GitLab

Sven Tennie pushed to branch wip/supersven/riscv-vectors at Glasgow Haskell Compiler / GHC

Commits:

ee2dc248

by Simon Hengel at 2025-07-31T06:25:35-04:00

Update comments on `OptKind` to reflect the code reality

b029633a

by Wen Kokke at 2025-07-31T06:26:21-04:00

rts: Disable --eventlog-flush-interval unless compiled with -threaded.

This commit fixes issue #26222:
Using --eventlog-flush-interval with the non-threaded RTS leads to eventlog corruption.
https://gitlab.haskell.org/ghc/ghc/-/issues/26222

This commit makes three changes when code is compiled against the non-threaded RTS:

1. It disables the --eventlog-flush-interval flag.
2. It disables the documentation for the --eventlog-flush-interval flag.
3. It disables the relevant state from RtsConfig and code from Timer.
4. It updates the entry for --eventlog-flush-interval in the users guide.

31159f1d

by Wen Kokke at 2025-07-31T06:26:21-04:00

rts: Split T20006 into tests with and without -threaded

618687ef

by Simon Hengel at 2025-07-31T06:27:03-04:00

docs/users_guide/win32-dlls.rst: Remove references to `readline`

083e40f1

by Rodrigo Mesquita at 2025-08-01T04:38:23-04:00

debugger: Uniquely identify breakpoints by internal id

Since b85b11994e0130ff2401dd4bbdf52330e0bcf776 (support inlining
breakpoints), a breakpoint has been identified at runtime by *two* pairs
of <module,index>.

- The first, aka a 'BreakpointId', uniquely identifies a breakpoint in
  the source of a module by using the Tick index. A Tick index can index
  into ModBreaks.modBreaks_xxx to fetch source-level information about
  where that tick originated.

    - When a user specifies e.g. a line breakpoint using :break, we'll reverse
      engineer what a Tick index for that line
    - We update the `BreakArray` of that module (got from the
      LoaderState) at that tick index to `breakOn`.
    - A BCO we can stop at is headed by a BRK_FUN instruction. This
      instruction stores in an operand the `tick index` it is associated
      to. We look it up in the associated `BreakArray` (also an operand)
      and check wheter it was set to `breakOn`.

- The second, aka the `ibi_info_mod` + `ibi_info_ix` of the
  `InternalBreakpointId`, uniquely index into the `imodBreaks_breakInfo`
  -- the information we gathered during code generation about the
  existing breakpoint *ocurrences*.

  - Note that with optimisation there may be many occurrences of the
    same source-tick-breakpoint across different modules. The
    `ibi_info_ix` is unique per occurrence, but the `bi_tick_ix` may be
    shared. See Note [Breakpoint identifiers] about this.

  - Note that besides the tick ids, info ids are also stored in
    `BRK_FUN` so the break handler can refer to the associated
    `CgBreakInfo`.

In light of that, the driving changes come from the desire to have the
info_id uniquely identify the breakpoint at runtime, and the source tick
id being derived from it:

- An InternalBreakpointId should uniquely identify a breakpoint just
  from the code-generation identifiers of `ibi_info_ix` and `ibi_info_mod`.
  So we drop `ibi_tick_mod` and `ibi_tick_ix`.
- A BRK_FUN instruction need only record the "internal breakpoint id",
  not the tick-level id.
  So we drop the tick mod and tick index operands.
- A BreakArray should be indexed by InternalBreakpointId rather than
  BreakpointId

That means we need to do some more work when setting a breakpoint.
Specifically, we need to figure out the internal ids (occurrences of a
breakpoint) from the source-level BreakpointId we want to set the
breakpoint at (recall :break refers to breaks at the source level).

Besides this change being an improvement to the handling of breakpoints
(it's clearer to have a single unique identifier than two competing
ones), it unlocks the possibility of generating "internal" breakpoints
during Cg (needed for #26042).
It should also be easier to introduce multi-threaded-aware `BreakArrays`
following this change (needed for #26064).

Se also the new Note [ModBreaks vs InternalModBreaks]

On i386-linux:

-------------------------
Metric Decrease:
    interpreter_steplocal
-------------------------

bf03bbaa

by Simon Hengel at 2025-08-01T04:39:05-04:00

Don't use MCDiagnostic for `ghcExit`

This changes the error message of `ghcExit` from

```
<no location info>: error:
Compilation had errors

```
to
```

Compilation had errors

```

a889ec75

by Simon Hengel at 2025-08-01T04:39:05-04:00

Respect `-fdiagnostics-as-json` for driver diagnostics (see #24113)

81577fe7

by Ben Gamari at 2025-08-02T04:29:39-04:00

configure: Allow override of CrossCompiling

As noted in #26236, the current inference logic is a bit simplistic. In
particular, there are many cases (e.g. building for a new libc) where
the target and host triples may differ yet we are still able to run the
produced artifacts as native code.

