-05:00
Document infelicities of instance Ord Double and workarounds

- - - - -
44688140 by Richard Eisenberg at 2023-01-03T12:55:19+00:00
Refactor the treatment of loopy superclass dicts

This patch completely re-engineers how we deal with loopy superclass
dictionaries in instance declarations. It fixes #20666.

The highlights are

* Recognise that the loopy-superclass business should use precisely
  the Paterson conditions.  This is much much nicer.  See
  Note [Recursive superclasses] in GHC.Tc.TyCl.Instance

* With that in mind, define "Paterson-smaller" in
  Note [Paterson conditions] in GHC.Tc.Validity, and the new
  data type `PatersonSize` in GHC.Tc.Utils.TcType, along with
  functions to compute and compare PatsonSizes

* Use the new PatersonSize stuff when solving superclass constraints
  See Note [Solving superclass constraints] in GHC.Tc.TyCl.Instance

* In GHC.Tc.Solver.Monad.lookupInInerts, add a missing call to
  prohibitedSuperClassSolve.  This was the original cause of #20666.

* Treat (TypeError "stuff") as having PatersonSize zero. See
  Note [Paterson size for type family applications] in GHC.Tc.Utils.TcType.

- - - - -


30 changed files:

- .gitlab-ci.yml
- .gitlab/gen_ci.hs
- .gitlab/jobs.yaml
- compiler/GHC/Builtin/Names/TH.hs
- compiler/GHC/CmmToLlvm/CodeGen.hs
- compiler/GHC/Core/DataCon.hs
- compiler/GHC/Core/TyCo/FVs.hs
- compiler/GHC/Core/TyCo/Rep.hs
- compiler/GHC/Core/TyCon.hs
- compiler/GHC/Core/Type.hs
- compiler/GHC/CoreToIface.hs
- compiler/GHC/HsToCore/Quote.hs
- compiler/GHC/Iface/Binary.hs
- compiler/GHC/Iface/Type.hs
- compiler/GHC/Parser/PostProcess.hs
- compiler/GHC/Rename/Expr.hs
- compiler/GHC/Rename/HsType.hs
- compiler/GHC/Rename/Module.hs
- compiler/GHC/Settings/IO.hs
- compiler/GHC/Tc/Deriv.hs
- compiler/GHC/Tc/Deriv/Infer.hs
- compiler/GHC/Tc/Errors.hs
- compiler/GHC/Tc/Errors/Ppr.hs
- compiler/GHC/Tc/Errors/Types.hs
- compiler/GHC/Tc/Gen/HsType.hs
- compiler/GHC/Tc/Solver.hs
- compiler/GHC/Tc/Solver/Canonical.hs
- compiler/GHC/Tc/Solver/InertSet.hs
- compiler/GHC/Tc/Solver/Interact.hs
- compiler/GHC/Tc/Solver/Monad.hs


The diff was not included because it is too large.


View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/a341d0e15a83dcd45d92398df48dd0051cb6b0a1...4468814072e208c7b86d6cd73e77f339be3c19e1

-- 
View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/a341d0e15a83dcd45d92398df48dd0051cb6b0a1...4468814072e208c7b86d6cd73e77f339be3c19e1
You're receiving this email because of your account on gitlab.haskell.org.


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.haskell.org/pipermail/ghc-commits/attachments/20230103/3f8bdf2f/attachment-0001.html>


More information about the ghc-commits mailing list