-
5fab2238
by Wolfgang Jeltsch at 2026-05-12T21:24:27+03:00
Introduce a cache of home module name providers
This contribution introduces to the module graph a cache that maps home
module names to sets of units providing them and changes the finder to
use that cache. This is a performance optimization, especially for
multi-home-unit builds.
The particular changes are as follows:
* In `GHC.Unit.Module.Graph`, `ModuleGraph` is extended with a new
field `mg_home_module_name_providers_map`, exposed as
`mgHomeModuleNameProvidersMap`. This is a cache that assigns to each
home module name the set of IDs of home units that define it.
Operations that construct module graphs are updated such that this
cache stays synchronized.
* In `GHC.Unit.Finder`, `findImportedModule` is changed to pull
`mgHomeModuleNameProvidersMap` from `hsc_mod_graph` and pass it to
`findImportedModuleNoHsc`, which now does not search home units in
arbitrary order but prioritizes those units that the cache mentions
as potential providers of the requested module.
In addition, this contribution adds variants of the two multi-component
compiler performance tests that use 100 units instead of 20, because
with just 20 units the benefits from caching of home module name
providers are still negligible.
The following table shows the total time needed for running both
multi-component tests before and after this contribution and with
different numbers of units:
| # of units | Before | After |
|-----------:|-------:|------:|
| 20 | 0:12 | 0:12 |
| 100 | 0:47 | 0:42 |
| 200 | 3:05 | 2:08 |
Note that there seems to be a general overhead of 12 seconds that is not
attributable to the actual tests, so that the real running times should
be 12 seconds smaller than shown above.
Resolves #27055.
Metric Decrease:
MultiComponentModules
MultiComponentModulesRecomp
Co-authored-by: Matthew Pickering <matthewtpickering@gmail.com>
Co-authored-by: Fendor <fendor@posteo.de>
-
38b76b2f
by Cheng Shao at 2026-05-13T17:48:48-04:00
testsuite: mark T22159 as fragile
This patch marks T22159 as fragile on Windows for issue described in #27248.
Before we get to the bottom of those failures, this unblocks newer
Windows runners.
-
50188615
by Ian Duncan at 2026-05-14T13:45:07+02:00
AArch64: use ASR not LSR for MO_U_Shr at W8/W16
The unsigned right shift (MO_U_Shr) for sub-word widths (W8, W16)
with a variable shift amount was emitting ASR (arithmetic/signed shift
right) after zero-extending with UXTB/UXTH. This should be LSR
(logical/unsigned shift right). After zero-extension the upper bits
happen to be 0 so ASR produces the same result, but it is semantically
wrong and would break if the zero-extension were ever optimized away.
Includes assembly output test (grep for lsr) and runtime test
verifying unsigned right shift of Word8 and Word16 values.
-
28666fbf
by Vladislav Zavialov at 2026-05-19T12:44:05-04:00
Add type families: Tuple, Constraints, Tuple#, Sum# (#27179)
These type families map tuples of types to the corresponding Tuple<N>,
Tuple<N>#, CTuple<N>, and Sum<N># types. Some examples at N=2:
Tuple (Int, Bool) = Tuple2 Int Bool
Constraints (Show a, Eq a) = CTuple2 (Show a) (Eq a)
Tuple# (Int#, Float#) = Tuple2# Int# Float#
Sum# (Int#, Float#) = Sum2# Int# Float#
See GHC Proposal #145 "Non-punning list and tuple syntax".
To make the Sum# instance at N=64 possible, this patch also introduces
the Sum64# constructor declaration and bumps mAX_SUM_SIZE from 63 to 64.
Metric Increase:
ghc_experimental_dir
-
41c2448b
by Wen Kokke at 2026-05-19T12:44:53-04:00
rts: Add IPE event class for -l
This commit adds a new IPE event class to the -l RTS flag.
Previously, IPE events were enabled unconditionally.
However, the IPE events can easily grow to hundreds or thousands of megabytes.
With the new event class you can pass, e.g., -l-I to disable IPE events.