Closes #26236.

01136779
by Andreas Klebinger at 2025-08-02T04:30:20-04:00
```
rts: Support COFF BigObj files in archives.
```

1f9e4f54

by Stephen Morgan at 2025-08-03T15:14:08+10:00

refactor: Modify Data.List.sortOn to use (>) instead of compare. (#26184)

This lets a more efficient (>) operation be used if one exists.

This is technically a breaking change for malformed Ord instances, where
x > y is not equivalent to compare x y == GT.

Discussed by the CLC in issue #332: https://github.com/haskell/core-libraries-committee/issues/332

4f6bc9cf

by fendor at 2025-08-04T17:50:06-04:00

Revert "base: Expose Backtraces constructor and fields"

This reverts commit 17db44c5b32fff82ea988fa4f1a233d1a27bdf57.

bcdec657

by Zubin Duggal at 2025-08-05T10:37:29+05:30

compiler: Export a version of `newNameCache` that is not prone to footguns.
`newNameCache` must be initialized with both a non-"reserved" unique tag, as well
as a list of known key names. Failing to do so results in hard to debug unique conflicts.

It is difficult for API users to tell which unique tags are safe to use. So instead of leaving
this up to the user to decide, we now export a version of `newNameCache` which uses a guaranteed
non-reserved unique tag. In fact, this is now the way the unique tag is initialized for all invocations
of the compiler.

The original version of `newNameCache` is now exported as `newNameCache'` for advanced users.

We also deprecate `initNameCache` as it is also prone to footguns and is completely subsumed in
functionality by `newNameCache` and `newNameCache'`.

Fixes #26135 and #26055

57d3b4a8

by Andrew Lelechenko at 2025-08-05T18:36:31-04:00

hadrian: bump Stackage snapshot to LTS 24.2 / GHC 9.10.2

In line with #25693 we should use GHC 9.10 as a boot compiler,
while Hadrian stack.yaml was stuck on GHC 9.6.

c2a78cea

by Peng Fan at 2025-08-05T18:37:27-04:00

NCG/LA64: implement atomic write with finer-grained DBAR hints

Signed-off-by: Peng Fan <fanpeng@loongson.cn>

95231c8e

by Teo Camarasu at 2025-08-06T08:35:58-04:00

CODEOWNERS: add CLC as codeowner of base

We also remove hvr, since I think he is no longer active

77df0ded
by Andrew Lelechenko at 2025-08-06T08:36:39-04:00
```
Bump submodule text to 2.1.3
```

8af260d0

by Nikolaos Chatzikonstantinou at 2025-08-06T08:37:23-04:00

docs: fix internal import in getopt examples

This external-facing doc example shouldn't mention GHC internals when
using 'fromMaybe'.

69cc16ca
by Marc Scholten at 2025-08-06T15:51:28-04:00
```
README: Add note on ghc.nix
```

93a2f450

by Daniel Díaz at 2025-08-06T15:52:14-04:00

Link to the "Strict Bindings" docs from the linear types docs

Strict Bidings are relevant for the kinds of multiplicity annotations
linear lets support.

246b7853

by Matthew Pickering at 2025-08-07T06:58:30-04:00

level imports: Check the level of exported identifiers

The level imports specification states that exported identifiers have to
be at level 0. This patch adds the requird level checks that all
explicitly mentioned identifiers occur at level 0.

For implicit export specifications (T(..) and module B), only level 0
identifiers are selected for re-export.

ghc-proposal: https://github.com/ghc-proposals/ghc-proposals/pull/705

Fixes #26090

358bc4fc
by fendor at 2025-08-07T06:59:12-04:00
```
Bump GHC on darwin CI to 9.10.1
```

1903ae35

by Matthew Pickering at 2025-08-07T12:21:10+01:00

ipe: Place strings and metadata into specific .ipe section

By placing the .ipe metadata into a specific section it can be stripped
from the final binary if desired.

```
objcopy --remove-section .ipe <binary>
upx <binary>
```

Towards #21766

c80dd91c

by Matthew Pickering at 2025-08-07T12:22:42+01:00

ipe: Place magic word at the start of entries in the .ipe section

The magic word "IPE\nIPE\n" is placed at the start of .ipe sections,
then if the section is stripped, we can check whether the section starts
with the magic word or not to determine whether there is metadata
present or not.

Towards #21766

cab42666

by Matthew Pickering at 2025-08-07T12:22:42+01:00

ipe: Use stable IDs for IPE entries

IPEs have historically been indexed and reported by their address.
This makes it impossible to compare profiles between runs, since the
addresses may change (due to ASLR) and also makes it tricky to separate
out the IPE map from the binary.

This small patch adds a stable identifier for each IPE entry.

The stable identifier is a single 64 bit word. The high-bits are a
per-module identifier and the low bits identify which entry in each
module.

