-
69561925
by sheaf at 2026-07-20T14:25:36+05:30
Don't drop ticks around variables of type `IO ()`
GHC.Core.Utils.mkTick is responsible for placing a tick on a Core
expression. It contains logic for dropping SCCs (non-counting profiling
ticks) around non-function variables, as such variables cannot
meaningfully contribute to profiles. However, the logic for what counts
as a function was incorrect: it used `isFunTy` which returns 'False' for
types such as 'IO ()' where the function arrow is hidden under a
newtype.
We now use 'mightBeFunTy' instead of 'isFunTy'. This ensures we don't
drop ticks in cases we aren't sure.
On the way, we improve the documentation of 'isFunTy', 'isPiTy' and
'mightBeFunTy', and update the latter's implementation to consistently
handle unary classes.
Fixes #27225
-------------------------
Metric Decrease:
T5642
-------------------------
(cherry picked from commit ce01ccb625514a09e76aded549691da4dfe87de7)
-
d035085a
by sheaf at 2026-07-20T14:25:43+05:30
Fix crash in Data.Data instance for HsCtxt
The Data.Data instance for HsCtxt contained an error for the 'toConstr'
method, which could trigger for example when looking at -ddump-tc-ast
traces. Replace it with the 'abstractConstr' pattern used in the rest of
the codebase.
(cherry picked from commit 98c20394e8a130df928272c360199fb5e9265f22)
-
4a70cf55
by sheaf at 2026-07-20T14:25:43+05:30
Fix AArch64 clobbering bug for MUL2
On AArch64, the code generator could clobber one of the input operands
when computing the lower bits of a MUL2 operation. This rendered invalid
the subsequent computation of the high bits.
This commit fixes that by using a temporary register. The register
allocator can remove the redundant move in the common case when the
registers do not conflict.
Fixes #27046
(cherry picked from commit c9015f0953e72829e89ac768b6ad9ece34c7e187)
-
f17ccdf3
by sheaf at 2026-07-20T14:25:43+05:30
Add type family performance test for #26426
Some GHC versions produced large numbers of coercions after typechecking
and desugaring when compiling the program in #26426:
Version | Typechecker time | Typechecker allocations | Coercions
-------:|-----------------:|------------------------:|---------:
9.6 | 47 ms | 48 MB | 110k
9.8 | 1000 ms | 486 MB | 10,437k
9.10 | 922 ms | 489 MB | 10,436k
9.12 | 906 ms | 482 MB | 10,437k
9.14 | 63 ms | 55 MB | 333k
10.0 | 47 ms | 64 MB | 35k
The improvement 9.12 -> 9.14 was due to commit 22d11fa818fae2c95c494fc0fac1f8cb4c6e7cb6,
while the improvement 9.14 -> 10.0 was due to commit 0b7df6db9e46df40e86fbff1a66dc10440b99db5.
As the behaviour of GHC seems better than it's ever been on this program,
we declare victory, adding this performance test to ensure we don't
regress on this program.
On the way, we update Note [Combining equalities] in GHC.Tc.SolveR.Equality
with the explanation of the 9.12 -> 9.14 improvement (getting rid of an
exponential blowup in coercion sizes), and we update
Note [Exploiting closed type families] in GHC.Tc.Solver.FunDeps with
the explanation of the 9.14 -> 10.0 improvement (bringing down coercion
size growth from cubic to quadratic).
(cherry picked from commit 40764930f10d73f6ca00dfbbce908f9516d262ce)
-
eca8e6eb
by mangoiv at 2026-07-20T14:25:48+05:30
libraries/process: bump submodule to v1.6.30.0
- bump the submodule to the appropriate tag
- suppress benign warning resulting from the change
(cherry picked from commit d9ea2d76545452a7df567b162340079cb024a40c)
-
7aba67c2
by ARATA Mizuki at 2026-07-20T14:25:57+05:30
RISC-V NCG: Zero-extend the result of castFloatToWord32
According to the ISA manual, FMV.X.W sign-extends the result.
We need to truncate the result to avoid creating an exotic Word32 value.
Fixes #27300
(cherry picked from commit 291ce3aafe4ca3d2562154a28c595a858987d9f1)
-
9dba4f96
by ARATA Mizuki at 2026-07-20T14:25:57+05:30
RISC-V NCG: Treat d28-d31 (ft8-ft11) as caller-saved
According to the calling convention, the registers d28-d31 (ft8-ft11) are caller-saved.