-
62536551
by Wen Kokke at 2026-05-19T12:44:53-04:00
ghc-internal: Add TraceFlags.traceIPE
-
e45312d1
by Wen Kokke at 2026-05-19T12:44:53-04:00
testsuite: Add test for TraceFlags.traceIpe
-
4768d9aa
by Wen Kokke at 2026-05-19T12:44:53-04:00
ghc-internal: Add DebugFlags.ipe
-
bc1b5c69
by Wen Kokke at 2026-05-19T12:44:53-04:00
testsuite: Add test for DebugFlags.ipe
-
0da1543f
by Duncan Coutts at 2026-05-19T12:45:37-04:00
Document removal of the signal-based interval timer
Update mentions within the RTS section of the users guide.
Add a changelog entry.
-
b2911514
by Duncan Coutts at 2026-05-19T12:45:37-04:00
Fix section for an recent changelog entry
-
d6d76a7a
by David Eichmann at 2026-05-19T12:46:19-04:00
ghc-toolchain: implement llvm program versioning logic
-
2dd36fa3
by Wolfgang Jeltsch at 2026-05-20T04:49:52-04:00
Turn `Trustworthy` into `Safe` in `base` where possible
-
f4399dd1
by Wolfgang Jeltsch at 2026-05-20T04:50:37-04:00
Make the current `base` buildable with GHC 10.0
-
1a7de232
by Duncan Coutts at 2026-05-20T12:26:25-04:00
Hadrian: remove legacy rts .so symlinks
For compatibility with the old makefile based build system, hadrian had
rules to generate symlinks from unversioned to versioned names for the
rts .so/.dynlib file, like libHSrts-ghcx.y.so -> libHSrts-1.0.3-ghcx.y.so
We no longer need these symlinks since the makefile build system has
been retired some time ago. The need for these symlinks is awkward on
windows where we cannot (in practice) create symlinks. So rather than
make them conditional (non-windows), just remove them entirely.
-
286f1adf
by fendor at 2026-05-20T12:27:09-04:00
Fix regression T27202: `:load` and `:add` work in GHCi
To fix the regression there are conceptually two major things that we
fix:
* We don't remove the `importDirs` from `interactive-session`
* When `:add`ing a module, we don't try to find them via PackageImports
* The PackageImport is wrong as we can't know the package-name at
this stage in ghc/UI.hs
What does it mean to not remove the `importDirs` from
`interactive-session`?
It means that, given some initial `DynFlags`, we will use those
`importDirs` in `interactive-session`.
The initial `DynFlags`, however, depend on how you initialise the GHC
session.
For a simple session, initialised by
ghc -isrc -this-unit-id main
It is simple, just use the `DynFlags` given on the cli.
Thus, `main` and `interactive-session` will have the same `DynFlags`,
except for the `homeUnitId` and `interactive-session` depends on `main`
by construction of the GHCi session.
What about a multiple home unit session, though?
ghc -unit @unit1 -unit @unit2
What are the `DynFlags` in this cli invocation? It shouldn't be either
`@unti1` nor `@unit2`, as the order shouldn't matter or any other
implicit condition.
For consistency, we decide that the initial `DynFlags` are the top
`DynFlags` on the cli, ignoring `-unit` flags.
Thus, in this example, there are no `importsDirs` regardless of what we
might find in `@unit1` and `@unit2`.
But in this invocation:
ghc -isrc -unit @unit1 -unit @unit2
The `interactive-session` will have the `importsDirs` `src`.
Note, `-isrc` will be inherited in `@unit1` and `@unit2`, so you need to
explicitly use `-i` to clear the `importsDirs`, in order to avoid
accidentally adding `src` as an import directory to all other home
units.
This fix has been made possible by the improvements introduced in
!15888, which avoids ambiguity when a home unit shares the `importsDirs`
with the `interactive-session`, on top of being much faster for multiple
home units.
Adds regression tests for T27202 for `:load`ing and `:add`ing modules
that are located in import directories.
-
728662de
by fendor at 2026-05-20T12:27:09-04:00
Use home unit package db stacks in GHCi prompt and session unit
In order to import modules from home unit dependencies (e.g., `Data.Map`),
the ghci prompt unit needs to populate its `UnitState`.
This is tricky to handle correctly, which `PackageDBFlag`s should we use
to populate the `UnitState`?
We decide, the most intuitive solution for users is to depend on all
`PackageDBFlag`s, so that any dependency can be imported in GHCi.
This assumes consistency in the `PackageDBFlag`s, so no two home units
specify `PackageDBFlag`s that are inconsistent with each other.