1. When a node is added into the IPE buffer it is assigned a unique
   identifier from an incrementing global counter.
2. Each entry already has an index by it's position in the
   `IpeBufferListNode`.

The two are combined together by the `IPE_ENTRY_KEY` macro.

Info table profiling uses the stable identifier rather than the address
of the info table.

The benefits of this change are:

* Profiles from different runs can be easily compared
* The metadata can be extracted from the binary (via the eventlog for
  example) and then stripped from the executable.

Fixes #21766

2860a9a5

by Simon Peyton Jones at 2025-08-07T20:29:18-04:00

In TcSShortCut, typechecker plugins should get empty Givens

Solving in TcShortCut mode means /ignoring the Givens/. So we
should not pass them to typechecker plugins!

Fixes #26258.

This is a fixup to the earlier MR:

    commit 1bd12371feacc52394a0e660ef9349f9e8ee1c06
    Author: Simon Peyton Jones <simon.peytonjones@gmail.com>
    Date:   Mon Jul 21 10:04:49 2025 +0100

        Improve treatment of SPECIALISE pragmas -- again!

2157db2d

by sterni at 2025-08-08T15:32:39-04:00

hadrian: enable terminfo if --with-curses-* flags are given

The GHC make build system used to support WITH_TERMINFO in ghc.mk which
allowed controlling whether to build GHC with terminfo or not. hadrian
has replaced this with a system where this is effectively controlled by
the cross-compiling setting (the default WITH_TERMINFO value was bassed
on CrossCompiling, iirc).

This behavior is undesireable in some cases and there is not really a
good way to work around it. Especially for downstream packagers,
modifying this via UserSettings is not really feasible since such a
source file has to be kept in sync with Settings/Default.hs manually
since it can't import Settings.Default or any predefined Flavour
definitions.

To avoid having to add a new setting to cfg/system.config and/or a new
configure flag (though I'm happy to implement both if required), I've
chosen to take --with-curses-* being set explicitly as an indication
that the user wants to have terminfo enabled. This would work for
Nixpkgs which sets these flags [1] as well as haskell.nix [2] (which
goes to some extreme measures [3] [4] to force terminfo in all scenarios).

In general, I'm an advocate for making the GHC build be the same for
native and cross insofar it is possible since it makes packaging GHC and
Haskell related things while still supporting cross much less
compilicated. A more minimal GHC with reduced dependencies should
probably be a specific flavor, not the default.

Partially addresses #26288 by forcing terminfo to be built if the user
explicitly passes configure flags related to it. However, it isn't built
by default when cross-compiling yet nor is there an explicit way to
control the package being built.

[1]: https://github.com/NixOS/nixpkgs/blob/3a7266fcefcb9ce353df49ba3f292d06443760bb/pkgs/development/compilers/ghc/common-hadrian.nix#L513-L515
[2]: https://github.com/input-output-hk/haskell.nix/blob/6eaafcdf04bab7be745d1aa4f74d2cc85700042b/compiler/ghc/default.nix#L185
[3]: https://github.com/input-output-hk/haskell.nix/blob/6eaafcdf04bab7be745d1aa4f74d2cc85700042b/compiler/ghc/default.nix#L678-L682
[4]: https://github.com/input-output-hk/haskell.nix/blob/6eaafcdf04bab7be745d1aa4f74d2cc85700042b/compiler/ghc/default.nix#L772-L773

b3c31488

by David Feuer at 2025-08-08T15:33:21-04:00

Add default QuasiQuoters

Add `defaultQuasiQuoter` and `namedDefaultQuasiQuoter` to make it easier
to write `QuasiQuoters` that give helpful error messages when they're
used in inappropriate contexts.

Closes #24434.

03555ed8

by Sylvain Henry at 2025-08-10T22:20:57-04:00

Handle non-fractional CmmFloats in Cmm's CBE (#26229)

Since f8d9d016305be355f518c141f6c6d4826f2de9a2, toRational for Float and
Double converts float's infinity and NaN into Rational's infinity and
NaN (respectively 1%0 and 0%0).

Cmm CommonBlockEliminator hashing function needs to take these values
into account as they can appear as literals now. See added testcase.

6c956af3
by J. Ryan Stinnett at 2025-08-10T22:21:42-04:00
```
Fix extensions list in `DoAndIfThenElse` docs
```

6dc420b1

by J. Ryan Stinnett at 2025-08-10T22:21:42-04:00

Document status of `RelaxedPolyRec` extension

This adds a brief extension page explaining the status of the
`RelaxedPolyRec` extension. The behaviour of this mode is already
explained elsewhere, so this page is mainly for completeness so that
various lists of extensions have somewhere to point to for this flag.

Fixes #18630

18036d52

by Simon Peyton Jones at 2025-08-11T11:31:20-04:00

Take more care in zonkEqTypes on AppTy/AppTy

This patch fixes #26256.

