-
d47bf776
by Matthew Pickering at 2025-04-14T16:44:41+01:00
driver: Use ModuleGraph for oneshot and --make mode
This patch uses the `hsc_mod_graph` field for both oneshot and --make
mode. Therefore, if part of the compiler requires usage of the module
graph, you do so in a uniform way for the two different modes.
The `ModuleGraph` describes the relationship between the modules in the
home package and units in external packages. The `ModuleGraph` can be
queried when information about the transitive closure of a package is
needed. For example, the primary use of the ModuleGraph from within the
compiler is in the loader, which needs to know the transitive closure of
a module so it can load all the relevant objects for evaluation.
In --make mode, downsweep computes the ModuleGraph before any
compilation starts.
In oneshot mode, a thunk is created at the start of compilation, which
when forced will compute the module graph beneath the current module.
The thunk is only forced at the moment when the user uses Template
Haskell.
Finally, there are some situations where we need to discover what
dependencies to load but haven't loaded a module graph at all. In this
case, there is a fallback which computes the transitive closure on the
fly and doesn't cache the result. Presumably if you are going to call
getLinkDeps a lot, you would compute the right ModuleGraph before you
started.
Importantly, this removes the ExternalModuleGraph abstraction. This was quite
awkward to work with since it stored information about the home package
inside the EPS.
This patch will also be very useful when implementing explicit level
imports, which requires more significant use of the module graph in
order to determine which level instances are available at.
Towards #25795
-------------------------
Metric Decrease:
MultiLayerModulesTH_Make
MultiLayerModulesTH_OneShot
-------------------------
-
395e0ad1
by sheaf at 2025-04-16T12:33:26-04:00
base: remove .Internal modules (e.g. GHC.TypeLits)
This commit removes the following internal modules from base,
as per CLC proposal 217:
- GHC.TypeNats.Internal
- GHC.TypeLits.Internal
- GHC.ExecutionStack.Internal
Fixes #25007
-
e0f3ff11
by Patrick at 2025-04-17T04:31:12-04:00
Refactor Handling of Multiple Default Declarations
Fixes: #25912, #25914, #25934
Previously, GHC discarded all loaded defaults (tcg_default) when local
defaults were encountered during typechecking. According to the
exportable-named-default proposal (sections 2.4.2 and 2.4.3), local
defaults should be merged into tcg_default, retaining any defaults
already present while overriding where necessary.
Key Changes:
* Introduce DefaultProvenance to track the origin of default declarations
(local, imported, or built-in), replacing the original cd_module
in ClassDefaults with cd_provenance :: DefaultProvenance.
* Rename tcDefaults to tcDefaultDecls, limiting its responsibility to only
converting renamed class defaults into ClassDefaults.
* Add extendDefaultEnvWithLocalDefaults to merge local defaults into the
environment, with proper duplication checks:
- Duplicate local defaults for a class trigger an error.
- Local defaults override imported or built-in defaults.
* Update and add related notes: Note [Builtin class defaults],
Note [DefaultProvenance].
* Add regression tests: T25912, T25914, T25934.
Thanks sam and simon for the help on this patch.
Co-authored-by: sheaf <sam.derbyshire@gmail.com>
-
386f1854
by Teo Camarasu at 2025-04-17T04:31:55-04:00
template-haskell: Remove `addrToByteArrayName` and `addrToByteArray`
These were part of the implementation of the `Lift ByteArray` instance and were errornously exported because this module lacked an explicit export list. They have no usages on Hackage.
Resolves #24782
-
b96e2f77
by Sylvain Henry at 2025-04-18T20:46:33-04:00
RTS: remove target info and fix host info (#24058)
The RTS isn't a compiler, hence it doesn't have a target and we remove
the reported target info displayed by "+RTS --info". We also fix the
host info displayed by "+RTS --info": the host of the RTS is the
RTS-building compiler's target, not the compiler's host (wrong when
doing cross-compilation).
-
6d9965f4
by Sylvain Henry at 2025-04-18T20:46:33-04:00
RTS: remove build info
As per the discussion in !13967, there is no reason to tag the RTS with
information about the build platform.
