- 03 Jun, 2021 3 commits
-
-
-
The tests `CountParserDeps.hs` and `CountAstDeps.hs` are implemented by calling `CountDeps`. In this MR, `CountDeps.printDeps` is updated such tat by uncommenting a line, you can print a module's dependency graph showing what includes what. The output is in a format suitable for use with graphviz.
-
As the now historic part of `NOTE [aBSENT_ERROR_ID]` explains, we used to have `exprIsHNF` respond True to `absentError` and give it a non-bottoming demand signature, in order to perform case-to-let on certain `case`s we used to emit that scrutinised `absentError` (Urgh). What changed, why don't we emit these questionable absent errors anymore? The absent errors in question filled in for binders that would end up in strict fields after being seq'd. Apparently, the old strictness analyser would give these binders an absent demand, but today we give them head-strict demand `1A` and thus don't replace with absent errors at all. This fixes items (1) and (2) of #19853.
-
- 02 Jun, 2021 9 commits
-
-
When compiling Agda we entered into an infinite loop as the stopping condition was a bit wrong in hptSomeModulesBelow. The bad situation was something like * We would see module A (NotBoot) and follow it dependencies * Later on we would encounter A (Boot) and follow it's dependencies, because the lookup would not match A (NotBoot) and A (IsBoot) * Somewhere in A (Boot)s dependencies, A (Boot) would appear again and lead us into an infinite loop. Now the state marks whether we have been both variants (IsBoot and NotBoot) so we don't follow dependencies for A (Boot) many times.
-
This causes hadrian to be built twice because the second time uses a different index state.
-
-
The rule before decided to build the whole stage1 compiler, but this was unecessary as we were just missing one header file which can be generated directly by calling configure. Before: 18 minutes After: 54s
-
Before this patch, GHC used to silently accept programs such as the following: data R where D1 :: { d1 :: Int } %1 -> R The %1 annotation was completely ignored. Now it is a proper error. One remaining issue is that in the error message (⊸) turns into (%1 ->). This is to be corrected with upcoming exactprint updates.
-
-
* For primops from `GHC.Prim` lookup the HValues in `GHC.PrimopWrappers`. * Add short error messages if a user tries to use a *Non-Id* value or a `pseudoop` in a `:print`, `:sprint` or `force`command. * Add additional test cases for `Magic Ids`.
-
The sequencing of monadic effects in foldlM and foldrM was described as respectively right-associative and left-associative, but this could be confusing, as in essence we're just composing Kleisli arrows, whose composition is simply associative. What matters therefore is the order of sequencing of effects, which can be described more clearly without dragging in associativity as such. This avoids describing these folds as being both left-to-right and right-to-left depending on whether we're tracking effects or operator application. The new text should be easier to understand.
-
- 29 May, 2021 14 commits
-
-
Ben Gamari authored
-
Ben Gamari authored
GHC's internal State monad benefits from oneShot annotations on its state, allowing for more aggressive eta expansion. We currently don't have monad transformers with the same optimisation, so we only change uses of the pure State monad here. See #19657 and 19380. Metric Decrease: hie002
-
Ben Gamari authored
-
This creates new modules GHC.Tc.Solver.InertSet and GHC.Tc.Solver.Types. The Monad module is still pretty big, but this is an improvement. Moreover, it means that GHC.HsToCore.Pmc.Solver.Types no longer depends on the constraint solver (it now depends on GHC.Tc.Solver.InertSet), making the error-messages work easier. This patch thus contributes to #18516.
-
-
We need to match on DataCon workers for the rules to be triggered. T13701 ghc/alloc decreases by ~2.5% on some archs Metric Decrease: T13701
-
This also demotes the error message about -fkeep-going to a trace message which matches the behaviour of other build systems (such as cabal-install and nix) which don't print any message like this on a failure. We want to remove the stable module check in a future patch, which is an approximation of `-fkeep-going`. At the moment this change shouldn't do very much.
-
-
-
-
-
The stg_ctoi_t and stg_ret_t procedures which convert unboxed tuples between the bytecode an native calling convention were causing a panic when using the LLVM backend. Fixes #19591
-
After !4741, it was no longer possible to silence a warning about a missing pattern synonym signature if the `-Wmissing-signatures` flag was on. Restore the previous semantics while still adhering to the principle "enabling an additional warning flag should never make prior warnings disappear". For more symmetry and granularity, introduce `-Wmissing-exported-pattern-synonym-signatures`. See Note [Missing signatures] for an overview of all flags involved.
-
- 28 May, 2021 2 commits
-
-
Using the following high-quality benchmark (with -O2): main :: IO () main = do let go 0 = "" go n@(W# n#) = showWord n# (go (n -1)) print $ length (go 10000000) I get the following performance results: - remWord+quotRem: 0,76s user 0,00s system 99% cpu 0,762 total - quotRemWord: 0,45s user 0,01s system 99% cpu 0,456 total Note that showSignedInt already uses quotRemInt.
-
-
- 27 May, 2021 4 commits
-
-
Alan Zimmerman authored
Closes #19903 Note: the normal ppr does not reproduce unicode linear arrows, so that part of the normal printing test is ommented out in the Makefile for this test. See #18846
-
This fixes #19823
-
Another change in a series improving record syntax in the AST. The key change in this commit is the renaming of `HsFieldLabel` to `DotFieldOcc`.
-
In the common case this is a straight performance win at a compile time cost so we enable it at -O2. In rare cases it can lead to compile time regressions because of changed inlining behaviour. Which can very rarely also affect runtime performance. Increasing the inlining threshold can help to avoid this which is documented in the user guide. In terms of measured results this reduced instructions executed for nofib by 1%. However for some cases (e.g. Cabal) enabling this by default increases compile time by 2-3% so we enable it only at -O2 where it's clear that a user is willing to trade compile time for runtime. Most of the testsuite runs without -O2 so there are few perf changes. Increases: T12545/T18698: We perform more WW work because dicts are now treated strict. T9198: Also some more work because functions are now subject to W/W Decreases: T14697: Compiling empty modules. Probably because of changes inside ghc. T9203: I can't reproduce this improvement locally. Might be spurious. ------------------------- Metric Decrease: T12227 T14697 T9203 Metric Increase: T9198 T12545 T18698a T18698b -------------------------
-
- 26 May, 2021 2 commits
-
-
This change aims to make source files relocatable w.r.t. to the interface files produced by the compiler. This is so that we can download interface files produced by a cloud build system and then reuse them in a local ghcide session catch another case of implicit includes actually use the implicit quote includes add another missing case recomp020 test that .hi files are reused even if .hs files are moved to a new location Added recomp021 to record behaviour with non implicit includes add a note additional pointer to the note Mention #16956 in Note
-
This commit converts the lexers and all the parser machinery to use the new parser types and diagnostics infrastructure. Furthermore, it cleans up the way the parser code was emitting hints. As a result of this systematic approach, the test output of the `InfixAppPatErr` and `T984` tests have been changed. Previously they would emit a `SuggestMissingDo` hint, but this was not at all helpful in resolving the error, and it was even confusing by just looking at the original program that triggered the errors. Update haddock submodule
-
- 25 May, 2021 6 commits
-
-
Richard Eisenberg authored
This is part of #18738 [skip ci]
-
This reverts commit 673ff667.
-
-
-
This prohuibits CC=clang to work generically and will always bake in the clang that is found on the build machine in PATH, what ever clang that might be. It might not even be on the final host.
-
This should prevent some other `bash` to leak into the binary distributions.
-