| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
By referential transparency, multiple syntactic occurrences of the same
expression evaluate to the same value. Global value numbering (GVN)
assigns each such expression the same unique number (a `Name` in our
case). Two expressions trivially have the same value if they are
assigned the same value number.
The term oracle `TmOracle` of the pattern match checker couldn't handle
any complex expression before this patch. It would just give up on
anything involving a function application whose head was not a
constructor, by falling back to `PmExprOther`. This means it could not
determine completeness of the following example:
```haskell
foo
| True <- id True
= 1
| False <- id True
= 2
```
This is simply because `TmOracle` couldn't figure out that `id True`
always evaluates to the same `Bool`.
In this patch, we desugar such `PmExprOther`s in pattern guards to
`CoreExpr`. We do so in order to utilise `CoreMap Name` for a
light-weight GVN pass without concern for subexpressions. `TmOracle`
only sees the representing variables, like so:
```haskell
x = id True
foo
| True <- x
= 1
| False <- x
= 2
```
So `TmOracle` still doesn't need to decide equality of complex
expressions, which allows it to stay dead simple.
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
| |
The bulk of the work was done in #712, making settings be make/Hadrian
controlled. This commit then just moves the unlit command rules in
make/Hadrian from the `Config.hs` generator to the `settings` generator
in each build system.
I think this is a good change because the crucial benefit is *settings*
don't affect the build: ghc gets one baby step closer to being a regular
cabal executable, and make/Hadrian just maintains settings as part of
bootstrapping.
|
| |
|
|
| |
It is no longer needed
|
| |
|
|
| |
This allows it to eventually become stage-specific
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This does four things:
1. Look at `idArity` instead of manifest lambdas to decide whether to use LetUp
2. Compute the strictness signature in LetDown assuming at least `idArity`
incoming arguments
3. Remove the special case for trivial RHSs, which is subsumed by 2
4. Don't perform the W/W split when doing so would eta expand a binding.
Otherwise we would eta expand PAPs, causing unnecessary churn in the
Simplifier.
NoFib Results
--------------------------------------------------------------------------------
Program Allocs Instrs
--------------------------------------------------------------------------------
fannkuch-redux +0.3% 0.0%
gg -0.0% -0.1%
maillist +0.2% +0.2%
minimax 0.0% +0.8%
pretty 0.0% -0.1%
reptile -0.0% -1.2%
--------------------------------------------------------------------------------
Min -0.0% -1.2%
Max +0.3% +0.8%
Geometric Mean +0.0% -0.0%
|
| |
|
|
| |
Previously hsc2hs failed when building against a system FFI.
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
|
| |
When encountering type signatures for unknown names, suggest similar
alternatives.
This fixes issue #16504
|
| |
|
|
|
| |
It seems that this was inadvertently dropped in
1285d6b95fbae7858abbc4722bc2301d7fe40425.
|
| |
|
|
| |
Scripts taken from autoconf a8d79c3130da83c7cacd6fee31b9acc53799c406
|
| | |
|
| |
|
|
|
|
| |
- Remove redundant casting in evacuate_static_object
- Remove redundant parens in STATIC_LINK
- Fix a typo in GC.c
|
| |
|
|
|
| |
There is a hint added to error messages reported in checkPattern.
Instead of passing it manually, we put it in a ReaderT environment inside PV.
|
| |
|
|
|
|
|
|
|
|
|
| |
Previously we defined type PV = P,
this had the downside that if we wanted to change PV,
we would have to modify P as well.
Now PV is free to evolve independently from P.
The common operations addError, addFatalError, getBit, addAnnsAt,
were abstracted into a class called MonadP.
|
| | |
|
| | |
|
| |
|
|
|
|
| |
rlim_t is a signed type on FreeBSD, and the build fails with a
sign-compare error. Add explicit (unsigned) cast to handle this
case.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
EWildPat is a constructor of HsExpr used in the parser to represent
wildcards in ambiguous positions:
* in expression context, EWildPat is turned into hsHoleExpr (see rnExpr)
* in pattern context, EWildPat is turned into WildPat (see checkPattern)
Since EWildPat exists solely for the needs of the parser, we could
remove it by improving the parser.
