| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This patch fixes #22745 and #15205, which are about GHC's
failure to discard unnecessary superclass selections that
yield coercions. See
GHC.Core.Utils Note [exprOkForSpeculation and type classes]
The main changes are:
* Write new Note [NON-BOTTOM_DICTS invariant] in GHC.Core, and
refer to it
* Define new function isTerminatingType, to identify those
guaranteed-terminating dictionary types.
* exprOkForSpeculation has a new (very simple) case for ClassOpId
* ClassOpId has a new field that says if the return type is
an unlifted type, or a terminating type.
This was surprisingly tricky to get right. In particular note
that unlifted types are not terminating types; you can write an
expression of unlifted type, that diverges. Not so for dictionaries
(or, more precisely, for the dictionaries that GHC constructs).
Metric Decrease:
LargeRecord
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The key change is that in GHC.Core.Opt.Specialise.specLookupRule
we were using realIdUnfolding, which ignores the loop-breaker
flag. When given a loop breaker, rule matching therefore
looped infinitely -- #22802.
In fixing this I refactored a bit.
* Define GHC.Core.InScopeEnv as a data type, and use it.
(Previously it was a pair: hard to grep for.)
* Put several functions returning an IdUnfoldingFun into
GHC.Types.Id, namely
idUnfolding
alwaysActiveUnfoldingFun,
whenActiveUnfoldingFun,
noUnfoldingFun
and use them. (The are all loop-breaker aware.)
|
|
|
|
|
|
|
|
|
|
|
| |
This patch fixes two issues in the way that `type data` declarations were
reified with Template Haskell:
* `type data` data constructors are now properly reified using `DataConI`.
This is accomplished with a special case in `reifyTyCon`. Fixes #22818.
* `type data` type constructors are now reified in `reifyTyCon` using
`TypeDataD` instead of `DataD`. Fixes #22819.
|
|
|
|
|
|
|
|
|
|
|
|
| |
Unlike most other data constructors, data constructors declared with `type data`
are represented in `TyThing`s as `ATyCon` rather than `ADataCon`. The `ATyCon`
case in `tyThingParent_maybe` previously did not consider the possibility of
the underlying `TyCon` being a promoted data constructor, which led to the
oddities observed in #22817. This patch adds a dedicated special case in
`tyThingParent_maybe`'s `ATyCon` case for `type data` data constructors to fix
these oddities.
Fixes #22817.
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Previously we tried to lower the alignment requirement as far as
possible, based on the section kind inferred from the CLabel. For info
tables, .p2align 1 was applied given the GC should only need the
lowest bit to tag forwarding pointers. But this would lead to
unaligned loads/stores, which has a performance penalty even if the
wasm spec permits it. Furthermore, the test suite has shown memory
corruption in a few cases when compacting gc is used.
This patch takes a more conservative approach: all data sections
except C strings align to word size.
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Asm-shortcutting may produce relative references to symbols defined in
other compilation units. This is not something that MachO relocations
support (see #21972). For this reason we disable the optimisation on
Darwin. We do so without a warning since this flag is enabled by `-O2`.
Another way to address this issue would be to rather implement a
PLT-relocatable jump-table strategy. However, this would only benefit
Darwin and does not seem worth the effort.
Closes #21972.
|
|
|
|
|
| |
Properly handle ForeignHints of ccall arguments/return value, insert
sign extends and truncations when handling signed subwords. Fixes #22852.
|
| |
|
|
|
|
|
|
|
| |
The CmmBlock datacon was not handled in lower_CmmLit, since I thought
it would have been eliminated after proc-point splitting. Turns out it
still occurs in very rare occasions, and this patch is needed to fix
T9329 for wasm.
