- Mar 02, 2021
-
-
-
-
Previously the eventlog infrastructure had a couple of races that could pop up when using the startEventLog/endEventLog interfaces. In particular, stopping and then later restarting logging could result in data preceding the eventlog header, breaking the integrity of the stream. To fix this we rework the invariants regarding the eventlog and generally tighten up the concurrency control surrounding starting and stopping of logging. We also fix an unrelated bug, wherein log events from disabled capabilities could end up never flushed.
-
- Mar 01, 2021
-
-
Consider (`T18610`): ```hs f :: Bool -> Int f x = case (x, x) of (True, True) -> 1 (False, False) -> 2 (True, False) -> 3 -- Warning: Redundant ``` The third clause will be flagged as redundant. Nevertheless, the programmer might intend to keep the clause in order to avoid bitrot. After this patch, the programmer can write ```hs g :: Bool -> Int g x = case (x, x) of (True, True) -> 1 (False, False) -> 2 (True, False) | GHC.Exts.considerAccessible -> 3 -- No warning ``` And won't be bothered any longer. See also `Note [considerAccessible]` and the updated entries in the user's guide. Fixes #18610 and #19228. -
If the context is missing it is captured as Nothing, rather than putting a noLoc in the ParsedSource. Updates haddock submodule
-
Ticket #19360 showed up a terrible bug in the occurrence analyser, in a situation like this Rec { f = g ; g = ..f... {-# RULE g .. = ...f... #-} } Then f was postInlineUnconditionally, but not in the RULE (which is simplified first), so we had a RULE mentioning a variable that was not in scope. This led me to review (again) the subtle loop-breaker stuff in the occurrence analyser. The actual changes are few, and are largely simplifications. I did a /lot/ of comment re-organising though. There was an unexpected amount of fallout. * Validation failed when compiling the stage2 compiler with profiling on. That turned to tickle a second latent bug in the same OccAnal code (at least I think it was always there), which led me to simplify still further; see Note [inl_fvs] in GHC.Core.Opt.OccurAnal. * But that in turn let me to some strange behaviour in CSE when ticks are in the picture, which I duly fixed. See Note [Dealing with ticks] in GHC.Core.Opt.CSE. * Then I got an ASSERT failure in CoreToStg, which again seems to be a latent bug. See Note [Ticks in applications] in GHC.CoreToStg * I also made one unforced change: I now simplify the RHS of a RULE in the same way as the RHS of a stable unfolding. This can allow a trivial binding to disappear sooner than otherwise, and I don't think it has any downsides. The change is in GHC.Core.Opt.Simplify.simplRules. -
This is a first step towards #18738.
-
Ticket #19364 helpfully points out that we do not currently take advantage of pushing the result type of an application into the arguments. This makes error messages notably less good. The fix is rather easy: move the result-type unification step earlier. It's even a bit more efficient; in the the checking case we now do one less zonk. See Note [Unify with expected type before typechecking arguments] in GHC.Tc.Gen.App This change generally improves error messages, but it made one worse: typecheck/should_fail/T16204c. That led me to the realisation that a good error can be replaced by a less-good one, which provoked me to change GHC.Tc.Solver.Interact.inertsCanDischarge. It's explained in the new Note [Combining equalities] One other refactoring: I discovered that KindEqOrigin didn't need a Maybe in its type -- a nice simplification.
-
As #19293 realises, this one keeps on flip flopping by 2.5% depending on how many modules there are within the GHC package. We should revert this once we figured out how to fix what's going on.
