Also disable parallelism during tests. The test suite
is not broken, per se, but the output is not synced,
so in the event of a failure the logs would be garbled.

Ref: #1233

Zach van Rijn authored 7 months ago and Anna Wilcox committed 7 months ago

This ensures the Mozilla build system can find tools like llvm-objcopy
without the desired LLVM version being the default one.

Closes: #1230

Zach van Rijn authored 7 months ago and Anna Wilcox committed 7 months ago

This ensures Qt Creator picks up the LLVM version we have selected for
it, instead of looking for a non-version-suffixed `llvm-config`.

Closes: #1231

Zach van Rijn authored 7 months ago and Anna Wilcox committed 7 months ago

Closes: #796

The LLVM version is now a variable that can be easily changed, as well.

Closes: #1225

Closes: #1228

Zach van Rijn authored 8 months ago and Anna Wilcox committed 7 months ago

Closes: #1224

Zach van Rijn authored 9 months ago and Anna Wilcox committed 7 months ago

Closes: #1197

Zach van Rijn authored 9 months ago and Anna Wilcox committed 7 months ago

Default optimization settings between GCC 8.5.0 and 13.3.0 appear
to have caused this test to fail; disable them for this test.

Closes: #1195

* The san library needs some adjustments to handle the fact we need
  frame pointers (r7 is used for the syscall number, and the FP).

* Fix GCC Go __go_mmap external linking since they changed the signature
  upstream for the C side.  This closes #1201.

Closes: #1222

Required for Erlang/OTP 27 support.

Closes: #1182, #1194

We can update this once we have newer Rust.

Ref: #1176

Closes: #1174

Closes: #1217, #1218

Backport the GCC miscompile workaround from LLVM 18.

Use the old x87 excess precision behaviour to fix most of the tests.
Still no idea why 32-bit Intel wants new shiny colours for the graph.

Closes: #1200, #1202, #1204, #1223

Zach van Rijn authored 9 months ago and Anna Wilcox committed 7 months ago

Closes: #1198

Closes: #1207

Closes: #1221, #1206

Based-ish on Chimera's patches, but a bit more thorough.

* 32-bit ports don't have liblsan or libtsan.
* We can't dep on utmps-dev for utmpx, so hard-code size.
  The size is the same on all six arches per my testing.

Closes: #1181

* Parallel build on as many cores as normal builds use.
* The module cache is contained within the builddir instead of polluting HOME.

The test data has changed upstream.  We use our locally cached version
for now, since we haven't updated GNU Cobol itself yet.