We could simply concat all the `PackageDBFlag`s of the existing home
units, but later `PackageDBFlag`s shadow earlier ones, leading to the
last processed home units' `PackageDBFlag`s to shadow the earlier ones.
This is hard to fix, we need to give users the capability to provide ghc
options for the ghci prompt home unit.
However, as this is considerably more work, we decided on an
approximation that should work out most of the time.
Package Db stacks in cabal and stack follow a certain structure:
-no-user-package-db > -package-db $cabal-store > -package-db $local-db
The first two arguments are always the same, namely the
`-no-user-package-db` and `-package-db`.
We compute the longest common prefix over all home units, and use that
as the start of the package db stack. Then, over the rest of the
`PackageDBFlag`s, we simply take the union and append them to our
initial stack.
We assume, that the rest of package dbs only defines very few, "local"
units that are usually not shadowing each other.
This allows us to get a relatively consistent package database stack for
the ghci prompt home unit.
Similar reasoning applies to the session unit in order to add modules to
the session and have dependencies available in the module.
We do something similar for `-package` flags, to make sure only the
correct units are actually visible in the ghci session.
This time, we simply take the union of all `PackageFlag`s, allowing us
to import modules from the home unit dependencies.
In the future, it would be beneficial to allow the user to provide the
exact ghc options to control the visibilities. For now, this will have
to do.
-
740d89a0
by Simon Peyton Jones at 2026-05-20T17:20:44-04:00
Do not use mkCast during typechecking
This commit fixes #27219. The problem was that the typechecker was using
`mkCast`, whose assertion checks legitimately fail when applied to types
that contain unification variables.
-
a50fdb06
by Simon Peyton Jones at 2026-05-20T17:20:45-04:00
Major refactor of the Simplifier
The main payload of this patch is to refactor the Simplifer to avoid
repeated simplification when using Plan (AFTER) for rule rewrites.
The need for this was shown up by #26989.
See Note [Avoid repeated simplification] in GHC.Core.Opt.Simplify.Iteration.
Related refactoring:
* Refactor the two fields `sc_dup` and `sc_env` in `ApplyToVal` into one, `sc_env`.
Reason: the envt is irrelevant in the "simplified" case, so the data type describes
the possiblitiies much more accurately now.
* Some refactoring in `knownCon` to split off `wrapDataConFloats`.
* Refactor `lookupRule` and its auxiliary functions to return `RuleMatch`,
a new data type. See Note [data RuleMatch] in GHC.Core. Ditto for BuiltinRule.
This RuleMatch returns fragments of the target in rm_args and rm_floats,
leaving `rm_rhs` to be the stuff from the RULE itself.
Doing this has routine consequences in GHC.Core.Opt.ConstantFold. Many changes
there but all routine.
* When doing occurrence analysis on RULEs, make the occ-info on the rule
binders relate just to the RHS, not the LHS. See (OUR1) in
Note Note [OccInfo in unfoldings and rules]
This means that Lint must not complain about the fact that the patterns
in the RULE mentions binders that are marked dead.
See Note [Dead occurrences] in GHC.Core.Lint.
I changed the Core pretty-printer so that it didn't suppress dead binders,
else I can't see those binders in RULEs. That led to quite a lot of testsuite wibbles.
* Refactor FloatBinds, so that it is used both by
`exprIsConApp_mabye` and by `lookupRule`
* Move the definition of FloatBinds out of GHc.Core.Make, into GHC.Core.
* Add FloatTick as an extra constructor.
* Refactor `lookupRule` to use `FloatBinds` instead of `BindWrapper`.
This refactor just shares more code.
(Rename GHC.Core.Opt.FloatOut.FloatBinds to FloatLets, to avoid gratuitious
name clash with GHC.Core.FloatBinds.)
Corecion optimisation
* In simpleOpt, when composing coercions, call new function `optTransCo`.
This is much lighter weight than full blown coercion optimisation.
* Make `GHC.Core.Opt.Arity.pushCoValArg` and `pushCoTyArg` return the
coercionLKind of the coercion. This saves recomputing that coercionLKind
at the key call sites in GHC.Core.Opt.Simplify.Iteration.pushCast.
* Rename `addCoerce` in GHC.Core.Simplify.Iteration to become `pushCast`.