-
- Feb 28, 2021
-
-
For years we have lived in a supposedly sweet spot that gave case binders the CPR property, unconditionally. Which is an optimistic hack that is now described in `Historical Note [Optimistic case binder CPR]`. In #19232 the concern was raised that this might do more harm than good and that might be better off simply by taking the CPR property of the scrutinee for the CPR type of the case binder. And indeed that's what we do now. Since `Note [CPR in a DataAlt case alternative]` is now only about field binders, I renamed and garbage collected it into `Note [Optimistic field binder CPR]`. NoFib approves: ``` NoFib Results -------------------------------------------------------------------------------- Program Allocs Instrs -------------------------------------------------------------------------------- anna +0.1% +0.1% nucleic2 -1.2% -0.6% sched 0.0% +0.9% transform -0.0% -0.1% -------------------------------------------------------------------------------- Min -1.2% -0.6% Max +0.1% +0.9% Geometric Mean -0.0% +0.0% ``` Fixes #19232. -
This test flip-flops by +-1% in arbitrary changes in CI. While playing around with `-dunique-increment`, I could reproduce variations of 3% in compiler allocations, so I set the acceptance window accordingly. Fixes #19414.
-
The previous code using TyCoMapper could promote the same metavar twice. Use a set instead.
-
- Feb 27, 2021
-
-
The test probably could have used `usleep` from `unistd.h` instead, but this seemed like the simplest solution. Fixes #19431.
-
When desugaring large overloaded literals we now avoid computing the `Rational` value. Instead prefering to store the significant and exponent as given where reasonable and possible. See Note [FractionalLit representation] for details.
-
It should be left to tooling to perform the filtering to remove these specific closure types from the profile if desired. Fixes #16795
-
- Feb 26, 2021
-
-
It was revealed in #19363 that the Template Haskell pretty-printer implemented in `Language.Haskell.TH.Ppr` did not pretty-print infix names or symbolic names correctly in certain situations, such as in data constructor declarations or fixity declarations. Easily fixed by using `pprName' Applied` (which always parenthesizes symbolic names in prefix position) or `pprName' Infix` (which always surrounds alphanumeric names with backticks in infix position) in the right spots. Fixes #19363.
-
Fixes #18966. Adds a new warning -Wambiguous-fields for uses of field selectors or record updates that will be rejected in the future, when the DuplicateRecordFields extension is simplified per https://github.com/ghc-proposals/ghc-proposals/pull/366.
-
- Feb 24, 2021
-
-
Ben Gamari authored
Needed by #19025.
-
Ben Gamari authored
This reverts commit 4a9d856d.
-
- Feb 23, 2021
-
-
Ben Gamari authored
-
- Feb 22, 2021
-
-
alignment clearly should be a power of two. This patch makes it so. We do so by using the #alignment directive instead of using the size of the type.
-
-
As noted in #18391, foreignInterruptible fails pretty regularly under GHCi.
-
-
We used to only show the fixity of an operator if it wasn't the default fixity. Usually this was when the fixity was undeclared, but it could also arise if one declared the fixity of an operator as infixl 9, the default fixity. This commit makes it so that :i always shows the fixity of an operator, even if it is unset. We may want in the future to keep track of whether an operator's fixity is defined, so that we can print a comment like infixl 9 # -- Assumed, since no fixity is declared. for operators with no specified fixity, and so that we can print fixity of a term with a non-symbolic term when its fixity has been manually specified as infixl 9. Implements #19200.