|
|
|
|
|
|
| |
This patch fixes the W8/W16 literal narrowing logic in the wasm NCG,
which used to lower it to something like i32.const -1, without
properly zeroing-out the unused higher bits. Fixes #22608.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
I removed all occurrences of TcRnUnknownMessage in GHC.Rename.Bind
module. Instead, these TcRnMessage messages were introduced:
TcRnMultipleFixityDecls
TcRnIllegalPatternSynonymDecl
TcRnIllegalClassBiding
TcRnOrphanCompletePragma
TcRnEmptyCase
TcRnNonStdGuards
TcRnDuplicateSigDecl
TcRnMisplacedSigDecl
TcRnUnexpectedDefaultSig
TcRnBindInBootFile
TcRnDuplicateMinimalSig
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Previously, when using `capi` calling convention in foreign declarations,
code generator failed to handle const-cualified pointer return types.
This resulted in CC toolchain throwing `-Wincompatible-pointer-types-discards-qualifiers`
warning.
`Foreign.C.Types.ConstPtr` newtype was introduced to handle these cases -
special treatment was put in place to generate appropritetly qualified C
wrapper that no longer triggers the above mentioned warning.
Fixes #22043.
|
|
|
|
|
|
| |
types (#22043)"
This reverts commit 99aca26b652603bc62953157a48e419f737d352d.
|
|
|
|
|
|
|
|
|
|
| |
This accepts the orphan type family instance
type instance DsForeignHook = ...
in GHC.HsToCore.Types.
See Note [The Decoupling Abstract Data Hack] in GHC.Driver.Hooks
|
|
|
|
|
| |
This moves Anno instances for PatBuilder from GHC.Parser.PostProcess
to GHC.Parser.Types to avoid orphans.
|
|
|
|
|
|
|
| |
This patch removes some orphan instances in the STG namespace
by introducing the GHC.Stg.Lift.Types module, which allows various
type family instances to be moved to GHC.Stg.Syntax, avoiding orphan
instances.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This fixes the fact that we were not reporting orphan family instances
at all. The fix here is easy, but touches a bit of code. I refactored
the code to be much more similar to the way that class instances are done:
- Add a fi_orphan field to FamInst, like the is_orphan field in ClsInst
- Make newFamInst initialise this field, just like newClsInst
- And make newFamInst report a warning for an orphan, just like newClsInst
- I moved newFamInst from GHC.Tc.Instance.Family to GHC.Tc.Utils.Instantiate,
just like newClsInst.
- I added mkLocalFamInst to FamInstEnv, just like mkLocalClsInst in InstEnv
- TcRnOrphanInstance and SuggestFixOrphanInstance are now parametrised
over class instances vs type/data family instances.
Fixes #19773
|
|
|
|
|
|
|
|
|
| |
We were treating a type-family instance as a non-orphan if there
was a type constructor on its /right-hand side/ that was local. Boo!
Utterly wrong. With this patch, we correctly check the /left-hand side/
instead!
Fixes #22717
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Problem: in 02279a9c the type-level [] syntax was changed from a built-in name
to an alias for the GHC.Types.List constructor. badOrigBinding assumes that if
a name is not built-in then it must have come from TH quotation, but this is
not necessarily the case with [].
The outdated assumption in badOrigBinding leads to incorrect error messages.
This code:
data []
Fails with "Cannot redefine a Name retrieved by a Template Haskell quote: []"
Unfortunately, there is not enough information in RdrName to directly determine
if the name was constructed via TH or by the parser, so this patch changes the
error message instead.
It unifies TcRnIllegalBindingOfBuiltIn and TcRnNameByTemplateHaskellQuote
into a new error TcRnBindingOfExistingName and changes its wording to avoid
guessing the origin of the name.
|
|
|
|
|
|
| |
See Note [Unwrap newtypes first], which has the details.
Close #22519.
|
|
|
|
| |
Towards #22530
|
|
|
|
|
|
|
| |
This occname has just been derived from an `Id`, so need to force it
promptly so we can release the Id back to the world.
Another symptom of the bug caused by #19619
|
|
|
|
|
|
|
|
|
|
|
| |
Doesn't force the lazy `OccName` field (#19619) which is already known
as a really bad source of leaks.