See Note [zonkEqTypes and the PKTI] in GHC.Tc.Solver.Equality

c8d76a29

by Zubin Duggal at 2025-08-11T11:32:02-04:00

ci: upgrade bootstrap compiler on windows to 9.10.1

34fc50c1

by Ben Gamari at 2025-08-11T13:36:25-04:00

Kill IOPort#

This type is unnecessary, having been superceded by `MVar` and a rework
of WinIO's blocking logic.

See #20947.

See https://github.com/haskell/core-libraries-committee/issues/213.

56b32c5a

by sheaf at 2025-08-12T10:00:19-04:00

Improve deep subsumption

This commit improves the DeepSubsumption sub-typing implementation
in GHC.Tc.Utils.Unify.tc_sub_type_deep by being less eager to fall back
to unification.

For example, we now are properly able to prove the subtyping relationship

  ((∀ a. a->a) -> Int) -> Bool  <=  β[tau] Bool

for an unfilled metavariable β. In this case (with an AppTy on the right),
we used to fall back to unification. No longer: now, given that the LHS
is a FunTy and that the RHS is a deep rho type (does not need any instantiation),
we try to make the RHS into a FunTy, viz.

  β := (->) γ

We can then continue using covariance & contravariance of the function
arrow, which allows us to prove the subtyping relationship, instead of
trying to unify which would cause us to error out with:

  Couldn't match expected type ‘β’ with actual type ‘(->) ((∀ a. a -> a) -> Int)

See Note [FunTy vs non-FunTy case in tc_sub_type_deep] in GHC.Tc.Utils.Unify.

The other main improvement in this patch concerns type inference.
The main subsumption logic happens (before & after this patch) in
GHC.Tc.Gen.App.checkResultTy. However, before this patch, all of the
DeepSubsumption logic only kicked in in 'check' mode, not in 'infer' mode.

This patch adds deep instantiation in the 'infer' mode of checkResultTy
when we are doing deep subsumption, which allows us to accept programs
such as:

  f :: Int -> (forall a. a->a)
  g :: Int -> Bool -> Bool

  test1 b =
    case b of
      True  -> f
      False -> g
  test2 b =
    case b of
      True  -> g
      False -> f

See Note [Deeply instantiate in checkResultTy when inferring].

Finally, we add representation-polymorphism checks to ensure that the
lambda abstractions we introduce when doing subsumption obey the
representation polymorphism invariants of Note [Representation polymorphism invariants]
in GHC.Core. See Note [FunTy vs FunTy case in tc_sub_type_deep].
This is accompanied by a courtesy change to `(<.>) :: HsWrapper -> HsWrapper -> HsWrapper`,
adding the equation:

  WpCast c1 <.> WpCast c2 = WpCast (c1 `mkTransCo` c2)

This is useful because mkWpFun does not introduce an eta-expansion when
both of the argument & result wrappers are casts; so this change allows
us to avoid introducing lambda abstractions when casts suffice.

Fixes #26225

d175aff8
by Sylvain Henry at 2025-08-12T10:01:31-04:00
```
Add regression test for #18619
```

a3983a26

by Sylvain Henry at 2025-08-12T10:02:20-04:00

RTS: remove some TSAN annotations (#20464)

Use RELAXED_LOAD_ALWAYS macro instead.

0434af81

by Ben Gamari at 2025-08-12T10:03:02-04:00

Bump time submodule to 1.15

Also required bumps of Cabal, directory, and hpc.

62899117

by Florian Ragwitz at 2025-08-13T21:01:34-04:00

Extend record-selector usage ticking to all binds using a record field

This extends the previous handling of ticking for RecordWildCards and
NamedFieldPuns to all var bindings that involve record selectors.

Note that certain patterns such as `Foo{foo = 42}` will currently not tick the
`foo` selector, as ticking is triggered by `HsVar`s.

Closes #26191.

b37b3af7

by Florian Ragwitz at 2025-08-13T21:01:34-04:00

Add release notes for 9.16.1 and move description of latest HPC changes there.

a5e4b7d9

by Ben Gamari at 2025-08-13T21:02:18-04:00

rts: Clarify rationale for undefined atomic wrappers

Since c06e3f46d24ef69f3a3d794f5f604cb8c2a40cbc the RTS has declared
various atomic operation wrappers defined by ghc-internal as undefined.
While the rationale for this isn't clear from the commit message, I
believe that this is necessary due to the unregisterised backend.
Specifically, the code generator will reference these symbols when
compiling RTS Cmm sources.

50842f83

by Andreas Klebinger at 2025-08-13T21:03:01-04:00

Make unexpected LLVM versions a warning rather than an error.

Typically a newer LLVM version *will* work so erroring out if
a user uses a newer LLVM version is too aggressive.