-
-
- Feb 19, 2021
-
-
Simon Peyton Jones authored
When implementing Quick Look I'd failed to remember that overloaded labels, like #foo, should be treated as a "head", so that they can be instantiated with Visible Type Application. This caused #19154. A very similar ticket covers overloaded literals: #19167. This patch fixes both problems, but (annoyingly, albeit temporarily) in two different ways. Overloaded labels I dealt with overloaded labels by buying fully into the Rebindable Syntax approach described in GHC.Hs.Expr Note [Rebindable syntax and HsExpansion]. There is a good overview in GHC.Rename.Expr Note [Handling overloaded and rebindable constructs]. That module contains much of the payload for this patch. Specifically: * Overloaded labels are expanded in the renamer, fixing #19154. See Note [Overloaded labels] in GHC.Rename.Expr. * Left and right sections used to have special code paths in the typechecker and desugarer. Now we just expand them in the renamer. This is harder than it sounds. See GHC.Rename.Expr Note [Left and right sections]. * Infix operator applications are expanded in the typechecker, specifically in GHC.Tc.Gen.App.splitHsApps. See Note [Desugar OpApp in the typechecker] in that module * ExplicitLists are expanded in the renamer, when (and only when) OverloadedLists is on. * HsIf is expanded in the renamer when (and only when) RebindableSyntax is on. Reason: the coverage checker treats HsIf specially. Maybe we could instead expand it unconditionally, and fix up the coverage checker, but I did not attempt that. Overloaded literals Overloaded literals, like numbers (3, 4.2) and strings with OverloadedStrings, were not working correctly with explicit type applications (see #19167). Ideally I'd also expand them in the renamer, like the stuff above, but I drew back on that because they can occur in HsPat as well, and I did not want to to do the HsExpanded thing for patterns. But they *can* now be the "head" of an application in the typechecker, and hence something like ("foo" @T) works now. See GHC.Tc.Gen.Head.tcInferOverLit. It's also done a bit more elegantly, rather than by constructing a new HsExpr and re-invoking the typechecker. There is some refactoring around tcShortCutLit. Ultimately there is more to do here, following the Rebindable Syntax story. There are a lot of knock-on effects: * HsOverLabel and ExplicitList no longer need funny (Maybe SyntaxExpr) fields to support rebindable syntax -- good! * HsOverLabel, OpApp, SectionL, SectionR all become impossible in the output of the typecheker, GhcTc; so we set their extension fields to Void. See GHC.Hs.Expr Note [Constructor cannot occur] * Template Haskell quotes for HsExpanded is a bit tricky. See Note [Quotation and rebindable syntax] in GHC.HsToCore.Quote. * In GHC.HsToCore.Match.viewLExprEq, which groups equal HsExprs for the purpose of pattern-match overlap checking, I found that dictionary evidence for the same type could have two different names. Easily fixed by comparing types not names. * I did quite a bit of annoying fiddling around in GHC.Tc.Gen.Head and GHC.Tc.Gen.App to get error message locations and contexts right, esp in splitHsApps, and the HsExprArg type. Tiresome and not very illuminating. But at least the tricky, higher order, Rebuilder function is gone. * Some refactoring in GHC.Tc.Utils.Monad around contexts and locations for rebindable syntax. * Incidentally fixes #19346, because we now print renamed, rather than typechecked, syntax in error mesages about applications. The commit removes the vestigial module GHC.Builtin.RebindableNames, and thus triggers a 2.4% metric decrease for test MultiLayerModules (#19293). Metric Decrease: MultiLayerModules T12545
-
- Feb 18, 2021
-
-
Rather than just display every 100 tests, work out how many to display based on the total number of tests. This improves the experience when running a small number of tests. For [0..100] - Report every test [100..1000] - Report every 10 tests [1000..10000] - Report every 100 tests and so on.. -
At a SPECIALSE pragma for an imported Id, we used to check that it was marked INLINABLE. But that turns out to interact badly with worker/wrapper: see Note [Worker-wrapper for INLINABLE functions] in GHC.Core.Opt.WorkWrap. So this small patch instead simply tests that we have an unfolding for the function; see Note [SPECIALISE pragmas for imported Ids] in GHC.Tc.Gen.Sig. Fixes #19246
-
- Feb 17, 2021
-
-
Previously, the desugarer was looking up names referenced in TH-quoted `ANN`s by using `globalVar`, which would allocate a fresh TH `Name`. In effect, this would prevent quoted `ANN`s from ever referencing the correct identifier `Name`, leading to #19377. The fix is simple: instead of `globalVar`, use `lookupLOcc`, which properly looks up the name of the in-scope identifier. Fixes #19377.
-
For now this just tests that the order of the callbacks is what we expect for a couple of synthetic heap graphs.
-
- Feb 16, 2021
-
-
This adds a new `otycon` production to the parser that allows for type constructor names that are either alphanumeric (`tycon`) or symbolic (`tyconsym`), where the latter must be parenthesized appropriately. `otycon` is much like the existing `oqtycon` production, except that it does not permit qualified names. The parser now uses `otycon` to parse type constructor names in `ANN type` declarations, which fixes #19374. To make sure that all of this works, I added three test cases: * `should_compile/T19374a`: the original test case from #19374 * `should_fail/T19374b`: a test that makes sure that an `ANN` with a qualified name fails to parse * `should_fail/T19374c`: a test that makes sure that an `ANN type` with a qualified name fails to parse
-
Fixes #17853. We mustn't discard the result of pickGREs, because doing so might lead to incorrect redundant import warnings.