When we slam the hammer storing Names on disk (in interface files or the
like), all this should be forced as otherwise a `Name` can easily retain
an `Id` and hence the entire world.
Fixes #22833
|
|
|
|
|
|
|
|
|
| |
This fixes a tricky leak in GHCi where we were retaining old copies of
HscEnvs when reloading. If not all modules were recompiled then these
hydrated fields in break points would retain a reference to the old
HscEnv which could double memory usage.
Fixes #22530
|
|
|
|
|
| |
Create and use moduleGraphModulesBelow in GHC.Unit.Module.Graph that
doesn't need anything from the driver to be used.
|
|
|
|
|
|
|
|
|
| |
Nothing deep here; I had failed to bring some
floated dictionary binders into scope.
Exposed by -fspecialise-aggressively
Fixes #22715.
|
|
|
|
|
|
|
|
| |
In the wasm NCG, we used to compile MO_F_Neg to 0.0-x. It was an
oversight, there actually exists f32.neg/f64.neg opcodes in the wasm
spec and those should be used instead! The old behavior almost works,
expect when GHC compiles the -0.0 literal, which will incorrectly
become 0.0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
We used to print the offset value to a platform word sized integer.
This is incorrect when the offset is negative (e.g. output of cmm
constant folding) and the register is 64-bit but on a 32-bit target,
and may lead to incorrect runtime result (e.g. #22607).
The fix is simple: just treat it as a proper MO_Add, with the correct
width info inferred from the register itself.
Metric Increase:
T12707
T13379
T4801
T5321FD
T5321Fun
|
|
|
|
|
|
|
|
|
|
| |
Lint was checking for duplicate external names by calling removeDups,
which needs a comparison function that is passed to Data.List.sortBy.
But the comparison was not a valid ordering - it returned LT
if one of the names was not external.
For example, the previous implementation won't find a duplicate in
[M.x, y, M.x].
Instead, we filter out non-external names before looking for duplicates.
|
|
|
|
| |
This is helpful when debugging multiple component issues.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
In order to preserve existing behaviour it's important to look within the current component before consideirng a module might come from an external component.
This already happened by accident in `downsweep`, (because roots are used to repopulated the cache) but in the `Finder` the logic was the wrong way around.
Fixes #22680
-------------------------
Metric Decrease:
MultiComponentModules
MultiComponentModulesRecomp
-------------------------p
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Multiple units can refer to the same files without any problem. Just
another assumption which needs to be updated when we may have multiple
home units.
However, there is the invariant that within each unit each file only
maps to one module, so as long as we also key the cache by UnitId then
we are all good.
This led to some confusing behaviour in GHCi when reloading,
multipleHomeUnits_shared distils the essence of what can go wrong.
Fixes #22679
|
|
|
|
|
|
|
|
|
|
| |
Currently the driver diagnostics don't give any indication about which unit they correspond to.
For example `-Wmissing-home-modules` can fire multiple times for each different home unit and gives no indication about which unit it's actually reporting about.
Perhaps a longer term fix is to generalise the providence information away from a SrcSpan so that these kind of whole project errors can be reported with an accurate provenance. For now we can just include the `UnitId` in the error message.
Fixes #22678
|
|
|
|
|
|
|
|
| |
We should not be producing object files when in interactive mode but we
still produced the dummy o-boot files. These never made it into a
`Linkable` but then confused the recompilation checker.
Fixes #22669
|
|
|
|
|
|
|
|
|
|
| |
hs-boot combo
In interactive mode we don't produce any linkables for hs-boot files. So
we also need to not going looking for them when we check to see if we
have all the right objects needed for recompilation.
Ticket #22669
|
|
|
|
|
|
|
|
|
|
|
| |
The `pruneCache` function assumes that the list of `CachedInfo` all have unique `ModuleName`, this is not true:
* In normal compilation, the same module name can appear for a file and it's boot file.