-
d52e9b3f
by Vladislav Zavialov at 2025-04-18T20:47:15-04:00
Diagnostics: remove the KindMismatch constructor (#25957)
The KindMismatch constructor was only used as an intermediate
representation in pretty-printing.
Its removal addresses a problem detected by the "codes" test case:
[GHC-89223] is untested (constructor = KindMismatch)
In a concious deviation from the usual procedure, the error code
GHC-89223 is removed entirely rather than marked as Outdated.
The reason is that it never was user-facing in the first place.
-
e2f2f9d0
by Vladislav Zavialov at 2025-04-20T10:53:39-04:00
Add name for -Wunusable-unpack-pragmas
This warning had no name or flag and was triggered unconditionally.
Now it is part of -Wdefault.
In GHC.Tc.TyCl.tcTyClGroupsPass's strict mode, we now have to
force-enable this warning to ensure that detection of flawed groups
continues to work even if the user disables the warning with the
-Wno-unusable-unpack-pragmas option. Test case: T3990c
Also, the misnamed BackpackUnpackAbstractType is now called
UnusableUnpackPragma.
-
6caa6508
by Adam Gundry at 2025-04-20T10:54:22-04:00
Fix specialisation of incoherent instances (fixes #25883)
GHC normally assumes that class constraints are canonical, meaning that
the specialiser is allowed to replace one dictionary argument with another
provided that they have the same type. The `-fno-specialise-incoherents`
flag alters INCOHERENT instance definitions so that they will prevent
specialisation in some cases, by inserting `nospec`.
This commit fixes a bug in 7124e4ad76d98f1fc246ada4fd7bf64413ff2f2e, which
treated some INCOHERENT instance matches as if `-fno-specialise-incoherents`
was in effect, thereby unnecessarily preventing specialisation. In addition
it updates the relevant `Note [Rules for instance lookup]` and adds a new
`Note [Canonicity for incoherent matches]`.
-
0426fd6c
by Adam Gundry at 2025-04-20T10:54:23-04:00
Add regression test for #23429
-
eec96527
by Adam Gundry at 2025-04-20T10:54:23-04:00
user's guide: update specification of overlapping/incoherent instances
The description of the instance resolution algorithm in the user's
guide was slightly out of date, because it mentioned in-scope given
constraints only at the end, whereas the implementation checks for
their presence before any of the other steps.
This also adds a warning to the user's guide about the impact of
incoherent instances on specialisation, and more clearly documents
some of the other effects of `-XIncoherentInstances`.
-
a00eeaec
by Matthew Craven at 2025-04-20T10:55:03-04:00
Fix bytecode generation for `tagToEnum# <LITERAL>`
Fixes #25975.
-
2e204269
by Andreas Klebinger at 2025-04-22T12:20:41+02:00
Simplifier: Constant fold invald tagToEnum# calls to bottom expr.
When applying tagToEnum# to a out-of-range value it's best to simply
constant fold it to a bottom expression. That potentially allows more
dead code elimination and makes debugging easier.
Fixes #25976
-
7250fc0c
by Matthew Pickering at 2025-04-22T16:24:04-04:00
Move -fno-code note into Downsweep module
This note was left behind when all the code which referred to it was
moved into the GHC.Driver.Downsweep module
-
d2dc89b4
by Matthew Pickering at 2025-04-22T16:24:04-04:00
Apply editing notes to Note [-fno-code mode] suggested by sheaf
These notes were suggested in https://gitlab.haskell.org/ghc/ghc/-/merge_requests/14241
-
91564daf
by Matthew Pickering at 2025-04-24T00:29:02-04:00
ghci: Use loadInterfaceForModule rather than loadSrcInterface in mkTopLevEnv
loadSrcInterface takes a user given `ModuleName` and resolves it to the
module which needs to be loaded (taking into account module
renaming/visibility etc).
loadInterfaceForModule takes a specific module and loads it.
The modules in `ImpDeclSpec` have already been resolved to the actual
module to get the information from during renaming. Therefore we just
need to fetch the precise interface from disk (and not attempt to rename
it again).
Fixes #25951
-
2e0c07ab
by Simon Peyton Jones at 2025-04-24T00:29:43-04:00
Test for #23298
-
0eef99b0
by Sven Tennie at 2025-04-24T07:34:36-04:00
RV64: Introduce J instruction (non-local jumps) and don't deallocate stack slots for J_TBL (#25738)
J_TBL result in local jumps, there should not deallocate stack slots
(see Note [extra spill slots].)