Fixes #25915

c91e2650

by fendor at 2025-08-13T21:03:43-04:00

Store `StackTrace` and `StackSnapshot` in `Backtraces`

Instead of decoding the stack traces when collecting the `Backtraces`,
defer this decoding until actually showing the `Backtraces`.
This allows users to customise how `Backtraces` are displayed by
using a custom implementation of `displayExceptionWithInfo`, overwriting
the default implementation for `Backtraces` (`displayBacktraces`).

dee28cdd

by fendor at 2025-08-13T21:03:43-04:00

Allow users to customise the collection of exception annotations

Add a global `CollectExceptionAnnotationMechanism` which determines how
`ExceptionAnnotation`s are collected upon throwing an `Exception`.

This API is exposed via `ghc-experimental`.

By overriding how we collect `Backtraces`, we can control how the
`Backtraces` are displayed to the user by newtyping `Backtraces` and
giving a different instance for `ExceptionAnnotation`.

A concrete use-case for this feature is allowing us to experiment with
alternative stack decoders, without having to modify `base`, which take
additional information from the stack frames.

This commit does not modify how `Backtraces` are currently
collected or displayed.

66024722

by fendor at 2025-08-13T21:03:43-04:00

Expose Backtraces internals from ghc-experimental

Additionally, expose the same API `base:Control.Exception.Backtrace`
to make it easier to use as a drop-in replacement.

a766286f

by Reed Mullanix at 2025-08-13T21:04:36-04:00

ghc-internal: Fix naturalAndNot for NB/NS case

When the first argument to `naturalAndNot` is larger than a `Word` and the second is `Word`-sized, `naturalAndNot` will truncate the
result:

```
>>> naturalAndNot ((2 ^ 65) .|. (2 ^ 3)) (2 ^ 3)
0
```

In contrast, `naturalAndNot` does not truncate when both arguments are larger than a `Word`, so this appears to be a bug.
Luckily, the fix is pretty easy: we just need to call `bigNatAndNotWord#` instead of truncating.

Fixes #26230

3506fa7d

by Simon Hengel at 2025-08-13T21:05:18-04:00

Report -pgms as a deprecated flag

(instead of reporting an unspecific warning)

Before:

    on the commandline: warning:
        Object splitting was removed in GHC 8.8

After:

    on the commandline: warning: [GHC-53692] [-Wdeprecated-flags]
        -pgms is deprecated: Object splitting was removed in GHC 8.8

51c701fe

by Zubin Duggal at 2025-08-13T21:06:00-04:00

testsuite: Be more permissive when filtering out GNU_PROPERTY_TYPE linker warnings

The warning text is slightly different with ld.bfd.

Fixes #26249

dfe6f464

by Simon Hengel at 2025-08-13T21:06:43-04:00

Refactoring: Don't misuse `MCDiagnostic` for lint messages

`MCDiagnostic` is meant to be used for compiler diagnostics.

Any code that creates `MCDiagnostic` directly, without going through
`GHC.Driver.Errors.printMessage`, side steps `-fdiagnostics-as-json`
(see e.g. !14475, !14492 !14548).

To avoid this in the future I want to control more narrowly who creates
`MCDiagnostic` (see #24113).

Some parts of the compiler use `MCDiagnostic` purely for formatting
purposes, without creating any real compiler diagnostics.  This change
introduces a helper function, `formatDiagnostic`, that can be used in
such cases instead of constructing `MCDiagnostic`.

a8b2fbae

by Teo Camarasu at 2025-08-13T21:07:24-04:00

rts: ensure MessageBlackHole.link is always a valid closure

We turn a MessageBlackHole into an StgInd in wakeBlockingQueue().
Therefore it's important that the link field, which becomes the
indirection field, always points to a valid closure.

It's unclear whether it's currently possible for the previous behaviour
to lead to a crash, but it's good to be consistent about this invariant nonetheless.

Co-authored-by: Andreas Klebinger <klebinger.andreas@gmx.at>

4021181e

by Teo Camarasu at 2025-08-13T21:07:24-04:00

rts: spin if we see a WHITEHOLE in messageBlackHole

When a BLACKHOLE gets cancelled in raiseAsync, we indirect to a THUNK.
GC can then shortcut this, replacing our BLACKHOLE with a fresh THUNK.

This THUNK is not guaranteed to have a valid indirectee field.

If at the same time, a message intended for the previous BLACKHOLE is
processed and concurrently we BLACKHOLE the THUNK, thus temporarily
turning it into a WHITEHOLE, we can get a segfault, since we look at the
undefined indirectee field of the THUNK

The fix is simple: spin if we see a WHITEHOLE, and it will soon be
replaced with a valid BLACKHOLE.

Resolves #26205

1107af89

by Oleg Grenrus at 2025-08-13T21:08:06-04:00

Allow defining HasField instances for naughty fields

Resolves #26295

... as HasField solver doesn't solve for fields with "naughty"
selectors, we could as well allow defining HasField instances for these
fields.

