[Git][ghc/ghc][wip/dmdanal-precise-exn-simpl] 42 commits: Refactor hole constraints.
Sebastian Graf
gitlab at gitlab.haskell.org
Wed May 13 15:04:25 UTC 2020
Sebastian Graf pushed to branch wip/dmdanal-precise-exn-simpl at Glasgow Haskell Compiler / GHC
Commits:
7ab6ab09 by Richard Eisenberg at 2020-05-06T04:39:32-04:00
Refactor hole constraints.
Previously, holes (both expression holes / out of scope variables and
partial-type-signature wildcards) were emitted as *constraints* via
the CHoleCan constructor. While this worked fine for error reporting,
there was a fair amount of faff in keeping these constraints in line.
In particular, and unlike other constraints, we could never change
a CHoleCan to become CNonCanonical. In addition:
* the "predicate" of a CHoleCan constraint was really the type
of the hole, which is not a predicate at all
* type-level holes (partial type signature wildcards) carried
evidence, which was never used
* tcNormalise (used in the pattern-match checker) had to create
a hole constraint just to extract it again; it was quite messy
The new approach is to record holes directly in WantedConstraints.
It flows much more nicely now.
Along the way, I did some cleaning up of commentary in
GHC.Tc.Errors.Hole, which I had a hard time understanding.
This was instigated by a future patch that will refactor
the way predicates are handled. The fact that CHoleCan's
"predicate" wasn't really a predicate is incompatible with
that future patch.
No test case, because this is meant to be purely internal.
It turns out that this change improves the performance of
the pattern-match checker, likely because fewer constraints
are sloshing about in tcNormalise. I have not investigated
deeply, but an improvement is not a surprise here:
-------------------------
Metric Decrease:
PmSeriesG
-------------------------
- - - - -
420b957d by Ben Gamari at 2020-05-06T04:40:08-04:00
rts: Zero block flags with -DZ
Block flags are very useful for determining the state of a block.
However, some block allocator users don't touch them, leading to
misleading values. Ensure that we zero then when zero-on-gc is set. This
is safe and makes the flags more useful during debugging.
- - - - -
740b3b8d by Ben Gamari at 2020-05-06T04:40:08-04:00
nonmoving: Fix incorrect failed_to_evac value during deadlock gc
Previously we would incorrectly set the failed_to_evac flag if we
evacuated a value due to a deadlock GC. This would cause us to mark more
things as dirty than strictly necessary. It also turned up a nasty but
which I will fix next.
- - - - -
b2d72c75 by Ben Gamari at 2020-05-06T04:40:08-04:00
nonmoving: Fix handling of dirty objects
Previously we (incorrectly) relied on failed_to_evac to be "precise".
That is, we expected it to only be true if *all* of an object's fields
lived outside of the non-moving heap. However, does not match the
behavior of failed_to_evac, which is true if *any* of the object's
fields weren't promoted (meaning that some others *may* live in the
non-moving heap).
This is problematic as we skip the non-moving write barrier for dirty
objects (which we can only safely do if *all* fields point outside of
the non-moving heap).
Clearly this arises due to a fundamental difference in the behavior
expected of failed_to_evac in the moving and non-moving collector.
e.g., in the moving collector it is always safe to conservatively say
failed_to_evac=true whereas in the non-moving collector the safe value
is false.
This issue went unnoticed as I never wrote down the dirtiness
invariant enforced by the non-moving collector. We now define this
invariant as
An object being marked as dirty implies that all of its fields are
on the mark queue (or, equivalently, update remembered set).
To maintain this invariant we teach nonmovingScavengeOne to push the
fields of objects which we fail to evacuate to the update remembered
set. This is a simple and reasonably cheap solution and avoids the
complexity and fragility that other, more strict alternative invariants
would require.
All of this is described in a new Note, Note [Dirty flags in the
non-moving collector] in NonMoving.c.