Fixes #27306
(cherry picked from commit 011be91fdaa7869dff2f30b8f2ecca3c9a713739)
-
4e7a3d23
by ARATA Mizuki at 2026-07-20T14:25:57+05:30
RISC-V NCG: Set rounding mode when emitting `truncate`
If we omit the rounding mode for `fcvt`, `dyn` will be used.
We do not want that for `truncate`, so we set `rtz`.
In other places, we set `rne` because we do not use the dynamic rounding mode.
Fixes #27303
(cherry picked from commit e8a547133031b7de8f6f9bbd70a7148eda0941ee)
-
b9d78ac1
by Zubin Duggal at 2026-07-20T14:25:57+05:30
rts: fix validate build with gcc 16. `__attribute__((regparm(1)))` is ignored on x86_64 and now
gcc warns that it is ignored:
rts/sm/Evac.h:35:1: error:
error: ‘regparm’ attribute ignored [-Werror=attributes]
See https://gcc.gnu.org/git/?p=gcc.git;a=commit;h=ccead81bbc39668376eb5cf47066acb446cc43f3
Fixes #27366
(cherry picked from commit 9438bec7117433bbc70449d4288f6ee32dbbbec4)
-
dfd1dd1b
by Ian Duncan at 2026-07-20T14:25:58+05:30
AArch64: use SXTH, not SXTW, for W32 signExtendReg
signExtendReg was using SXTH (sign-extend halfword, 16-bit) for
W32-to-W64 sign extension. This should be SXTW (sign-extend word,
32-bit). SXTH only sign-extends the lower 16 bits, producing wrong
results for 32-bit values whose bit 15 differs from bit 31.
Other fixes:
- At sub-W64, code gen for MO_S_Mul2 should use W32 registers for
SMULL source operands as per the ARM spec (SMULL Xd, Wn, Wm),
and not W64.
- Ensure signExtendReg uses the source width for the source operand
in SXTW/SXTH/SXTB instructions. GNU as requires sxtw Xd,Wn (not
sxtw Xd,Xn), while LLVM's integrated assembler on macOS is lenient.
- Fix overflow flag computation for `MO_S_Mul2`. The overflow bit
was exactly inverted for sub-W64 operands.
Fixes #26978 and #27047
(cherry picked from commit 636c1c7ae47495f022affa501ea0a40cb55cd4a4)
-
36de60c2
by Luite Stegeman at 2026-07-20T14:25:58+05:30
tag inference: don't confuse functions with their return values
inferTagRhs was mixing up taggedness for closures and return values
for function closures. We really shouldn't assign TagTuple to a
properly tagged function returning a tuple.
We fix this by keeping track of functions (TagFun) separately from
values (TagVal) and keeping track of their return value. TagFun is
also used for join points.
fixes #27005
(cherry picked from commit 7fe4f2ec3ce12ea138177c81a019dcfc148fe5d4)
-
caac944f
by Sebastian Graf at 2026-07-20T14:26:25+05:30
Desugar a `case` scrutinee only once (#27383, #20251)
In `dsExpr` for `HsCase` we desugared the scrutinee /twice/: once to
build the Core `case` itself, and again inside `matchWrapper`, which
re-desugared the source scrutinee (via `addHsScrutTmCs`) purely to
record long-distance information for the pattern-match checker.
For a single `case` that is merely wasteful. But for nested cases it
is catastrophic. Consider
case (case (case e of ... ) of ... ) of ...
Desugaring the outer scrutinee desugars the middle `case` twice, each
of which desugars the inner `case` twice, and so on. The work doubles
at every level, so desugaring takes O(2^n) time in the nesting depth.
That is the blowup reported in #27383; it is also what makes the
machine-generated program in #20251 take an age to compile.
The fix is simple. `matchWrapper` is handed the scrutinee anyway, so
we give it the Core expression we have /already/ desugared, and record
the long-distance term constraint with `addCoreScrutTmCs` instead of
re-desugaring from source. This is just what `matchSinglePatVar`
already does for single-pattern matches.
So:
* `matchWrapper` now takes `Maybe [CoreExpr]` rather than
`Maybe [LHsExpr GhcTc]`.
* The `HsCase` equation of `dsExpr` passes the already-desugared
`core_discrim`; the arrow desugarer passes its match variables.
* `addHsScrutTmCs` had no other use, so it is gone.