020e7587
by Sylvain Henry at 2025-08-13T21:09:00-04:00
```
Fix Data.List unqualified import warning
```

fd811ded

by Simon Peyton Jones at 2025-08-14T17:56:47-04:00

Make injecting implicit bindings into its own pass

Previously we were injecting "impliicit bindings" (data constructor
worker and wrappers etc)

  - both at the end of CoreTidy,
  - and at the start of CorePrep

This is unpleasant and confusing.  This patch puts it it its own pass,
addImplicitBinds, which runs between the two.

The function `GHC.CoreToStg.AddImplicitBinds.addImplicitBinds` now takes /all/
TyCons, not just the ones for algebraic data types.  That change ripples
through to
     - corePrepPgm
     - doCodeGen
     - byteCodeGen
All take [TyCon] which includes all TyCons

9bd7fcc5

by Simon Peyton Jones at 2025-08-14T17:56:47-04:00

Implement unary classes

The big change is described exhaustively in

   Note [Unary class magic] in GHC.Core.TyCon

Other changes

* We never unbox class dictionaries in worker/wrapper.  This has been true for some
  time now, but the logic is now centralised in functions in
  GHC.Core.Opt.WorkWrap.Utils, namely `canUnboxTyCon`, and `canUnboxArg`
  See Note [Do not unbox class dictionaries] in GHC.Core.Opt.WorkWrap.Utils.

* Refactored the `notWorthFloating` logic in GHc.Core.Opt.SetLevels.
  I can't remember if I actually changed any behaviour here, but if so it's
  only in a corner cases.

* Fixed a bug in `GHC.Core.TyCon.isEnumerationTyCon`, which was wrongly returning
  True for (##).

* Remove redundant Role argument to `liftCoSubstWithEx`.  It was always
  Representational.

* I refactored evidence generation in the constraint solver:
  * Made GHC.Tc.Types.Evidence contain better abstactions for evidence
    generation.
  * I deleted the file `GHC.Tc.Types.EvTerm` and merged its (small) contents
    elsewhere. It wasn't paying its way.
  * Made evidence for implicit parameters go via a proper abstraction.

* Fix inlineBoringOk; see (IB6) in Note [inlineBoringOk]
  This fixes a slowdown in `countdownEffectfulDynLocal`
  in the `effectful` library.

Smaller things

* Rename `isDataTyCon` to `isBoxedDataTyCon`.
* GHC.Core.Corecion.liftCoSubstWithEx was only called with Representational role,
  so I baked that into the function and removed the argument.
* Get rid of `GHC.Core.TyCon.tyConSingleAlgDataCon_maybe` in favour of calling
  `not isNewTyCon` at the call sites; more explicit.
* Refatored `GHC.Core.TyCon.isInjectiveTyCon`; but I don't think I changed its
  behaviour
* Moved `decomposeIPPred` to GHC.Core.Predicate

Compile time performance changes:
  geo. mean        +0.1%
    minimum        -6.8%
    maximum       +14.4%

The +14% one is in T21839c, where it seems that a bit more inlining
is taking place.  That seems acceptable; and the average change is small

Metric Decrease:
    LargeRecord
    T12227
    T12707
    T16577
    T21839r
    T5642
Metric Increase:
    T15164
    T21839c
    T3294
    T5321FD
    T5321Fun
    WWRec

b4075d71

by Simon Peyton Jones at 2025-08-14T17:56:47-04:00

Slight improvement to pre/postInlineUnconditionally

Avoids an extra simplifier iteration

9e443596

by Simon Peyton Jones at 2025-08-14T17:56:47-04:00

Fix a long-standing assertion error in normSplitTyConApp_maybe

91310ad0
by Simon Peyton Jones at 2025-08-14T17:56:47-04:00
```
Add comment to coercion optimiser
```

5b841d82

by Teo Camarasu at 2025-08-14T17:57:56-04:00

template-haskell: move some identifiers from ghc-internal to template-haskell

These identifiers are not used internally by the compiler. Therefore we
have no reason for them to be in ghc-internal.

By moving them to template-haskell, we benefit from it being easier to
change them and we avoid having to build them in stage0.

Resolves #26048

33e2c7e5

by Teo Camarasu at 2025-08-14T17:57:56-04:00

template-haskell: transfer $infix note to public module

This Haddock note should be in the public facing module

2a411fc4

by Sylvain Henry at 2025-08-14T17:59:09-04:00

JS: export HEAP8 symbol (#26290)

Newer Emscripten requires this.

248f78ca

by Ben Gamari at 2025-08-14T17:59:51-04:00

users-guide: Drop the THREAD_RUNNABLE event

As of f361281c89fbce42865d8b8b27b0957205366186 it is no longer emitted.

