| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| | |
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This introduces three new modules:
- basicTypes/Predicate.hs describes predicates, moving
this logic out of Type. Predicates don't really exist
in Core, and so don't belong in Type.
- typecheck/TcOrigin.hs describes the origin of constraints
and types. It was easy to remove from other modules and
can often be imported instead of other, scarier modules.
- typecheck/Constraint.hs describes constraints as used in
the solver. It is taken from TcRnTypes.
No work other than module splitting is in this patch.
This is the first step toward homogeneous equality, which will
rely more strongly on predicates. And homogeneous equality is the
next step toward a dependently typed core language.
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
In #17343 we saw that we didn't handle the pattern guard `!_ <-
undefined` correctly: The `undefined` was never evaluated. Indeed,
elaboration failed to insert the invisible type aruments to `undefined`.
So `undefined` was trivially a normal-form and in turn never entered.
The problem is that we used to infer a sigma-type for the RHS of the
guard, the leading qualifiers of which will never be useful in a pattern
match situation. Hence we infer a rho-type now.
Fixes #17343.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
For backends maintaining the CFG during codegen
we can now find loops and their nesting level.
This is based on the Cmm CFG and dominator analysis.
As a result we can estimate edge frequencies a lot better
for methods, resulting in far better code layout.
Speedup on nofib: ~1.5%
Increase in compile times: ~1.9%
To make this feasible this commit adds:
* Dominator analysis based on the Lengauer-Tarjan Algorithm.
* An algorithm estimating global edge frequences from branch
probabilities - In CFG.hs
A few static branch prediction heuristics:
* Expect to take the backedge in loops.
* Expect to take the branch NOT exiting a loop.
* Expect integer vs constant comparisons to be false.
We also treat heap/stack checks special for branch prediction
to avoid them being treated as loops.
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Issue #17355 occurred because the control flow for
`TcValidity.check_valid_inst_head` was structured in such a way that
whenever it checked a special, built-in class (like `Generic` or
`HasField`), it would skip the most important check of all:
`checkValidTypePats`, which rejects nonsense like this:
```hs
instance Generic (forall a. a)
```
This fixes the issue by carving out `checkValidTypePats` from
`check_valid_inst_head` so that `checkValidTypePats` is always
invoked. `check_valid_inst_head` has also been renamed to
`check_special_inst_head` to reflect its new purpose of _only_
checking for instances headed by special classes.
Fixes #17355.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The latest installment in my quest to clean up the code in
`TcDeriv*`. This time, my sights are set on
`TcDerivInfer.inferConstraints`, which infers the context for derived
instances. This function is a wee bit awkward at the moment:
* It's not terribly obvious from a quick glance, but
`inferConstraints` is only ever invoked when using the `stock` or
`anyclass` deriving strategies, as the code for inferring the
context for `newtype`- or `via`-derived instances is located
separately in `mk_coerce_based_eqn`. But there's no good reason
for things to be this way, so I moved this code from
`mk_coerce_based_eqn` to `inferConstraints` so that everything
related to inferring instance contexts is located in one place.
* In this process, I discovered that the Haddocks for the auxiliary
function `inferConstraintsDataConArgs` are completely wrong. It
claims that it handles both `stock` and `newtype` deriving, but
this is completely wrong, as discussed above—it only handles
`stock`. To rectify this, I renamed this function to
`inferConstraintsStock` to reflect its actual purpose and created
a new `inferConstraintsCoerceBased` function to specifically
handle `newtype` (and `via`) deriving.
Doing this revealed some opportunities for further simplification:
* Removing the context-inference–related code from
`mk_coerce_based_eqn` made me realize that the overall structure
of the function is basically identical to `mk_originative_eqn`.
In fact, I was easily able to combine the two functions into a
single `mk_eqn_from_mechanism` function.
As part of this merger, I now invoke
`atf_coerce_based_error_checks` from `doDerivInstErrorChecks1`.
