Draft: try to eliminate all usages of `addFatalError' and substitue with `addError' followed by `return badValue' for a suitable `badValue'

The motivation is that some parsing errors still halt the whole pipeline due to
wrong interaction between the catch mechansim and the layout rule requiring
emission of `error' tokens, So we try to keep as much info in the source up
until the occurrence of error.

Use the new `%error {} {}` resumption facilities with `ValD`, `HsUnboundVar', `WildPat' and `HsWildCardTy' constructors representing parse errors.

Ben Gamari authored 3 months ago and Marge Bot committed 3 months ago

While the testsuite driver already normalizes these away, they are
nevertheless a severe nuisance when diffing outside of the testsuite.

Intriguingly, this doesn't completely eliminate the unit IDs; some
wired-in names are still printed. However, this is a cheap and helpful
improvement over the status quo so I am simply going to accept this.

Fixes #25334.

Rodrigo Mesquita authored 4 months ago and Marge Bot committed 3 months ago

Fixes #25574

Ben Gamari authored 3 months ago and Marge Bot committed 3 months ago

This makes it considerably easier to grok the structure of the heap
when IPE information is available.

We found that speculative evaluation can increase the amount of
allocations in some circumstances. This patch adds new flags for
selectively disabling speculative evaluation, allowing us to
test the effect of the optimization.

The new flags are:

  -fspec-eval
     globally enable speculative evaluation

  -fspec-eval-dictfun
     enable speculative evaluation for dictionary functions (no effect
     if speculative evaluation is globally disabled)

The new flags are on by default for all optimisation levels.

See #25284

Bryan R authored 3 months ago and Marge Bot committed 3 months ago

Fixes #25620

Simon Peyton Jones authored 3 months ago and Marge Bot committed 3 months ago

Addresses #25630

In particular,

* Introduce ConArgKind and use it.

* Make kcConDecls and tcConDecls work the same way
  concerning the kind of argument types

Matthew Pickering authored 4 months ago and Marge Bot committed 3 months ago

There were two ad-hoc mechanisms used to determine which modules were in
the interactive scope.

1. Look at everything in the GRE, to see what is imported qualified.
2. Look at the last loaded module in the HPT.

(1) Is very inefficient, GlobalRdrEnvs can be very big.
(2) is incorrect, there is no reason to assume the "last" thing added to
the HPT has any relevance to module loading order.

Happily, the same checks can be implemented directly by looking at the
interactive imports from the interactive context. This mirrors what
happens for normal imports.

Arguably, the error reporting code shouldn't be doing this kind of
processing and it should be an option is set when rendering the error
message. However, this just improves the situation and doesn't block
progress on that front in future.

See #14225 and #15611

Fixes #25600

Krzysztof Gogolewski authored 3 months ago and Marge Bot committed 3 months ago

The issue has been fixed by commit f5d3e03c.
Only T23883a is the actual regression test, the remaining ones are
tricky cases found during development of an independent fix !11313.

Simon Peyton Jones authored 4 months ago and Marge Bot committed 3 months ago

Ticket #25468 showed an assertion failure in CSE because a top-level
Id was being used before it was defined.  Reason: Note [Glomming] in
GHC.Core.Opt.OccurAnal.

Solution (used in many places): just put all the top-level bindings in
scope at the beginning of CSE.

Compile-time allocation wobbles up and down a tiny bit; geo mean is
zero. But MultiLayerModulesTH_OneShot and hard_hole_fits increase (on
some architectures only) by a bit oever 2% .  I think these are just a
random fluctuations.

Metric Increase:
    MultiLayerModulesTH_OneShot
    hard_hole_fits

Zubin authored 4 months ago and Zubin committed 3 months ago

os-string has redundant imports

(cherry picked from commit dde3796b)

Zubin authored 4 months ago and Zubin committed 3 months ago

(cherry picked from commit cbfd0829)

Zubin authored 4 months ago and Zubin committed 3 months ago

Otherwise we link against boot inplace and boot unix as boot haskeline
depends on boot unix.

(cherry picked from commit 90b49376)

Zubin authored 4 months ago and Zubin committed 3 months ago

(cherry picked from commit 77f340a2)

Zubin authored 6 months ago and Zubin committed 3 months ago

Also bump various submodules.