J is for non-local jumps, these may need to deallocate stack slots.
-
1bd3d13e
by fendor at 2025-04-24T07:35:17-04:00
Add `UnitId` to `EvalBreakpoint`
The `EvalBreakpoint` is used to communicate that a breakpoint was
encountered during code evaluation.
This `EvalBreakpoint` needs to be converted to an `InternalBreakpointId`
which stores a `Module` to uniquely find the correct `Module` in the
Home Package Table.
The `EvalBreakpoint` used to store only a `ModuleName` which is then
converted to a `Module` based on the currently active home unit.
This is incorrect in the face of multiple home units, the break point
could be in an entirely other home unit!
To fix this, we additionally store the `UnitId` of the `Module` in
`EvalBreakpoint` to later reconstruct the correct `Module`
All of the changes are the consequence of extending `EvalBreakpoint`
with the additional `ShortByteString` of the `UnitId`.
For performance reasons, we store the `ShortByteString` backing the
`UnitId` directly, avoiding marshalling overhead.
-
fe6ed8d9
by Sylvain Henry at 2025-04-24T18:04:12-04:00
Doc: add doc for JS interruptible calling convention (#24444)
-
6111c5e4
by Ben Gamari at 2025-04-24T18:04:53-04:00
compiler: Ensure that Panic.Plain.assertPanic' provides callstack
In 36cddd2ce1a3bc62ea8a1307d8bc6006d54109cf @alt-romes removed CallStack
output from `GHC.Utils.Panic.Plain.assertPanic'`. While this output is
redundant due to the exception backtrace proposal, we may be
bootstrapping with a compiler which does not yet include this machinery.
Reintroduce the output for now.
Fixes #25898.
-
217caad1
by Matthew Pickering at 2025-04-25T18:58:42+01:00
Implement Explicit Level Imports for Template Haskell
This commit introduces the `ExplicitLevelImports` and
`ImplicitStagePersistence` language extensions as proposed in GHC
Proposal #682.
Key Features
------------
- `ExplicitLevelImports` adds two new import modifiers - `splice` and
`quote` - allowing precise control over the level at which imported
identifiers are available
- `ImplicitStagePersistence` (enabled by default) preserves existing
path-based cross-stage persistence behavior
- `NoImplicitStagePersistence` disables implicit cross-stage
persistence, requiring explicit level imports
Benefits
--------
- Improved compilation performance by reducing unnecessary code generation
- Enhanced IDE experience with faster feedback in `-fno-code` mode
- Better dependency tracking by distinguishing compile-time and runtime dependencies
- Foundation for future cross-compilation improvements
This implementation enables the separation of modules needed at
compile-time from those needed at runtime, allowing for more efficient
compilation pipelines and clearer code organization in projects using
Template Haskell.
Implementation Notes
--------------------
The level which a name is availble at is stored in the 'GRE', in the normal
GlobalRdrEnv. The function `greLevels` returns the levels which a specific GRE
is imported at. The level information for a 'Name' is computed by `getCurrentAndBindLevel`.
The level validity is checked by `checkCrossLevelLifting`.
Instances are checked by `checkWellLevelledDFun`, which computes the level an
instance by calling `checkWellLevelledInstanceWhat`, which sees what is
available at by looking at the module graph.
Modifications to downsweep
--------------------------
Code generation is now only enabled for modules which are needed at
compile time.
See the Note [-fno-code mode] for more information.
Uniform error messages for level errors
---------------------------------------
All error messages to do with levels are now reported uniformly using
the `TcRnBadlyStaged` constructor.
Error messages are uniformly reported in terms of levels.
0 - top-level
1 - quote level
-1 - splice level
The only level hard-coded into the compiler is the top-level in
GHC.Types.ThLevelIndex.topLevelIndex.
Uniformly refer to levels and stages
------------------------------------
There was much confusion about levels vs stages in the compiler.
A level is a semantic concept, used by the typechecker to ensure a
program can be evaluated in a well-staged manner.
A stage is an operational construct, program evaluation proceeds in
stages.