- - - - -
9f3e6884 by Zubin Duggal at 2020-05-06T04:41:08-04:00
Allow atomic update of NameCache in readHieFile
The situation arises in ghcide where multiple different threads may need to
update the name cache, therefore with the older interface it could happen
that you start reading a hie file with name cache A and produce name cache
A + B, but another thread in the meantime updated the namecache to A +
C. Therefore if you write the new namecache you will lose the A' updates
from the second thread.
Updates haddock submodule
- - - - -
edec6a6c by Ryan Scott at 2020-05-06T04:41:57-04:00
Make isTauTy detect higher-rank contexts
Previously, `isTauTy` would only detect higher-rank `forall`s, not
higher-rank contexts, which led to some minor bugs observed
in #18127. Easily fixed by adding a case for
`(FunTy InvisArg _ _)`.
Fixes #18127.
- - - - -
a95e7fe0 by Ömer Sinan Ağacan at 2020-05-06T04:42:39-04:00
ELF linker: increment curSymbol after filling in fields of current entry
The bug was introduced in a8b7cef4d45 which added a field to the
`symbols` array elements and then updated this code incorrectly:
- oc->symbols[curSymbol++] = nm;
+ oc->symbols[curSymbol++].name = nm;
+ oc->symbols[curSymbol].addr = symbol->addr;
- - - - -
cab1871a by Sylvain Henry at 2020-05-06T04:43:21-04:00
Move LeadingUnderscore into Platform (#17957)
Avoid direct use of DynFlags to know if symbols must be prefixed by an
underscore.
- - - - -
94e7c563 by Sylvain Henry at 2020-05-06T04:43:21-04:00
Don't use DynFlags in showLinkerState (#17957)
- - - - -
9afd9251 by Ryan Scott at 2020-05-06T04:43:58-04:00
Refactoring: Use bindSigTyVarsFV in rnMethodBinds
`rnMethodBinds` was explicitly using `xoptM` to determine if
`ScopedTypeVariables` is enabled before bringing type variables
bound by the class/instance header into scope. However, this `xoptM`
logic is already performed by the `bindSigTyVarsFV` function. This
patch uses `bindSigTyVarsFV` in `rnMethodBinds` to reduce the number
of places where we need to consult if `ScopedTypeVariables` is on.
This is purely refactoring, and there should be no user-visible
change in behavior.
- - - - -
6f6d72b2 by Brian Foley at 2020-05-08T15:29:25-04:00
Remove further dead code found by a simple Python script.
Avoid removing some functions that are part of an API even
though they're not used in-tree at the moment.
- - - - -
78bf8bf9 by Julien Debon at 2020-05-08T15:29:28-04:00
Add doc examples for Bifoldable
See #17929
- - - - -
66f0a847 by Julien Debon at 2020-05-08T15:29:29-04:00
doc (Bitraversable): Add examples to Bitraversable
* Add examples to Data.Bitraversable
* Fix formatting for (,) in Bitraversable and Bifoldable
* Fix mistake on bimapAccumR documentation
See #17929
- - - - -
9749fe12 by Baldur Blöndal at 2020-05-08T15:29:32-04:00
Specify kind variables for inferred kinds in base.
- - - - -
4e9aef9e by John Ericson at 2020-05-08T15:29:36-04:00
HsSigWcTypeScoping: Pull in documentation from stray location
- - - - -
f4d5c6df by John Ericson at 2020-05-08T15:29:36-04:00
Rename local `real_fvs` to `implicit_vs`
It doesn't make sense to call the "free" variables we are about to
implicitly bind the real ones.
- - - - -
20570b4b by John Ericson at 2020-05-08T15:29:36-04:00
A few tiny style nits with renaming
- Use case rather than guards that repeatedly scrutenize same thing.
- No need for view pattern when `L` is fine.
- Use type synnonym to convey the intent like elsewhere.