* I discovered that GHC defined this function:
```hs
typeToTypeKind = liftedTypeKind `mkVisFunTy` liftedTypeKind
```
No fewer than four times in different modules. I consolidated all
of these definitions in a single location in `TysWiredIn`.
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
Statements can change the basic block in which instructions
are placed during instruction selection.
We have to keep track of this switch of the current basic block
as we need this information in order to properly update the CFG.
This commit implements this change and fixes #17334.
We do so by having stmtToInstr return the new block id
if a statement changed the basic block.
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
GHC has a wonderful-but-delicate mechanism for building recursive
dictionaries by adding a goal to the "solved dictionaries" before
solving the sub-goals. See Note [Solved dictionaries] in TcSMonad
Ticket #17267 showed that if you use this mechanism for local
/quantified/ constraints you can get a loop -- or even unsafe
coerce. This patch fixes the bug.
Specifically
* Make TcSMonad.addSolvedDict be conditional on using a
/top level/ instance, not a quantified one.
* Moreover, we /also/ don't want to add a solved dict
for equalities (a~b).
* Add lots more comments to Note [Solved dictionaries]
to explain the above cryptic stuff.
* Extend InstanceWhat to identify those strange built-in
equality instances.
A couple of other things along the way
* Delete the unused Type.isIPPred_maybe.
* Stop making addSolvedDict conditional on not being an
impolicit parameter. This comes from way back. But
it's irrelevant now because IP dicts are never solved
via an instance.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
GHC is suposed to uphold the principle that an /inferred/ type
for a let-binding should obey the rules for that module. E.g.
we should only accept an inferred higher rank type if we have
RankNTypes on.
But we were failing to check this: TcValidity.checkValidType
allowed arbitrary rank for inferred types.
This patch fixes the bug. It might in principle cause some breakage,
but if so that's good: the user should add RankNTypes and/or a
manual signature. (And almost every package has explicit user
signatures for all top-level things anyway.) Let's see.
Fixes #17213.
Metric Decrease:
T10370
|
| |
|
|
|
|
|
|
|
|
|
| |
Issue #17339 was caused by using a slightly different version of
`newDFunName` for derived instances that, confusingly enough, did not
take all arguments to the class into account when generating the
`DFun` name. I cannot think of any good reason for doing this, so
this patch uses `newDFunName` uniformly for both derived instances
and manually written instances alike.
Fixes #17339.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Simon realised that the simple language composed of let bindings, bang
patterns and flat constructor patterns is enough to capture the
semantics of the source pattern language that are important for
pattern-match checking. Well, given that the Oracle is smart enough to
connect the dots in this less informationally dense form, which it is
now.
So we transform `translatePat` to return a list of `PmGrd`s relative to
an incoming match variable. `pmCheck` then trivially translates each of
the `PmGrd`s into constraints that the oracle understands.
Since we pass in the match variable, we incidentally fix #15884
(coverage checks for view patterns) through an interaction with !1746.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
If you derive an instance like this:
```hs
deriving <...> instance Foo C
```
And the data constructors for `C` aren't in scope, then
`doDerivInstErrorChecks1` throws an error. Moreover, it will
_only_ throw an error if `<...>` is either `stock` or `newtype`.
This is because the code that the `anyclass` or `via` strategies
would generate would not require the use of the data constructors
for `C`.
However, `doDerivInstErrorChecks1` has another purpose. If you
write this:
```hs
import M (C(MkC1, ..., MkCn))
deriving <...> instance Foo C
```
Then `doDerivInstErrorChecks1` will call `addUsedDataCons` on
`MkC1` through `MkCn` to ensure that `-Wunused-imports` does not
complain about them. However, `doDerivInstErrorChecks1` was doing
this for _every_ deriving strategy, which mean that if `<...>` were
`anyclass` or `via`, then the warning about `MkC1` through `MkCn`
being unused would be suppressed!
The fix is simple enough: only call `addUsedDataCons` when the
strategy is `stock` or `newtype`, just like the other code paths
in `doDerivInstErrorChecks1`.