* In multiple home unit compilation the same ModuleName can appear in different units
The fix is to use a `NodeKey` as the actual key for the interfaces which includes `ModuleName`, `IsBoot` and `UnitId`.
Fixes #22677
|
|
|
|
|
|
|
|
|
|
|
|
| |
satisfies target
This fixes a spurious warning in -Wmissing-home-modules.
This is a simple oversight where when looking for the target in the
first place we augment the search by the -working-directory flag but
then fail to do so when checking this warning.
Fixes #22676
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The key part of this change is to store a UnitId in the
`UsageHomeModule` and `UsageHomeModuleInterface`.
* Fine-grained dependency tracking is used if the dependency comes from
any home unit.
* We actually look up the right module when checking whether we need to
recompile in the `UsageHomeModuleInterface` case.
These scenarios are both checked by the new tests (
multipleHomeUnits_recomp and multipleHomeUnits_recomp_th )
Fixes #22675
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
runtimeRepLevity_maybe was panicing unnecessarily; and
the error printing code made use of the case when it should
return Nothing rather than panicing.
For some bizarre reason perf/compiler/T21839r shows a 10% bump in runtime
peak-megagbytes-used, on a single architecture (alpine). See !9753 for
commentary, but I'm going to accept it.
Metric Increase:
T21839r
|
|
|
|
|
|
| |
To be able to capture string literals with possible escape codes as labels.
Close #22771
|
|
|
|
|
| |
These flags did not make it into the 9.6 release series,
so the "since" annotations must be corrected.
|
|
|
|
|
|
|
|
|
|
|
|
| |
Problem: In 2463df2f, the Solo data constructor was renamed to MkSolo,
and Solo was turned into a pattern synonym for backwards compatibility.
Since pattern synonyms can not be promoted, the old code that pretty-printed
promoted single-element tuples started producing ill-typed code:
t :: Proxy ('Solo Int)
This fails with "Pattern synonym ‘Solo’ used as a type"
The solution is to track the distinction between type constructors and data
constructors more carefully when printing single-element tuples.
|
|
|
|
|
|
|
|
|
|
|
| |
In #22764 a user noticed that a program implementing a simple atomic
counter via an STRef regressed significantly due to the introduction of
necessary atomic operations in the MutVar# primops (#22468). This
regression was caused by a bug in the NCG, which emitted an unnecessary
MFENCE instruction for a release-ordered atomic write. MFENCE is rather
only needed to achieve sequentially consistent ordering.
Fixes #22764.
|
| |
|
| |
|
|
|
|
|
|
|
|
| |
This adds `-Werror=<group>` and `-fwarn-<group>` flags for warning
groups as well as individual warnings. Previously these were defined
on an ad hoc basis so for example we had `-Werror=compat` but not
`-Werror=unused-binds`, whereas we had `-fwarn-unused-binds` but not
`-fwarn-compat`. Fixes #22182.
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This adds support for calling Cmm code from bytecode using the native
calling convention, allowing modules that use `foreign import prim`
to be loaded and debugged in GHCi.
This patch introduces a new `PRIMCALL` bytecode instruction and
a helper stack frame `stg_primcall`. The code is based on the
existing functionality for dealing with unboxed tuples in bytecode,
which has been generalised to handle arbitrary calls.
Fixes #22051
|
|
|
|
|
|
|
|
| |
We need to ensure that the output of `cvtSigTypeKind` is parenthesized (at
precedence `sigPrec`) so that any type signatures with an outermost, explicit
kind signature can parse correctly.
Fixes #22784.
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Following the plan in GHC Proposal #143 "Remove the * kind syntax",
which states:
In the next release (or 3 years in), enable -fwarn-star-is-type by default.
The "next release" happens to be 9.6.1
I also moved the T21583 test case from should_fail to should_compile,
because the only reason it was failing was -Werror=compat in our test
suite configuration.
|
|
|
|
| |
This MR is in response to the discussion on #22719
|