* In the `ApplyToVal` case of `pushCast` we had a very unsavoury call to `simplArg`.
I eliminated it by adding a field `sc_cast` to `ApplyToVal` that records any
pending casts. Much nicer now. See Note [The sc_cast field of ApplyToVal].
* Don't optimise coercions if the type-substitution is empty.
See Note [Optimising coercions] in GHC.Core.Opt.Simplify.Iteration.
The fix for #26838 is dramatic. For the test in perf/compiler/T26839 we have
Compiler allocs: Before: 7,363M
After: 688M
Compile time goes down generally. Here are compiler-alloc changes
over 0.5%:
CoOpt_Read(normal) 729,184,920 -0.7%
CoOpt_Singletons(normal) 666,916,960 -4.6% GOOD
LargeRecord(normal) 1,227,056,876 +1.1%
T12227(normal) 256,827,604 -4.6% GOOD
T12425(optasm) 76,879,410 -0.8%
T12545(normal) 787,826,918 -10.8% GOOD
T12707(normal) 775,186,464 -0.9%
T13253(normal) 318,599,596 -0.8%
T14766(normal) 685,857,320 -1.0%
T15304(normal) 1,123,333,422 -2.2%
T15630(normal) 123,142,330 -2.6%
T15630a(normal) 123,092,100 -2.6%
T15703(normal) 299,751,682 -2.9% GOOD
T17516(normal) 964,072,280 +1.0%
T18223(normal) 367,016,820 -6.2% GOOD
T18730(optasm) 130,643,770 -3.3% GOOD
T20261(normal) 535,608,584 -0.7%
T21839c(normal) 340,340,436 -0.9%
T24984(normal) 85,568,392 -1.9%
T3064(normal) 174,631,992 -1.2%
T3294(normal) 1,215,886,432 -0.7%
T5030(normal) 141,449,704 -17.2% GOOD
T5321Fun(normal) 258,484,744 -1.9%
T8095(normal) 770,532,232 -2.7%
T9630(normal) 858,423,408 -14.5% GOOD
T9872c(normal) 1,591,709,448 +0.7%
info_table_map_perf(normal) 19,700,614,458 -1.3%
geo. mean -0.7%
minimum -17.2%
maximum +1.1%
However, strangely there seems to be a 5.0% increase in CoOpt_Read in
the x86_64-linux-fedora43-validate+debug_info+ubsan job, although
there generally a /decrease/ in this test in other builds. The baseline
value looks strange. Anyway I'll just accept it.
Metric Decrease:
CoOpt_Singletons
T12227
T12545
T12707
T15703
T18223
T18730
T21839c
T5030
T9630
Metric Increase:
CoOpt_Read
-
834623d4
by Mrjtjmn at 2026-05-20T17:21:41-04:00
users-guide: Fix weird notation in "Summary of stolen syntax"
-
6f9d7c71
by Markus Läll at 2026-05-21T15:25:34-04:00
Use "grimily" instead of "grimly"
Fixes https://gitlab.haskell.org/ghc/ghc/-/issues/27221
-
50e999ca
by fendor at 2026-05-21T15:26:18-04:00
Speed up 'closure' computation in `ghc-pkg`
Cache the set of already seen `UnitId`s and use `Set` operations to
speed up 'closure' computation.
Further simplify the implementation of 'closure' to account for the
actual usage.
As a consequence, we rename 'closure' to 'brokenPackages' to reflect its
purpose better after the simplification.
-
7ecc6184
by sheaf at 2026-05-21T15:27:10-04:00
TcMPluginHandling: be more lenient when no plugins
This change ensures that, if a function such as 'typecheckModule' was
invoked with 'NoTcMPlugins', GHC doesn't spuriously complain about TcM
plugins having already been stopped, as there were none to start with.
-
89e7a6ba
by Wolfgang Jeltsch at 2026-05-23T15:38:37+03:00
Allow `downsweep` to use nodes of an existing module graph
To this end, `downsweep` has not been able to use the nodes of a module
graph obtained from a previous downsweeping round. In some GHC API
applications, downsweeping is performed somewhat incrementally and
therefore could profit from reusing such existing results. This
contribution makes this possible.
Resolves #27054.
Co-authored-by: Matthew Pickering <matthewtpickering@gmail.com>