Fixes #17324.
|
| | |
|
| | |
|
| |
|
|
| |
To avoid polluting the macro namespace
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
The vanilla COMPLETE set is attached to the representation TyCon of a
data family instance, whereas the user-defined COMPLETE sets are
attached to the parent data family TyCon itself.
Previously, we weren't trying particularly hard to get back to the
representation TyCon to the parent data family TyCon, resulting in bugs
like #17207. Now we should do much better.
Fixes the original issue in #17207, but I found another related bug that
isn't so easy to fix.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Previously, all constraints from all top-level groups (as
separated by top-level splices) were lumped together and solved
at the end. This could leak metavariables to TH, though, and
that's bad. This patch solves each group's constraints before
running the next group's splice.
Naturally, we now report fewer errors in some cases.
One nice benefit is that this also fixes #11680, but in a much
simpler way than the original fix for that ticket. Admittedly,
the error messages degrade just a bit from the fix from #11680
(previously, we informed users about variables that will be
brought into scope below a top-level splice, and now we just
report an out-of-scope error), but the amount of complexity
required throughout GHC to get that error was just not worth it.
This patch thus reverts much of
f93c9517a2c6e158e4a5c5bc7a3d3f88cb4ed119.
Fixes #16980
Test cases: th/T16980{,a}
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
Currently, newtype constructors are not marked as used when they are
accessed under the hood by uses of `coerce`, as described in #10347.
This fixes #10347 by co-opting the `tcg_keep` field of `TcGblEnv`
to track uses of newtype constructors in the `Coercible` solver.
See `Note [Tracking unused binding and imports]` in `TcRnTypes`.
Since #10347 is fixed, I was able to simplify the code in `TcDeriv`
slightly, as the hack described in
`Note [Newtype deriving and unused constructors]`
is no longer necessary.
|
| | |
|
| |
|
|
|
|
|
|
|
|
| |
By introducing a `CoreMap Id` to the term oracle, we can represent
syntactically equivalent expressions by the same `Id`. Combine that with
`CoreOpt.simpleCoreExpr` and it might even catch non-trivial semantic
equalities.
Unfortunately due to scoping issues, this will not solve #17208 for
view patterns yet.
|
| |
|
|
|
|
| |
* `foldBal` contains needless partiality that can easily be avoided.
* `mkProd_E` and `mkProd_P` both contain unique supply arguments that
are completely unused, which can be removed.
|
| |
|
|
| |
This way, we can be sure we don't miss a case.
|
| |
|
|
|
|
|
|
|
|
|
|
| |
As described in the new Note [LLVM Configuration] in SysTools, we now
load llvm-targets and llvm-passes lazily to avoid the overhead of doing
so when -fllvm isn't used (also known as "the common case").
Noticed in #17003.
Metric Decrease:
T12234
T12150
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Among other uses, `coreFlattenTyFamApp` is used by Core Lint as a
part of its check to ensure that each type family axiom reduces
according to the way it is defined in the source code. Unfortunately,
the logic that `coreFlattenTyFamApp` uses to flatten type family
applications disagreed with the logic in `TcFlatten`, which caused
it to spuriously complain this program:
```hs
type family Param :: Type -> Type
type family LookupParam (a :: Type) :: Type where
LookupParam (f Char) = Bool
LookupParam x = Int
foo :: LookupParam (Param ())
foo = 42
```
This is because `coreFlattenTyFamApp` tries to flatten the `Param ()`
in `LookupParam (Param ())` to `alpha` (where `alpha` is a flattening
skolem), and GHC is unable to conclude that `alpha` is apart from
`f Char`. This patch spruces up `coreFlattenTyFamApp` so that it
instead flattens `Param ()` to `alpha ()`, which GHC _can_ know for
sure is apart from `f Char`. See
`Note [Flatten], wrinkle 3` in `FamInstEnv`.
|
| |
|
|
|
|
|
|
| |
- Remove unneeded ones
- Use <..> for inter-package.
Besides general clean up, helps distinguish between the RTS we link
against vs the RTS we compile for.
|
| |
|
|
|
| |
If a main module doesn't contain a header, we omit the check whether the main module is exported.