-
This is a small fix that depends on the previous commit, because it corrected the rnExpr free variable calculation for HsVars which refer to ambiguous fields. Fixes #19213.
-
Fixes #5972. This adds an extension NoFieldSelectors to disable the generation of selector functions corresponding to record fields. When this extension is enabled, record field selectors are not accessible as functions, but users are still able to use them for record construction, pattern matching and updates. See Note [NoFieldSelectors] in GHC.Rename.Env for details. Defining the same field multiple times requires the DuplicateRecordFields extension to be enabled, even when NoFieldSelectors is in use. Along the way, this fixes the use of non-imported DuplicateRecordFields in GHCi with -fimplicit-import-qualified (fixes #18729). Moreover, it extends DisambiguateRecordFields to ignore non-fields when looking up fields in record updates (fixes #18999), as described by Note [DisambiguateRecordFields for updates]. Co-authored-by:
Simon Hafner <hafnersimon@gmail.com> Co-authored-by:
Fumiaki Kinoshita <fumiexcel@gmail.com>
-
- Feb 14, 2021
-
-
Daniel Gröber (dxld) authored
Alignments passed to alloca and friends must be a power of two for the code in allocatePinned to work properly. Commit 41230e26 ("Zero out pinned block alignment slop when profiling") introduced an ASSERT for this but this test was still violating it.
-
In GHC.Core.SimpleOpt, I found that its inlining could duplicate an arbitary redex inside a lambda! Consider (\xyz. x+y). The occurrence-analysis treats the lamdda as a group, and says that both x and y occur once, even though the occur under the lambda-z. See Note [Occurrence analysis for lambda binders] in OccurAnal. When the lambda is under-applied in a call, the Simplifier is careful to zap the occ-info on x,y, because they appear under the \z. (See the call to zapLamBndrs in simplExprF1.) But SimpleOpt missed this test, resulting in #19347. So this patch * commons up the binder-zapping in GHC.Core.Utils.zapLamBndrs. * Calls this new function from GHC.Core.Opt.Simplify * Adds a call to zapLamBndrs to GHC.Core.SimpleOpt.simple_app This change makes test T12990 regress somewhat, but it was always very delicate, so I'm going to put up with that. In this voyage I also discovered a small, rather unrelated infelicity in the Simplifier: * In GHC.Core.Opt.Simplify.simplNonRecX we should apply isStrictId to the OutId not the InId. See Note [Dark corner with levity polymorphism] It may never "bite", because SimpleOpt should have inlined all the levity-polymorphic compulsory inlnings already, but somehow it bit me at one point and it's generally a more solid thing to do. Fixing the main bug increases runtime allocation in test perf/should_run/T12990, for (acceptable) reasons explained in a comement on Metric Increase: T12990
-
-
Fix the following rule: "fromIntegral/Int->Natural" fromIntegral = naturalFromWord . fromIntegral Its type wasn't constrained to Int hence #19345.
-
The roughMatchTcs function enables a quick definitely-no-match test in lookupInstEnv. Unfortunately, it didn't account for type families. This didn't matter when type families were flattened away, but now they aren't flattened it matters a lot. The fix is very easy. See INVARIANT in GHC.Core.InstEnv Note [ClsInst laziness and the rough-match fields] Fixes #19336 The change makes compiler perf worse on two very-type-family-heavy benchmarks, T9872{a,d}: T9872a(normal) ghc/alloc 2172536442.7 2216337648.0 +2.0% T9872d(normal) ghc/alloc 614584024.0 621081384.0 +1.1% (Everything else is 0.0% or at most 0.1%.) I think we just have to put up with this. Some cases were being wrongly filtered out by roughMatchTcs that might actually match, which could lead to false apartness checks. And it only affects these very type-family-heavy cases. Metric Increase: T9872a T9872d
-
