GitLab

Ord Unique can be a source of invisible, accidental
nondeterminism as explained in Note [No Ord for Unique].
This removes it, leaving a note with rationale.

It's unfortunate that I had to write Ord instances for
codegen data structures by hand, but I believe that it's a
right trade-off here.

Test Plan: ./validate

Reviewers: simonmar, austin, bgamari

Reviewed By: simonmar

Subscribers: thomie

Differential Revision: https://phabricator.haskell.org/D2370

GHC Trac Issues: #4012

The cause of #12213 is in dump_stdout and dump_stderr:

      print(read_no_crs(<filename>))

Commit 6f6f5154 changed read_no_crs to
return a unicode string. Printing a unicode strings works fine as long
as sys.stdout.encoding is 'UTF-8'.

There are two reasons why sys.stdout.encoding might not be 'UTF-8'.

* When output is going to a file, sys.stdout and sys.stdout do not respect
  the locale:

  $ LC_ALL=en_US.utf8 python -c 'import sys; print(sys.stderr.encoding)'
  UTF-8
  $ LC_ALL=en_US.utf8 python -c 'import sys; print(sys.stderr.encoding)' 2>/dev/null
  None

* When output is going to the terminal, explicitly reopening sys.stdout has
  the side-effect of changing sys.stdout.encoding from 'UTF-8' to 'None'.

      sys.stdout = os.fdopen(sys.__stdout__.fileno(), "w", 0)

  We currently do this to set a buffersize of 0 (the actual
  buffersize used is irrelevant for the sys.stdout.encoding problem).

Solution: fix dump_stdout and dump_stderr to not use read_no_crs.

As Trac #12242 showed, there was a missing case in
the pretty printer for IfaceCo. I've refactored it
so that the pattern-match ovelap checker will spot
it next time.

The problem with ignore_output is that it hides errors for WAY=ghci.
GHCi always returns with exit code 0 (unless it is broken itself).

For example: ghci015 must have been failing with compile errors for
years, but we didn't notice because all output was ignored.

Therefore, replace all uses of ignore_output with either ignore_stderr
or ignore_stdout. In some cases I opted for adding the expected output.

Update submodule hpc and stm.

Reviewed by: simonmar

Differential Revision: https://phabricator.haskell.org/D2367

I have get test failures on `openFile008` with
`openFile: resource exhausted (Too many open files)` when running
inside `./validate`, but not when I run the test individually.
I suspect that's because with `./validate` parallelism of 33 threads
I go just above the `1024` file descriptor limit.
This is probably related to the recent change:
`58f0086b: Testsuite: open/close stdin/stdout/stderr explicitly`
but I haven't looked deep enough to understand exactly how.
I think bumping this is harmless, but I don't really know why it's
necessary at all.

Test Plan: ./validate

Reviewers: austin, thomie, hvr, bgamari, simonmar

Reviewed By: simonmar

Subscribers: simonmar

Differential Revision: https://phabricator.haskell.org/D2368

Unique order can introduce nondeterminism.
As a step towards removing the Ord Unique instance
I've refactored the code to use deterministic sets instead.

Test Plan: ./validate

Reviewers: simonmar, austin, bgamari

Subscribers: thomie

Differential Revision: https://phabricator.haskell.org/D2369

GHC Trac Issues: #4012

Summary:
We have the FloatOut pass create exported ids for floated StaticPtr
bindings. The simplifier doesn't try to remove those.

This patch also improves on 7fc20b by making a common definition
collectStaticPtrSatArgs to test for StaticPtr binds.

Fixes #12207.

Test Plan: ./validate

Reviewers: simonpj, austin, bgamari, simonmar, goldfire

Reviewed By: simonpj

Subscribers: thomie

Differential Revision: https://phabricator.haskell.org/D2366

GHC Trac Issues: #12207

I'm not quite sure why this changed with my two recent commits,
but it /has/ changed (in a benign way) so I'm accepting it.
Maybe it wasn't me anyway... but life is short and I'm not inclined
to dig further.

We only do newtype unwrapping for Coercible constraints if
the newtype's data constructor is in scope.  We were trying to
record the fact that the data constructor was thereby 'used', so
that an import statement would not be flagged as unnecsssary
(by -Wunused-imports).

But the code was simply wrong. It was wrong because it assumed
that only one level of unwrapping happened, whereas
tcTopNormaliseNewTypeTF_maybe actually unwraps multiple layers.
So we need to return a /list/ of data constructors that are used.

This entailed a bit of refactoring, as usual.

Fixes Trac #12067

This fixes Trac #12156, where we were omitting to make an
evidence binding (because cec_suppress was on), but yet the
program was compiled and run.