With this patch GHC, GHCi and runghc use the same code.
|
| |
|
|
|
|
| |
You can always just not use or even build `iserv`. I don't think the
maintenance cost of the CPP is worth...I can't even tell what the
benefit is.
|
| |
|
|
| |
not found.
|
| |
|
|
|
|
|
|
|
|
|
| |
This has two benefits:
1. One less hunk of code dependent on DynFlags
2. Add a little bit of error granularity to distrinugish between missing
data and bad data. This could someday be shared with ghc-pkg which
aims to work even with a missing file. I also am about to to make
--supported-extensions use this too.
|
| | |
|
| |
|
|
|
|
| |
not used.
This fixes #10913.
|
| |
|
|
| |
They are only used in a file we construct directly, so just skip CPP.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The generated headers are now generated per stage, which means we can
skip hacks like `ghc_boot_platform.h` and just have that be the stage 0
header as proper. In general, stages are to be embraced: freely generate
everything in each stage but then just build what you depend on, and
everything is symmetrical and efficient. Trying to avoid stages because
bootstrapping is a mind bender just creates tons of bespoke
mini-mind-benders that add up to something far crazier.
Hadrian was pretty close to this "stage-major" approach already, and so
was fairly easy to fix. Make needed more work, however: it did know
about stages so at least there was a scaffold, but few packages except
for the compiler cared, and the compiler used its own counting system.
That said, make and Hadrian now work more similarly, which is good for
the transition to Hadrian. The merits of embracing stage aside, the
change may be worthy for easing that transition alone.
|
| |
|
|
|
| |
Previously we would throw an error which seems a bit harsh. As reported
in #17283.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
This patch also changes the way we handle empty lists, simplifying
them somewhat. See Note [Empty lists]. Previously, we had to
special-case empty lists in the type-checker. Now no more!
Finally, this patch improves some documentation around the ir_inst
field used in the type-checker.
This breaks a test case, but I really think the problem is #17251,
not really related to this patch.
Test case: typecheck/should_compile/T13680
|
| |
|
|
| |
This is a good convenience for testing.
|
| |
|
|
|
|
|
|
|
|
| |
Previously, we would sometimes flatten 1-tuples and sometimes
not. This didn't cause damage because there is no way to
generate HsSyn with 1-tuples. But, with the upcoming fix to #16881,
there will be. Without this patch, obscure lint errors would
have resulted.
No test case, as there is not yet a way to tickle this.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This patch:
* Implements a refactoring (suggested in
https://gitlab.haskell.org/ghc/ghc/merge_requests/1199#note_207345)
that moves all functions from `TcTypeableValidity` back to
`TcTypeable`, as the former module doesn't really need to live on its
own.
* Adds `Note [Typeable instances for casted types]` to `TcTypeable`
explaining why the `Typeable` solver currently does not support
types containing casts.
Resolves #16835.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Simon writes:
Currently we check for a type arg rather than isTyCoArg. This in turn
makes INLINE things look bigger than they should be, and stops them
being inlined into boring contexts when they perfectly well could be.
E.g.
f x = g <refl> x
{-# INLINE g #-}
... (map (f x) xs) ...
The context is boring, so don't inline unconditionally. But f's RHS is
no bigger than its call, provided you realise that the coercion argument
is ultimately cost-free.
This happens in practice for $WHRefl. It's not a big deal: at most it
means we have an extra function call overhead. But it's untidy, and
actually worse than what happens without an INLINE pragma.
Fixes #17182
This makes 0.0% change in nofib binary sizes.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
* `mk_eqn_stock`, `mk_eqn_anyclass`, and `mk_eqn_no_mechanism` all
took a continuation of type
`DerivSpecMechanism -> DerivM EarlyDerivSpec` to represent its
primary control flow. However, in practice this continuation was
always instantiated with the `mk_originative_eqn` function, so
there's not much point in making this be a continuation in the
first place.
This patch removes these continuations in favor of invoking
`mk_originative_eqn` directly, which is simpler.