Desugaring is now linear in the nesting depth. (The coverage checker
still runs `simpleOptExpr` over each scrutinee, which leaves the total
at O(n^2); that is ample.) The long-distance information itself is
unchanged: the checker sees precisely the Core that backs the
generated code.
Test: deSugar/should_compile/T27383
(cherry picked from commit 67d41299be96798702377e6e2b826f9c8e070821)
-
26c22c6e
by sheaf at 2026-07-20T14:26:26+05:30
Avoid mkTick in Core Prep breaking ANF (part II)
Hotfix for 2f9579765f55b3920ceb2e04995ff41a9d0e2d4e fixing a small
oversight in the call to tickTickedExpr from mkTick, in which we
improperly recursively called mkTick without passing on the preserve_anf
flag.
Fixes #27386
(cherry picked from commit 473b97ebc742305f56e30d5b1bbf95b7681312f0)
-
4872ef31
by Simon Hengel at 2026-07-20T15:39:29+05:30
Reference correct package in error messages for reexported modules
(fixes #27417)
(cherry picked from commit a805b2a25021606b30d250e084d4beecbfac0d0a)
-
9f158de1
by mangoiv at 2026-07-20T15:39:29+05:30
compiler: rename ZonkAny to UnusedType and add pretty printing logic
ZonkAny is a hard to understand name for users who do not know how the
compiler works internally. Additionally, it is confusing that ZonkAny,
while being a concrete type *represents* a meta variable, espeically in
the compiler output.
This patch changes the name of ZonkAny to UnusedType which is closer to
its intended semantics and adds special pretty printing logic to display
this type in the same fashion the compiler displays meta variables in
other places, whenever they leak from the implementation to the user.
It also exports the type from ghc-internal:GHC.Internal.Types in order
to expose documentation.
Fixes #27390
Co-Authored-By: Sam Derbyshire <sam.derbyshire@gmail.com>
(cherry picked from commit a72ff58fa86172b1557e405f7246d27914fdae6e)
-
7f19139e
by Luite Stegeman at 2026-07-20T15:39:29+05:30
rts: handle large AP closures in compacting GC
The function update_fwd_large in the compacting GC could run into
an unexpected object with the following error:
internal error: update_fwd_large: unknown/strange object 24
Closure type 24 is the AP closure, which was not handled in
upd_fwd_large. This patch adds handling them.
fixes #27434
(cherry picked from commit cca0d58963f802a8b2e43aa2dbc58592f8ad07bb)
-
34a825d4
by Cheng Shao at 2026-07-20T15:39:29+05:30
compiler: fix missing handling of CmmUnsafeForeignCall node in LayoutStack
This patch fixes missing handling of `CmmUnsafeForeignCall` middle
node in the `LayoutStack` pass.
Before proc-points splitting, this pass computes liveliness of local
registers, and spills those alive across a Cmm native call onto the
stack. It need to traverse all middle nodes in each block and check
whether a local register is an assignee, if so then the previous
mapping in `sm_regs` is invalidated and needs to be dropped. However,
it didn't handle `CmmUnsafeForeignCall` node which may also assign to
a local register. When proc-points splitting is enabled, this can
produce an invalid basic block that doesn't properly backup the
updated local register to the stack before doing a Cmm call, resulting
in completely invalid runtime behavior.
The patch also adds a `T27447` regression test. With no-TNTC or with
LLVM backend, without the fix the test case would output a stale
0x1111111111111111 value, instead of the expected 0x2222222222222222
output.
Fixes #27447.
Co-authored-by: Codex <codex@openai.com>
(cherry picked from commit 3f00f234d0d5b3b3b2a23a5dc70ce372eb9bbdb4)
-
835ed6f2
by mangoiv at 2026-07-20T15:39:29+05:30
ci: retry fetching test metrics
Retry fetching test metrics to make the CI not fail if the services is
temporarily unavailable
(cherry picked from commit b7e24044fde064cb3f0d44c36872a86d024cd7d4)
-
89a634fc
by Zubin Duggal at 2026-07-20T15:39:29+05:30
Bump semaphore-compat submodule to 2.0.1
This versions includes some cruicial fixes for darwin
(cherry picked from commit 4180af3f71754472dbd49b85179b25fd29bd9998)
-
e5b1fab6
by Zubin Duggal at 2026-07-20T15:39:29+05:30
CorePrep: Don't speculatively evaluate bindings that we have already discovered to be absent
In #25924, we segfault because speculation forces a projection out of a RUBBISH dictionary
(which we generated because it absent).