706d33e3

by Recursion Ninja at 2025-08-15T04:12:12-04:00

Resolving issues #20645 and #26109

Correctly sign extending and casting smaller bit width types for LLVM operations:

  - bitReverse8#
  - bitReverse16#
  - bitReverse32#
  - byteSwap16#
  - byteSwap32#
  - pdep8#
  - pdep16#
  - pext8#
  - pext16#

1cdc6f46

by Cheng Shao at 2025-08-15T04:12:56-04:00

hadrian: enforce have_llvm=False for wasm32/js

This patch fixes hadrian to always pass have_llvm=False to the
testsuite driver for wasm32/js targets. These targets don't really
support the LLVM backend, and the optllvm test way doesn't work. We
used to special-case wasm32/js to avoid auto-adding optllvm way in
testsuite/config/ghc, but this is still problematic if someone writes
a new LLVM-related test and uses something like when(have_llvm(),
extra_ways(["optllvm"])). So better just enforce have_llvm=False for
these targets here.

ca03226d
by Ben Gamari at 2025-08-18T13:43:20+00:00
```
configure: Allow use of LLVM 20
```

783cd7d6

by Cheng Shao at 2025-08-18T20:13:14-04:00

compiler: use `UniqMap` instead of `Map` for `BCEnv` in bytecode compiler

The bytecode compiler maintains a `BCEnv` which was previously `Map Id
StackDepth`. Given `Id` is `Uniquable`, we might as well use `UniqMap`
here as a more efficient data structure, hence this patch.

Co-authored-by: Codex <codex@openai.com>

58e46da9

by fendor at 2025-08-18T20:13:56-04:00

rts: Strip lower three bits when hashing Word instead of lower eight bits

45dbfa23

by Cheng Shao at 2025-08-18T20:14:37-04:00

libffi: update to 3.5.2

Bumps libffi submodule.

54be78ef

by Ben Gamari at 2025-08-19T16:28:05-04:00

testsuite: Fix T20006b

This test is supposed to fail for non-threaded ways yet it
was previously marked as only failing in `normal`.
Fix this.

f4bac607

by Simon Peyton Jones at 2025-08-19T16:28:47-04:00

Take yet more care with reporting redundant constraints

This small patch fixes #25992, which relates to reporting redundant
constraints on default-method declarations.

See (TRC5) in Note [Tracking redundant constraints]

ab130fec

by fendor at 2025-08-19T16:29:29-04:00

Bump dependencies of hadrian-bootstrap-gen to use GHC 9.6.7

6d02ac6f

by fendor at 2025-08-19T16:29:29-04:00

Bump required GHC version for test-bootstrap jobs to 9.10.1

Include test-bootstrap job for GHC 9.12.2.

Update hadrian bootstrap plans use GHC 9.10 and 9.12

Remove older GHC bootstrap configurations.
We require at least GHC 9.10.1 to build GHC.

Adds plans for:

* 9.10.1
* 9.10.2
* 9.12.1
* 9.12.2

9e857171

by Brandon Chinn at 2025-08-20T11:47:46-04:00

Don't warn unused-imports with used generated imports

Fixes #21730

* The old notion of "implicit" import has been renamed to "generated". See Note [Generated imports] in GHC.Hs.ImpExp.

* ImportMap now keeps track of generated and user-written imports separately. This avoids the fake SrcSpan we used to give the implicit Prelude import, and the hack that went with it.

* -ddump-minimal-imports now considers generated imports (but still only
  warns on + prints user-written imports)

* bestImport considers generated imports to take priority over user-written imports.

9fb3bad4

by Ben Gamari at 2025-08-20T11:48:31-04:00

mailmap: Use ben@well-typed.com more liberally

Nearly all of this work was done while working for Well-Typed.

774fec37

by Ben Gamari at 2025-08-20T11:49:15-04:00

Add primop to annotate the call stack with arbitrary data

We introduce a new primop `annotateStack#` which allows us to push
arbitrary data onto the call-stack.
This allows us to extract the data later when decoding the stack, for
example when an exception is thrown, showing more information to the
user without having to annotate the full call-stack with `HasCallStack`
constraints.

A new stack frame value is introduced `AnnFrame`, which consists of
nothing but a generic payload.

The primop has a small wrapper API that allows users to annotate their
call-stack in programs.
There is a pure API and an IO-based one. The former is a little bit
dubious, as it affects the evaluation of a program, so use with care.
The latter is "safe", as it doesn't change the evaluation of the
program.

The stack annotation mechanism is similarly implemented to the
`ExceptionAnnotation` and `Exception`, there is a typeclass to indicate
something can be pushed onto the call-stack and all values are wrapped
in the existential `SomeStackAnnotation`, which recover the type of the
annotation payload.

There is currently no builtin way to show the stack annotations when
`Backtraces` are displayed (i.e., when showing stack traces to the user),
which we will address in a follow-up MR.

-------------------------
Metric Increase:
    ghc_experimental_so
-------------------------

We increase the size of the package, so this is not unreasonable.