- - - - -
09ac8de5 by John Ericson at 2020-05-08T15:29:36-04:00
Add `forAllOrNothing` function with note
- - - - -
bb35c0e5 by Joseph C. Sible at 2020-05-08T15:29:40-04:00
Document lawlessness of Ap's Num instance
- - - - -
cdd229ff by Joseph C. Sible at 2020-05-08T15:29:40-04:00
Apply suggestion to libraries/base/Data/Monoid.hs
- - - - -
926d2aab by Joseph C. Sible at 2020-05-08T15:29:40-04:00
Apply more suggestions from Simon Jakobi
- - - - -
7a763cff by Adam Gundry at 2020-05-08T15:29:41-04:00
Reject all duplicate declarations involving DuplicateRecordFields (fixes #17965)
This fixes a bug that resulted in some programs being accepted that used the same
identifier as a field label and another declaration, depending on the order they
appeared in the source code.
- - - - -
88e3c815 by Simon Peyton Jones at 2020-05-08T15:29:41-04:00
Fix specialisation for DFuns
When specialising a DFun we must take care to saturate the
unfolding. See Note [Specialising DFuns] in Specialise.
Fixes #18120
- - - - -
86c77b36 by Greg Steuck at 2020-05-08T15:29:45-04:00
Remove unused SEGMENT_PROT_RWX
It's been unused for a year and is problematic on any OS which
requires W^X for security.
- - - - -
9d97f4b5 by nineonine at 2020-05-08T15:30:03-04:00
Add test for #16167
- - - - -
aa318338 by Ryan Scott at 2020-05-08T15:30:04-04:00
Bump exceptions submodule so that dist-boot is .gitignore'd
`exceptions` is a stage-0 boot library as of commit
30272412fa437ab8e7a8035db94a278e10513413, which means that building
`exceptions` in a GHC tree will generate a `dist-boot` directory.
However, this directory was not specified in `exceptions`'
`.gitignore` file, which causes it to dirty up the current `git`
working directory.
Accordingly, this bumps the `exceptions` submodule to commit
ghc/packages/exceptions at 23c0b8a50d7592af37ca09beeec16b93080df98f,
which adds `dist-boot` to the `.gitignore` file.
- - - - -
ea86360f by Ömer Sinan Ağacan at 2020-05-08T15:30:30-04:00
Linker.c: initialize n_symbols of ObjectCode with other fields
- - - - -
951c1fb0 by Sylvain Henry at 2020-05-09T21:46:38-04:00
Fix unboxed-sums GC ptr-slot rubbish value (#17791)
This patch allows boot libraries to use unboxed sums without implicitly
depending on `base` package because of `absentSumFieldError`.
See updated Note [aBSENT_SUM_FIELD_ERROR_ID] in GHC.Core.Make
- - - - -
b352d63c by Ben Gamari at 2020-05-09T21:47:14-04:00
rts: Make non-existent linker search path merely a warning
As noted in #18105, previously this resulted in a rather intrusive error
message. This is in contrast to the general expectation that search
paths are merely places to look, not places that must exist.
Fixes #18105.
- - - - -
cf4f1e2f by Ben Gamari at 2020-05-13T02:02:33-04:00
rts/CNF: Fix fixup comparison function
Previously we would implicitly convert the difference between two words
to an int, resulting in an integer overflow on 64-bit machines.
Fixes #16992
- - - - -
a03da9bf by Ömer Sinan Ağacan at 2020-05-13T02:03:16-04:00
Pack some of IdInfo fields into a bit field
This reduces residency of compiler quite a bit on some programs.
Example stats when building T10370:
Before:
2,871,242,832 bytes allocated in the heap
4,693,328,008 bytes copied during GC
33,941,448 bytes maximum residency (276 sample(s))
375,976 bytes maximum slop
83 MiB total memory in use (0 MB lost due to fragmentation)
After:
2,858,897,344 bytes allocated in the heap
4,629,255,440 bytes copied during GC
32,616,624 bytes maximum residency (278 sample(s))
314,400 bytes maximum slop
80 MiB total memory in use (0 MB lost due to fragmentation)
So -3.9% residency, -1.3% bytes copied and -0.4% allocations.
Fixes #17497
Metric Decrease:
T9233
T9675
- - - - -
670c3e5c by Ben Gamari at 2020-05-13T02:03:54-04:00
get-win32-tarballs: Fix base URL
Revert a change previously made for testing purposes.