The fix is easy, and involves deleting code :-).

* T7837 is still broken for prof_ways (#9406)
* T11627b is broken on Windows for WAY=prof_hc_hb (#12236)
* T8089 is also broken for WAY=profasm on Windows

This prevents `cannot find -lHSrts_debug_p` when running `make
TEST=T9078 WAY=profasm` (#9078).

I don't know why some ways were omitted before.

* add framework failures to unexpected results list
* report errors in .T files as framework failures (show in summary)
* don't report missing tests when framework failures in .T files

Just use a simple list of tuples, instead of a nested map.

-90 lines of code.

This allows run_command's to contain `|`, and `no_stdin` isn't necessary
anymore.

Unfortunately it doesn't fix T7037 on Windows which I had hoped it would
(testsuite driver tries to read a file that it just created itself, but
the OS says it doesn't exist).

The only drawback of this commit is that the command that the testsuite
prints to the terminal (for debugging purposes) doesn't mention the
files that stdout and stderr are redirected to anymore. This is probably
ok.

Update submodule unix.

Differential Revision: https://phabricator.haskell.org/D1234

This is a followup to e1293bbf, but then
for Windows timeout.

Before, `extra_files(['.hpc/Main.mix'])` meant copy `Main.mix` to
`<testdir>/.hpc/Main.mix`. This feature wasn't really necessary, so now
it just means copy `Main.mix` to `<testdir>/Main.mix`. This simplifies
the implementation.

Some small other cleanups as well. -40 lines of code.

Refactoring only. Move try/except out of do_test.

And use os.walk instead of calling os.listdir many times. The testsuite
driver should be able to handle backward slashes on Windows now.

Summary:
Rename it to `USED_IF_x86_64_HOST_ARCH` to make it more like the existing
`USED_IF` macros as suggested by Simon Marlow.

Test Plan: Build on x86_64 windows

Reviewers: Phyx, bgamari, simonmar, austin

Subscribers: thomie

Differential Revision: https://phabricator.haskell.org/D2365

These get annoying when `undefined` is actually used as placeholder in WIP code.
Some of these were also completely redundant (just call `deAnnotate'` instead of
`deAnnotate` etc.).

Summary:
T8761 seems to get different interleaving in its output on
Phabricator, so this switches it to use stderr consistently.

Test Plan: worksforme

Reviewers: goldfire, austin, bgamari, erikd

Subscribers: thomie

Differential Revision: https://phabricator.haskell.org/D2363

Test Plan: Build on x86_64/linux, x86_64/darwin and powerpc/linux

Reviewers: Phyx, bgamari, simonmar, austin

Reviewed By: Phyx

Subscribers: thomie

Differential Revision: https://phabricator.haskell.org/D2364

Summary:
Commit 90538d86, seems to have broken static linking.
The introduction of `argFixup` in `runLink` rearranges libs, and considers
anything with an `-l` prefix or `.a` suffix a lib, which fails for libs that are
just being linked together (e.g. `-o lib.a`).

The proposed solution explicitly checks for the existance of the `-o` flag.

Reviewers: rwbarton, erikd, Phyx, bgamari, austin

Reviewed By: Phyx

Subscribers: thomie

Differential Revision: https://phabricator.haskell.org/D2362

This removes an unnecessary loop looking for invisible binders
and tries to clarify what the very closely-related functions
tcInferArgs, tc_infer_args, tcInferApps all do.

This renames VisibilityFlag from

> data VisibilityFlag = Visible | Specified | Invisible

to

> data ArgFlag = Required | Specified | Inferred

The old name was quite confusing, because both Specified
and Invisible were invisible! The new names are hopefully clearer.

* Replaced 2.11's *TBA* with *May 2016*
* Removed an outdated TODO comment
* Removed lines which incorrectly stated that some strictness-related
  functions had been removed (they have been deprecated instead)

Summary: Add more Notes and signposts across the codebase to help navigation.

Test Plan: validate

Reviewers: goldfire, simonpj, austin, ezyang, hvr, bgamari, erikd

Subscribers: thomie

Differential Revision: https://phabricator.haskell.org/D2358

Summary:
From a suggestion by @goldfire: clean up the message types, so that
rather than one Message type with all the messages, we have a separate
THMessage type for messages sent back to GHC during TH execution.  At
the same time I also removed the QDone/QFailed/QException messages
into their own type, and made the result type of RunTH more accurate.

Test Plan: validate

Reviewers: goldfire, ezyang, austin, niteria, bgamari, erikd

Subscribers: thomie, goldfire

Differential Revision: https://phabricator.haskell.org/D2356