* There were several parts of `TcDeriv` that took different code
paths if compiling an `.hs-boot` file. But this is silly, because
ever since 101a8c770b9d3abd57ff289bffea3d838cf25c80 we simply error
eagerly whenever attempting to derive any instances in an
`.hs-boot` file.
This patch removes all of the unnecessary `.hs-boot` code paths,
leaving only one (which errors out).
* Remove various error continuation arguments from `mk_eqn_stock`
and related functions.
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
| |
We were failing to correctly implement Note [Unbound RULE binders]
in Rules.hs. In particular, when cooking up a fake Refl,
were were failing to apply the substitition.
This patch fixes that problem, and simultaneously tidies
up the impedence mis-match between RuleSubst and TCvSubst.
Thanks to Sebastian!
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This commit refactors interface file generation to allow information
from the later passed (NCG, STG) to be stored in interface files.
We achieve this by splitting interface file generation into two parts:
* Partial interfaces, built based on the result of the core pipeline
* A fully instantiated interface, which also contains the final
fingerprints and can optionally contain information produced by the backend.
This change is required by !1304 and !1530.
-dynamic-too handling is refactored too: previously when generating code
we'd branch on -dynamic-too *before* code generation, but now we do it
after.
(Original code written by @AndreasK in !1530)
Performance
~~~~~~~~~~~
Before this patch interface files where created and immediately flushed
to disk which made space leaks impossible.
With this change we instead use NFData to force all iface related data
structures to avoid space leaks.
In the process of refactoring it was discovered that the code in the
ToIface Module allocated a lot of thunks which were immediately forced
when writing/forcing the interface file. So we made this module more
strict to avoid creating many of those thunks.
Bottom line is that allocations go down by about ~0.1% compared to
master.
Residency is not meaningfully different after this patch.
Runtime was not benchmarked.
Co-Authored-By: Andreas Klebinger <klebinger.andreas@gmx.at>
Co-Authored-By: Ömer Sinan Ağacan <omer@well-typed.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
In #14998 I realised that the notion of speculative execution
*exactly matches* eager evaluation of expressions in a case alternative
where the scrutinee is an IO action.
Normally we have to `deferIO` any result from that single case
alternative to prevent this speculative execution, so we had a special
case in place in the demand analyser that would check if the scrutinee
was a prim-op, in which case we assumed that it would be ok to do the
eager evaluation.
Now we just check if the scrutinee is `exprOkForSpeculation`,
corresponding to the notion that we want to push evaluation of the
scrutinee *after* eagerly evaluating stuff from the case alternative.
This fixes #14988, because it resolves the last open Item 4 there.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
`pmcheck` used to call `refineToAltCon` which would refine the knowledge
we had about a variable by equating it to a `ConLike` application.
Since we weren't particularly smart about this in the Check module, we
simply freshened the constructors existential and term binders utimately
through a call to `mkOneConFull`.
But that instantiation is unnecessary for when we match against a
concrete pattern! The pattern will already have fresh binders and field
types. So we don't call `refineToAltCon` from `Check` anymore.
Subsequently, we can simplify a couple of call sites and functions in
`PmOracle`. Also implementing `computeCovered` becomes viable and we
don't have to live with the hack that was `addVarPatVecCt` anymore.
A side-effect of not indirectly calling `mkOneConFull` anymore is that
we don't generate the proper strict argument field constraints anymore.
Instead we now desugar ConPatOuts as if they had bangs on their strict
fields. This implies that `PmVar` now carries a `HsImplBang` that we
need to respect by a (somewhat ephemeral) non-void check. We fix #17234
in doing so.
|
| |
|
|
|
|
|
|
|
| |
Basically do what we currently only do for -XEmptyCase in other cases
where adding the type signature won't distract from pattern
matches in other positions.
We use the precedence to guide us, equating "need to parenthesise" with
"too much noise".
|
| |
|
|
|
| |
`rts.conf` already contains this exact information in its
`extra-libraries` stanza.
|