- - - - -
8ad8dc41 by Ben Gamari at 2020-05-13T02:03:54-04:00
get-win32-tarballs: Improve diagnostics output
- - - - -
8c0740b7 by Simon Jakobi at 2020-05-13T02:04:33-04:00
docs: Add examples for Data.Semigroup.Arg{Min,Max}
Context: #17153
- - - - -
cb22348f by Ben Gamari at 2020-05-13T02:05:11-04:00
Add few cleanups of the CAF logic
Give the NameSet of non-CAFfy names a proper newtype to distinguish it
from all of the other NameSets floating about.
- - - - -
90e38b81 by Emeka Nkurumeh at 2020-05-13T02:05:51-04:00
fix printf warning when using with ghc with clang on mingw
- - - - -
86d8ac22 by Sebastian Graf at 2020-05-13T02:06:29-04:00
CprAnal: Don't attach CPR sigs to expandable bindings (#18154)
Instead, look through expandable unfoldings in `cprTransform`.
See the new Note [CPR for expandable unfoldings]:
```
Long static data structures (whether top-level or not) like
xs = x1 : xs1
xs1 = x2 : xs2
xs2 = x3 : xs3
should not get CPR signatures, because they
* Never get WW'd, so their CPR signature should be irrelevant after analysis
(in fact the signature might even be harmful for that reason)
* Would need to be inlined/expanded to see their constructed product
* Recording CPR on them blows up interface file sizes and is redundant with
their unfolding. In case of Nested CPR, this blow-up can be quadratic!
But we can't just stop giving DataCon application bindings the CPR property,
for example
fac 0 = 1
fac n = n * fac (n-1)
fac certainly has the CPR property and should be WW'd! But FloatOut will
transform the first clause to
lvl = 1
fac 0 = lvl
If lvl doesn't have the CPR property, fac won't either. But lvl doesn't have a
CPR signature to extrapolate into a CPR transformer ('cprTransform'). So
instead we keep on cprAnal'ing through *expandable* unfoldings for these arity
0 bindings via 'cprExpandUnfolding_maybe'.
In practice, GHC generates a lot of (nested) TyCon and KindRep bindings, one
for each data declaration. It's wasteful to attach CPR signatures to each of
them (and intractable in case of Nested CPR).
```
Fixes #18154.
- - - - -
e34bf656 by Ben Gamari at 2020-05-13T02:07:08-04:00
users-guide: Add discussion of shared object naming
Fixes #18074.
- - - - -
5d0f2445 by Ben Gamari at 2020-05-13T02:07:47-04:00
testsuite: Print sign of performance changes
Executes the minor formatting change in the tabulated performance
changes suggested in #18135.
- - - - -
9e4b981f by Ben Gamari at 2020-05-13T02:08:24-04:00
testsuite: Add testcase for #18129
- - - - -
266310c3 by Ivan-Yudin at 2020-05-13T02:09:03-04:00
doc: Reformulate the opening paragraph of Ch. 4 in User's guide
Removes mentioning of Hugs
(it is not helpful for new users anymore).
Changes the wording for the rest of the paragraph.
Fixes #18132.
- - - - -
4e1fb536 by Sebastian Graf at 2020-05-13T17:04:14+02:00
DmdAnal: Improve handling of precise exceptions
This patch does two things: Fix possible unsoundness in what was called
the "IO hack" and implement part 2.1 of the "fixing precise exceptions"
plan in
https://gitlab.haskell.org/ghc/ghc/wikis/fixing-precise-exceptions,
which, in combination with !2956, supersedes !3014 and !2525.
**IO hack**
The "IO hack" (which is a fallback to preserve precise exceptions
semantics and thus soundness, rather than some smart thing that
increases precision) is called `exprMayThrowPreciseException` now.
I came up with two testcases exemplifying possible unsoundness (if
twisted enough) in the old approach:
- `T13380d`: Demonstrating unsoundness of the "IO hack" when resorting
to manual state token threading and direct use of primops.
More details below.