(cherry picked from commit 6fc1fa3b)

Bump base bound to 4.21 for GHC 9.12

(cherry picked from commit 473a201c)

Bump binary submodule to 0.8.9.2

(cherry picked from commit 7199869a)
(cherry picked from commit ec2f40b4)

Bump exceptions submodule to 0.10.9

(cherry picked from commit f5b5d1dc)

Bump file-io submodule to 0.1.4

(cherry picked from commit ba786681)

bump os-string submodule to 2.0.6

(cherry picked from commit 3a7ffdbb)

bump transformers submodule to 0.6.1.2

(cherry picked from commit 53b46fd4)

Bump directory submodule to v1.3.9.0

(cherry picked from commit 27dc2664)

Bump Win32 submodule to v2.14.1.0

(cherry picked from commit 80df8808)

Bump filepath submodule to 1.5.3.0

(cherry picked from commit 29bfae2c)

Bump file-io submodule to avoid usage of QuasiQuotes

(cherry picked from commit 97b0dff2)

Bump unix submodule to 2.8.6.0

(cherry picked from commit a1f56d6d)

Bump os-string submodule to 2.0.8

(cherry picked from commit 0121b76f)

Bump file-io submodule to avoid usage of QuasiQuotes

(cherry picked from commit 962ceb50)

Bump filepath submodule to 1.5.4.0

(cherry picked from commit 7bc6877f)

Bump file-io submodule to 0.1.5

(cherry picked from commit 9478b5ae)

Bump Cabal submodule to 3.14.1.0

(cherry picked from commit 5c9c3e3f)

Bump directory submodule to 0.12.2.0

(cherry picked from commit 89790626)

Bump array submodule for base bump

Bump stm submodule for base bump

Bump process submodule for base bump

ARATA Mizuki authored 3 months ago and Marge Bot committed 3 months ago

With a recent LLVM, `llc -version` emits the version on the first line
if the vendor is set.  It emits the version on the second line
otherwise.

Therefore, we need to check the both lines to detect the version.

GHC now emits a warning if it fails to detect the LLVM version,
so we can notice if the output of `llc -version` changes in the future.

Also, the warning for using LLVM < 10 on s390x is removed, because
we assume LLVM >= 13 now.

This fixes the definition of __GLASGOW_HASKELL_LLVM__ macro.

Fixes #25606

The Monad instance shouldn't produce the outer :| unless f a reduces
to WHNF. (Notice that the b :| bs match is implicitly lazy.)

Implementation of https://github.com/haskell/core-libraries-committee/issues/107

Simon Peyton Jones authored 4 months ago and Marge Bot committed 3 months ago

This MR fixes yet another manifestation of the trickiness caused
by Note [Fundeps with instances, and equality orientation].

I wish there was a more robust way to do this, but this fix is
a definite improvement.

Fixes #25597

Ben Gamari authored 4 months ago and Marge Bot committed 3 months ago

This slightly eases updating of the `latest` symlinks.

Andrei Borzenkov authored 4 months ago and Marge Bot committed 3 months ago

GHC Proposal 583 "HasField redesign" specifies the
following order of a setField function arguments as this:

  setField :: forall fld a b. SetField fld a b. b -> a -> a

This patch flips the application order to match the spec.

Rodrigo Mesquita authored 5 months ago and Marge Bot committed 4 months ago

This eta expansion was introduced with the initial commit for Linear
types.

I believe this isn't needed any longer. My guess is it is an artifact
from the initial linear types implementation: data constructors are
linear, but they shouldn't need to be eta expanded to be used as higher
order functions. I suppose in the early days this wasn't true.

For instance, this works now:

    data T x = T x
    f = \(x :: forall y. y -> T y) -> x True
    f T -- ok!

T is linear, but can be passed where an unrestricted higher order
function is expected. I recall there being some magic around to make
this work for data constructors...

Since this works, there's no need to eta_expand the data constructors in
the derived Data instances.

Rodrigo Mesquita authored 5 months ago and Marge Bot committed 4 months ago

Significantly improves performance of deriving Show instances by
avoiding using the very polymorphic `.` operator in favour of inlining
its definition. We were generating tons of applications of it, each
which had 3 type arguments!

Improves on #9557

-------------------------
Metric Decrease:
    InstanceMatching
    T12707
    T3294
------------------------

Sylvain Henry authored 4 months ago and Marge Bot committed 4 months ago

First step towards merging ghc-bignum and ghc-prim into ghc-internal.

After this patch, ghc-bignum is deprecated and is just a shallow package
reexporting modules from ghc-internal and base. Use those directly
instead.

Move `gmp` submodule into ghc-internal directory.

Zubin authored 4 months ago and Marge Bot committed 4 months ago

Redhat 9 doesn't have libtinfo.so.5 anymore

(cherry picked from commit dc86785e)

Zubin authored 4 months ago and Marge Bot committed 4 months ago

(cherry picked from commit 52b58a66)