- Sep 21, 2023
-
-
-
-
-
libiserv has been merged into the ghci package.
-
Programs produced by the JavaScript backend were returning incorrect file sizes and modification times, causing cabal related tests to fail. This fixes the problem and adds an additional test that verifies basic file information operations. fixes #23980
-
-
-
-
This is much more terse because we are programmatically handling the leading underscore. `findPtr` however is still handled in the Cabal file because we need a newer Cabal to pass flags to the configure script automatically. Co-Authored-By:Ben Gamari <ben@well-typed.com>
-
This is a hack to get around the cabal function we're calling *decreasing* the verbosity it passes to another function, which is the stuff we often actually care about. Sigh. Keeping this a separate commit so if this makes things too verbose it is easy to revert.
-
Right now, output is squashed. This make per-package configure scripts extremely hard to maintain, because we get vague "library is missing" errors when the actually probably is usually completely unrelated except for also involving the C/C++ toolchain. (I can always pass `-VVV` to Hadrian locally, but these errors are subtle and I often cannot reproduce them locally!) `--disable-option-checking` was added back in 75c6e068 but seems to be a bit overkill; if other flags are passed that are not recognized behind the two from Cabal mentioned in the former comment, we *do* want to know about it.
-
When the script is changed, we should reconfigure.
-
We need the newer version for `CABAL_FLAG_*` env vars for #17191.
-
- Sep 19, 2023
-
-
Fixes #23977
-
-
It is dead code since the Make build system was removed. I tried to go over every match of `git grep -i ghc-cabal` to find other stray bits. Some of those might be workarounds that can be further removed.
-
-
This commit changes how we perform some validity checking for coercion axioms to mirror how we handle default declarations for associated type families. This allows us to keep track of whether type variables in type and data family instances were user-written or not, in order to only report the user-written ones in "unused type variable" error messages. Consider for example: {-# LANGUAGE PolyKinds #-} type family F type instance forall a. F = () In this case, we get two quantified type variables, (k :: Type) and (a :: k); the second being user-written, but the first is introduced by the typechecker. We should only report 'a' as being unused, as the user has no idea what 'k' is. Fixes #23734 -
This commit adjusts the validity checking of data family instances to improve the reporting of unused type variables. See Note [Out of scope tvs in data family instances] in GHC.Tc.Validity. The problem was that, in a situation such as data family D :: Type data instance forall (d :: Type). D = MkD the RHS passed to 'checkFamPatBinders' would be the TyCon app R:D d which mentions the type variable 'd' quantified in the user-written forall. Thus, when computing the set of unused type variables in the RHS of the data family instance, we would find that 'd' is used, and report a strange error message that would say that 'd' is not bound on the LHS. To fix this, we special-case the data-family instance case, manually extracting all the type variables that appear in the arguments of all the data constructores of the data family instance. Fixes #23778
-
-
This commit passes the correct set of quantified type variables written by the user in associated type default declarations for validity checking. This ensures that validity checking of associated type defaults mirrors that of standalone type family instances. Fixes #23768 (see testcase T23734 in subsequent commit)
-
closes: #23259 - adds -fprint-error-index-links={auto|always|never} flag
-
-
This adds a debian 11 release job for aarch64. Fixes #22005
-
This is dynamically linked and makes creating statically linked executables more straightforward. Fixes #23482
-
- Sep 18, 2023
-
-
Since the fix was backported to 9.8.1
-
This function never fires, so we can delete it: #23965.
-
As noted in #23538, `readFloat` has runtime that scales nonlinearly in the size of its input. Consequently, its use on untrusted input can be exploited as a denial-of-service vector. Point this out and suggest use of `read` instead. See #23538.
-
-
Closes #23885 Updates haddock submodule
-
- Sep 16, 2023
-
-
-
As the Lint error in #23952 showed, the function adjustJoinPointType was failing to adjust the FunTyFlag when adjusting the type. I don't think this caused the seg-fault reported in the ticket, but it is definitely. This patch fixes it. It is tricky to come up a small test case; Krzysztof came up with this one, but it only triggers a failure in GHC 9.6.
-
- Sep 15, 2023
- Sep 14, 2023
-
-
-
There are some ways in which unusual characters like quotes or others can make it into cost centre names. So properly escape these. Fixes #23924
-
- Sep 13, 2023
-
-
Issue #23922 showed that GHC was bogusly eta-reducing a join point. We should never eta-reduce (\x -> j x) to j, if j is a join point. It is extremly difficult to trigger this bug. It took me 45 mins of trying to make a small tests case, here immortalised as T23922a.
-
When the user adds a SPEC argument to a function, they are informing us that they expect the function to be specialised. However, previously this instruction could be preempted by the specialised-argument limit (sc_max_args). Fix this. This fixes #14003.
-
canEtaReduceToArity was only called internally, and always with two arguments equal to zero. This patch just specialises the function, and renames it to cantEtaReduceFun. No change in behaviour.
-
This -fno-cse change is to avoid these performance tests depending on flukey CSE stuff. Each contains several independent tests, and we don't want them to interact. See #23925. By killing CSE we expect a 400% increase in T15426, and 100% in T18964. Metric Increase: T15426 T18964
-