- `T13380e`: Demonstrating unsoundness of the "IO hack" when we have
Nested CPR. Not currently relevant, as we don't have Nested
CPR yet.
- `T13380f`: Demonstrating unsoundness of the "IO hack" for safe FFI
calls.
Basically, the IO hack assumed that precise exceptions can only be
thrown from a case scrutinee of type `(# State# RealWorld, _ #)`. I
couldn't come up with a program using the `IO` abstraction that violates
this assumption. But it's easy to do so via manual state token threading
and direct use of primops, see `T13380d`. Also similar code might be
generated by Nested CPR in the (hopefully not too) distant future, see
`T13380e`. Hence, we now have a more careful test in `forcesRealWorld`
that passes `T13380{d,e}` (and will hopefully be robust to Nested CPR).
**Precise exceptions**
In #13380 and #17676 we saw that we didn't preserve precise exception
semantics in demand analysis. We fixed that with minimal changes in
!2956, but that was terribly unprincipled.
That unprincipledness resulted in a loss of precision, which is tracked
by these new test cases:
- `T13380b`: Regression in dead code elimination, because !2956 was too
syntactic about `raiseIO#`
- `T13380c`: No need to apply the "IO hack" when the IO action may not
throw a precise exception (and the existing IO hack doesn't
detect that)
Fixing both issues in !3014 turned out to be too complicated and had
the potential to regress in the future. Hence we decided to only fix
`T13380b` and augment the `Divergence` lattice with a new middle-layer
element, `ExnOrDiv`, which means either `Diverges` (, throws an
imprecise exception) or throws a *precise* exception.
See the wiki page on Step 2.1 for more implementational details:
https://gitlab.haskell.org/ghc/ghc/wikis/fixing-precise-exceptions#dead-code-elimination-for-raiseio-with-isdeadenddiv-introducing-exnordiv-step-21
- - - - -
30 changed files:
- compiler/GHC.hs
- compiler/GHC/Builtin/Names.hs
- compiler/GHC/Builtin/primops.txt.pp
- compiler/GHC/Cmm/CLabel.hs
- compiler/GHC/Cmm/Info/Build.hs
- compiler/GHC/Cmm/Utils.hs
- compiler/GHC/CmmToAsm/PPC/Cond.hs
- compiler/GHC/CmmToAsm/SPARC/Cond.hs
- compiler/GHC/CmmToAsm/X86/Cond.hs
- compiler/GHC/CmmToLlvm/Base.hs
- compiler/GHC/Core/Arity.hs
- compiler/GHC/Core/Lint.hs
- compiler/GHC/Core/Make.hs
- compiler/GHC/Core/Opt/CallArity.hs
- compiler/GHC/Core/Opt/CprAnal.hs
- compiler/GHC/Core/Opt/DmdAnal.hs
- compiler/GHC/Core/Opt/FloatIn.hs
- compiler/GHC/Core/Opt/FloatOut.hs
- compiler/GHC/Core/Opt/LiberateCase.hs
- compiler/GHC/Core/Opt/SetLevels.hs
- compiler/GHC/Core/Opt/Simplify.hs
- compiler/GHC/Core/Opt/Simplify/Utils.hs
- compiler/GHC/Core/Opt/SpecConstr.hs
- compiler/GHC/Core/Opt/Specialise.hs
- compiler/GHC/Core/Opt/WorkWrap/Utils.hs
- compiler/GHC/Core/SimpleOpt.hs
- compiler/GHC/Core/Type.hs
- compiler/GHC/Core/Unfold.hs
- compiler/GHC/Core/Utils.hs
- compiler/GHC/Driver/Main.hs
The diff was not included because it is too large.
View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/1d5265f9e8c3074c2129054880ae8dc9ed7ac4e8...4e1fb5362886e37d24d2a68881e674fbdca2bb41
--
View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/1d5265f9e8c3074c2129054880ae8dc9ed7ac4e8...4e1fb5362886e37d24d2a68881e674fbdca2bb41
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/20200513/4dd6a223/attachment-0001.html>
More information about the ghc-commits
mailing list