Co-Authored-By: fendor <fendor@posteo.de>
Co-Authored-By: Ben Gamari <bgamari.foss@gmail.com>

fdfa3892
by Ben Gamari at 2025-08-20T11:49:57-04:00
```
testsuite: Add regression test for #24606
```

39b2e382

by Cheng Shao at 2025-08-20T11:50:40-04:00

compiler: only use `Name` instead of `Id` in `SptEntry`

As a part of #26298, this patch refactors `SptEntry` to only carry a
`Name` instead of `Id`: we do not care about extra information like
caffyness or type at all in any static pointer related codegen logic.
This is necessary to make `SptEntry` serializable, as a part of the
grand plan of serializable bytecode.

Co-authored-by: Codex <codex@openai.com>

276f8ea8
by Vekhir -- at 2025-08-20T11:51:35-04:00
```
Bump Cabal dependency
```

0b9c7437

by Zubin Duggal at 2025-08-20T11:52:18-04:00

ci: Teach ci.sh to fetch FreeBSD artifacts from ghcup unofficial bindists and bootstrap compiler on FreeBSD to 9.10.1

Also refactor fetch_ghc logic in ci.sh, renaming the GHC_VERSION enviorment configuration variable to FETCH_GHC_VERSION,
making it clear that it is intended for use on platforms like Windows and FreeBSD where we don't want to use the GHC
excecutable from the platform environment and instead need to download and install GHC-$FETCH_GHC_VERSION from a release
bindist.

Fixes #26296

b2914797

by Cheng Shao at 2025-08-20T11:53:00-04:00

driver: use UniqSet for hiddenModules in DynFlags/FinderOpts

This patch replaces Set ModuleName with UniqSet ModuleName in
DynFlags.hiddenModules and FinderOpts.finder_hiddenModules for
improved efficiency.

Co-authored-by: Codex <codex@openai.com>

0335d899

by Cheng Shao at 2025-08-20T11:53:00-04:00

driver: use UniqMap ModuleName in the finder

This patch replaces Map ModuleName with UniqMap ModuleName in the
finder for improved efficiency.

Co-authored-by: Codex <codex@openai.com>

91f4faaa

by Cheng Shao at 2025-08-20T11:53:43-04:00

configure: check python3 version and require minimal 3.7

Since !9515, the testsuite driver requires python3 version to be at
least 3.7, though this has never been checked by configure logic. This
patch implements the version check. Fixes #23234.

Co-authored-by: Codex <codex@openai.com>

df4ee9b4

by Cheng Shao at 2025-08-20T11:54:25-04:00

compiler: use zero cost coerce in GHC.CmmToAsm.CFG.loopInfo

This patch refactors GHC.CmmToAsm.CFG.loopInfo to use zero cost coerce
and thus addresses the TODO. For coerce to work, constructors of
Label/LabelMap/LabelSet from GHC.Cmm.Dataflow.Label are exposed,
though I believe it's a worthy tradeoff to avoid unnecessary runtime
cost without using unsafeCoerce, since the latter could be a landmine
for future refactoring.

Co-authored-by: Codex <codex@openai.com>

ccda188d

by Simon Peyton Jones at 2025-08-20T11:55:07-04:00

Start with empty inerts in shortcut solving

When short-cut solving we were starting with an inert set that had
unsolved Wanteds.  This caused an infinite loop (#26314), because a
typechecker plugin kept being given that unsolved Wanted.

It's better just to start with an empty inert set

c8882ed7

by Ben Gamari at 2025-08-20T11:55:49-04:00

configure: Bump minimal bootstrap GHC version to 9.8

f0a19d74

by fendor at 2025-08-20T19:55:00-04:00

Remove deprecated functions from the ghci package

ebeb991b

by fendor at 2025-08-20T19:55:00-04:00

base: Remove unstable heap representation details from GHC.Exts

e368e247

by Rodrigo Mesquita at 2025-08-20T19:55:42-04:00

bytecode: Use 32bits for breakpoint index

Fixes #26325

42724462

by Simon Hengel at 2025-08-21T17:52:11-04:00

Serialize wired-in names as external names when creating HIE files

Note that the domain of de-serialized names stays the same.
Specifically, for known-key names, before `lookupKnownKeyName` was used,
while now this is handled by `lookupOrigNameCache` which captures the
same range provided that the OrigNameCache has been initialized with
`knownKeyNames` (which is the case by default).

(fixes #26238)

6a43f8ec

by Cheng Shao at 2025-08-21T17:52:52-04:00

compiler: fix closure C type in SPT init code

This patch fixes the closure C type in SPT init code to StgClosure,
instead of the previously incorrect StgPtr. Having an incorrect C type
makes SPT init code not compatible with other foreign stub generation
logic, which may also emit their own extern declarations for the same
closure symbols and thus will clash with the incorrect prototypes in
SPT init code.

7268ec00

by Sven Tennie at 2025-08-25T11:41:58+02:00

Add CPU vector support (RVV) detection for RISC-V

The strategy is to try an operation that requires vector support and be
prepared for an illegal instruction signal.

29bfb99c