Solution: Don't speculate on bindings we already know are absent.
Fixes 25924
(cherry picked from commit 9b714c4c833461c621f0a050680848d7248aa57e)
-
ebe99a26
by Zubin Duggal at 2026-07-20T15:39:29+05:30
Don't make absent fillers for terminating types
In #25924 we discovered that we could speculatively evaluate an absent filler
for a dictionary, and project a field (a superclass selector) out of it,
resulting in segfaults.
Solution: Never make an absent filler or rubbish literal for a terminating type
like a dictionary. mkAbsentFiller returns Nothing for isTerminatingType, so
worker/wrapper and the specialiser keep the real argument instead.
Some small metric decreases because we do a little less work in the
simplifier now.
Metric Decrease:
T9872a
T9872b
T9872c
TcPlugin_RewritePerf
(cherry picked from commit 4a59b3eece9b7106fcbe73d2d06a49755be4ea8f)
-
2d4a6f31
by Andreas Klebinger at 2026-07-20T15:39:29+05:30
Fix a profiling race condition resulting in segfaults.
StgToCmm: Don't assume tagged FUN closures in closureCodeBody.
When entering a closure the self/node pointer might not be tagged in
some situations when a thunk is evaluated by multiple threads.
So we most AND away the tag bits rather than subtracting an expected tag.
Apply.cmm: Fix a race condition occuring when a thunk is mutated during GC.
In stg_ap_0_fast when might need to run GC before entering a thunk. If this happens
another thread or the GC itself might mutate the closure making entering it no longer
valid. We now check for this.
Add test and changelog for #27123 fixes.
(cherry picked from commit ed09895d7de1ca116a561868c151fd825a16ad0c)
-
5775b5ca
by Cheng Shao at 2026-07-20T15:39:29+05:30
ghc-heap: fix invalid srtlen returned by peekItbl when no-TNTC
This patch fixes the no-TNTC code path of `peekItbl` so that it looks
at the right memory address when reading the `srt` field from the
`StgInfoTable_` struct. Also adds a `T27465` regression test that
reproduces the bug on no-TNTC builds before the fix. Fixes #27465.
Co-authored-by: Codex <codex@openai.com>
(cherry picked from commit 67c03eb2c762fdfeb646eb8345341173dd4268b2)
-
cb19b6c5
by Andrea Vezzosi at 2026-07-20T15:39:29+05:30
[Fix #27287] preserve ModBreaks in ModIface
(cherry picked from commit 4396a6f2a4c7799908e1e0b88a218a51d063fdca)
-
e401ef1c
by Zubin Duggal at 2026-07-20T15:39:29+05:30
hadrian: Remove old package.conf files when generating new ones
Old package.conf files might exists with different hashes, causing issues like #26661
Fixes #26661
(cherry picked from commit 5ac9ce7d3821c75e2d2cc17fae00b9e865a02987)
-
7f50e6d6
by Cheng Shao at 2026-07-20T15:39:29+05:30
compiler: fix miscompiled %load_relaxed, add missing %store_relaxed
This patch fixes the %load_relaxed cmm primop compilation logic to
correctly use relaxed memory ordering, and adds the missing
%store_relaxed primop. Parsing logic of %load/%store with explicit
ordering is covered in the AtomicFetch test case. Fixes #27483.
Co-authored-by: Codex <codex@openai.com>
(cherry picked from commit eee8ec5b25ef0f83ba4822e7a0a941df7b0bec5f)
-
d184ea5d
by Cheng Shao at 2026-07-20T15:39:29+05:30
rts: fix missing UNTAG in stg_readTVarIOzh
This patch fixes missing UNTAG on the current value closure read from
StgTVar. UNTAG is a no-op when it's stg_TREC_HEADER_info which is word
aligned; it may be a tagged closure, and reading info table from the
tagged address is an unaligned load which may cause issues on
platforms with strict alignment requirements.
Co-authored-by: Codex <codex@openai.com>
(cherry picked from commit d377e83e51d39a06e1f0bf2e35a923a3210b21a2)
-
2036f996
by Cheng Shao at 2026-07-20T15:39:29+05:30
rts: fix missing UNTAG in stg_control0zh_ll
This patch fixes missing UNTAG on the cont closure returned by
captureContinuationAndAbort. In case it's not NULL,
captureContinuationAndAbort returns a tagged StgContinuation closure,
in which case it must be untagged before accessing the
apply_mask_frame field.