Deprecate -Wbadly-staged-types
------------------------------
`-Wbadly-staged-types` is deprecated in favour of `-Wbadly-levelled-types`.
Lift derivation changed
-----------------------
Derived lift instances will now not generate code with expression
quotations.
Before:
```
data A = A Int deriving Lift
=>
lift (A x) = [| A $(lift x) |]
```
After:
```
lift (A x) = conE 'A `appE` (lift x)
```
This is because if you attempt to derive `Lift` in a module where
`NoImplicitStagePersistence` is enabled, you would get an infinite loop
where a constructor was attempted to be persisted using the instance you
are currently defining.
GHC API Changes
---------------
The ModuleGraph now contains additional information about the type of
the edges (normal, quote or splice) between modules. This is abstracted
using the `ModuleGraphEdge` data type.
Fixes #25828
-------------------------
Metric Increase:
MultiLayerModulesTH_Make
-------------------------
-
7641a74a
by Simon Peyton Jones at 2025-04-26T22:05:19-04:00
Get a decent MatchContext for pattern synonym bindings
In particular when we have a pattern binding
K p1 .. pn = rhs
where K is a pattern synonym. (It might be nested.)
This small MR fixes #25995. It's a tiny fix, to an error message,
removing an always-dubious `unkSkol`.
The bug report was in the context of horde-ad, a big program,
and I didn't manage to make a small repro case quickly. I decided
not to bother further.
-
ce616f49
by Simon Peyton Jones at 2025-04-27T21:10:25+01:00
Fix infelicities in the Specialiser
On the way to #23109 (unary classes) I discovered some infelicities
(or maybe tiny bugs, I forget) in the type-class specialiser.
I also tripped over #25965, an outright bug in the rule matcher
Specifically:
* Refactor: I enhanced `wantCallsFor`, whih previously always said
`True`, to discard calls of class-ops, data constructors etc. This is
a bit more efficient; and it means we don't need to worry about
filtering them out later.
* Fix: I tidied up some tricky logic that eliminated redundant
specialisations. It wasn't working correctly. See the expanded
Note [Specialisations already covered], and
(MP3) in Note [Specialising polymorphic dictionaries].
See also the new top-level `alreadyCovered`
function, which now goes via `GHC.Core.Rules.ruleLhsIsMoreSpecific`
I also added a useful Note [The (CI-KEY) invariant]
* Fix #25965: fixed a tricky bug in the `go_fam_fam` in
`GHC.Core.Unify.uVarOrFam`, which allows matching to succeed
without binding all type varibles.
I enhanced Note [Apartness and type families] some more
* #25703. This ticket "just works" with -fpolymorphic-specialisation;
but I was surprised that it worked! In this MR I added documentation
to Note [Interesting dictionary arguments] to explain; and tests to
ensure it stays fixed.
-
22d11fa8
by Simon Peyton Jones at 2025-04-28T18:05:19-04:00
Track rewriter sets more accurately in constraint solving
The key change, which fixed #25440, is to call `recordRewriter` in
GHC.Tc.Solver.Rewrite.rewrite_exact_fam_app. This missing call meant
that we were secretly rewriting a Wanted with a Wanted, but not really
noticing; and that led to a very bad error message, as you can see
in the ticket.
But of course that led me into rabbit hole of other refactoring around
the RewriteSet code:
* Improve Notes [Wanteds rewrite Wanteds]
* Zonk the RewriterSet in `zonkCtEvidence` rather than only in GHC.Tc.Errors.
This is tidier anyway (e.g. de-clutters debug output), and helps with the
next point.
* In GHC.Tc.Solver.Equality.inertsCanDischarge, don't replace a constraint
with no rewriters with an equal constraint that has many. See
See (CE4) in Note [Combining equalities]
* Move zonkRewriterSet and friends from GHC.Tc.Zonk.Type into
GHC.Tc.Zonk.TcType, where they properly belong.
A handful of tests get better error messages.
For some reason T24984 gets 12% less compiler allocation -- good
Metric Decrease:
T24984
-
cb5ed017
by Simon Peyton Jones at 2025-04-29T11:27:58+01:00
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
-
3a62ad7e
by Simon Peyton Jones at 2025-04-29T11:29:29+01:00
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.
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