However, EWildPat has also been used for a different purpose since
8a50610: to represent patterns that the coverage checker cannot handle.
Not only this is a misuse of EWildPat, it also stymies the removal of
EWildPat.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
4 haddock tests assume that .haddock files have been produced, by using the
'req_haddock' modifier. The testsuite driver assumes that this condition is
satisfied if 'config.haddock' is non-empty, but before this patch Hadrian was
always passing the path to where the haddock executable should be, regardless
of whether it is actually there or not.
Instead, we now pass an empty config.haddock when we can't find all of
<build root>/docs/html/libraries/<pkg>/<pkg>.haddock>, where <pkg> ranges over
array, base, ghc-prim, process and template-haskell, and pass the path
to haddock when all those file exists. This has the (desired) effect of skipping
the 4 tests (marked as 'missing library') when the docs haven't been built,
and running the haddock tests when they have.
|
| |
|
|
| |
Fixes #16569
|
| | |
|
| | |
|
| |
|
|
| |
returned element may be two different things
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| |
|
|
| |
Due to #15934.
|
| | |
|
| | |
|
| |
|
|
| |
Due to allowed failure of Windows job.
|
| |
|
|
| |
Previously the failure was quite nondescript.
|
| | |
|
| |
|
|
|
|
|
|
|
| |
Rather than massaging the output of the parser to re-arrange docs and
bangs, it is simpler to patch the two places in which the strictness
info is needed (to accept that the `HsBangTy` may be inside an
`HsDocTy`).
Fixes #16585.
|
| |
|
|
|
|
|
|
|
|
|
| |
Before this change, we used a roundabout encoding:
1. a GADT (ExpCmdG)
2. a class to pass it around (ExpCmdI)
3. helpers to match on it (ecHsApp, ecHsIf, ecHsCase, ...)
It is more straightforward to turn these helpers into class methods,
removing the need for a GADT.
|
| |
|
|
|
|
|
| |
Makes the width of STG dumps slightly saner.
Especially for things like unboxing.
Fixes #16580
|
| |
|
|
|
|
|
|
|
|
|
| |
Previously we would generate a local variable pointing after the array
header and use it to initialize the array elements. But we already use
stores with offset, so it's easy to just add the header to those offsets
during compilation and avoid generating the local variable (which would
become a LEA instruction when using native codegen; LLVM already
optimizes it away).
Signed-off-by: Michal Terepeta <michal.terepeta@gmail.com>
|
| |
|
|
|
|
|
|
| |
Now that `TExp` accepts unlifted types, #16384 is fixed. Since the real
issue there was GHC letting through an ill-kinded type which
`-dcore-lint` rightly rejected, a reasonable regression test is that
the program from #16384 can now be accepted without `-dcore-lint`
complaining.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Besides the obvious benefits of being able to manipulate `TExp`'s of
unboxed types, this also simplified `-XDeriveLift` all while making
it more capable.
* `ghc-prim` is explicitly depended upon by `template-haskell`
* The following TH things are parametrized over `RuntimeRep`:
- `TExp(..)`
- `unTypeQ`
- `unsafeTExpCoerce`
- `Lift(..)`
* The following instances have been added to `Lift`:
- `Int#`, `Word#`, `Float#`, `Double#`, `Char#`, `Addr#`
- unboxed tuples of lifted types up to arity 7
- unboxed sums of lifted types up to arity 7
Ideally we would have levity-polymorphic _instances_ of unboxed
tuples and sums.
* The code generated by `-XDeriveLift` uses expression quotes
instead of generating large amounts of TH code and having
special hard-coded cases for some unboxed types.
|
| |
|
|
|
| |
We used to pass YES/NO, while that particular field is set to True/False. This
happens to fix an unexpected pass, T9208.
|
| | |
|
| |
|
|
| |
[skip ci]
|
| |
|
|
| |
When printf debugging this can be helpful.
|
| |
|
|
| |
E.g. use `\(\mathcal{O}(n^2)\)` instead of `/O(n^2)/`.
|