In the past it worked out of luck: when apply_mask_frame was NULL then
mask_frame_offset is also 0 so the control flow didn't diverge to a
wrong path. Still, this is horribly wrong and will crash once
StgContinuation struct is refactored and fields are shuffled around.
Co-authored-by: Codex <codex@openai.com>
(cherry picked from commit 8ed038421a20e3e4e681973b2f5098e5bd2144b5)
-
3b82fd7f
by Cheng Shao at 2026-07-20T15:39:29+05:30
compiler: fix redundant AP thunk codegen when not using -ticky-ap-thunk
This patch fixes a double negation confusion in !7525 that results in
some redundant AP thunk code generation when not using
-ticky-ap-thunk. Now, we use `stgToCmmUseStdApThunk` to indicate
whether precomputed AP thunks in the RTS should be used, which
defaults to `True`, unless `-ticky-ap-thunk` is passed.
`-finfo-table-map` now also implies `-ticky-ap-thunk`, since when
doing IPE profiling we want the generated AP thunks to be unique.
Fixes #27502.
-------------------------
Metric Decrease:
T3064
-------------------------
Co-authored-by: Codex <codex@openai.com>
(cherry picked from commit 5aa7000ae246ae6a706437799338295b5801a629)
-
c25a2d92
by Brian McKenna at 2026-07-20T15:39:29+05:30
Strip ticks when desugaring bool guards
The special `considerAccessible` pattern was broken when compiling
with debug info. Compiling with debug info wraps expressions with
`SourceNote` ticks, which broke the internals of the
`desugarBoolGuard` function. Ticks are now ignored within this
function.
Fixes #27360
(cherry picked from commit d43a7b7ad90047f1445c88d796ee7b15a0a605d7)
-
42778740
by Ben Gamari at 2026-07-20T15:39:29+05:30
base: Display ExceptionContext in WhileHandling's textual description
As originally-implemented the implementation for
`WhileHandling(displayExceptionAnnotation)` would display the
`ExceptionContext` of the exception which it carries (as this was the
behavior of `displayException`, in terms of which
`displayExceptionAnnotation` was implemented).
However, in 284ffab3 the definition of `SomeException(displayException)`
was changed to exclude the `ExceptionContext`. This means that
`WhileHandling(displayExceptionAnnotation)` fails to describe the
provenance of the exception which it captures, greatly limiting its
utility.
Return the implementation to its originally-specified behavior by
implementing `WhileHandling(displayExceptionAnnotation)` in terms of
`displayExceptionWithInfo`.
Fixes #27456.
(cherry picked from commit ede4b17bdc54c055e44ad95711409c7d1dcb1128)
-
cd1fdc26
by Brian McKenna at 2026-07-20T15:39:29+05:30
Ignore ticks in the pattern-match term oracle
The term-oracle in the pattern-match checker is keyed by a canonical
form of the scrutinee, computed by `makeDictsCoherent`. That canonical
form was tick-sensitive: two occurrences of an otherwise identical
expression that happened to carry different ticks were treated as
distinct values, breaking long-distance information.
This shows up in practice under `-finfo-table-map`, because the
desugarer wraps every record-selector use site in a `SourceNote`
carrying that site's span. For example:
data Box = Box { unBox :: Maybe Int }
f b = case unBox b of
Nothing -> 0
Just _ -> let Just x = unBox b in x
The two `unBox b` expressionss carry different SourceNote spans, the
pattern-match checker sees them as different, the long-distance
information from the outer `Just _` branch never reaches the
let-pattern, and `Just x = unBox b` is wrongly reported as
non-exhaustive.
We now strip all ticks in `makeDictsCoherent`. This is documented as
Wrinkle (UD1) of Note [Unique dictionaries in the TmOracle CoreMap].
Fixes #27314
(cherry picked from commit b388d0934c0933c1cdb47226489634e229239d0a)
-
2a27443f
by Duncan Coutts at 2026-07-20T15:39:29+05:30
Fix failing test GcStaticPointers for non-moving GC
Minor mistake in asserting something before checking for that same
thing.
Specifically, Bdescr asserts HEAP_ALLOCED_GC, but Bdescr was being used
prior to a guard that checks HEAP_ALLOCED_GC. The solution is just to
move the use of Bdescr after the guard.
Thanks to Simon Jakobi for identifying the problem.
(cherry picked from commit 42918646169c7c5ba61912b39b5f5147bf1d255c)