- User Since
- Jul 29 2013, 1:59 AM (383 w, 5 d)
I like your approach. It will indeed make it easier to add new tests and update current ones.
- Rebase with added version
Thu, Dec 3
It would be maybe possible to:
Nice findings! Apparently all of barrier, latch, semaphore and simd are missing _LIBCPP_PUSH_MACROS #include <__undef_macros>. And semaphore includes <cassert>.
OK, that's clear now. Thank you for the answer, Louis.
Wed, Dec 2
Sorry for late comments. Just stumbled upon this patch.
@MyDeveloperDay, is there anything I can do to help you on that?
LGTM. But wait for an approval from a higher instance :).
It looks pretty good for me, just minor comments.
And please put the relevant papers in the patch description, so that they appear in the commit later.
- Fix status page.
- Set -o pipefail for the whole script.
- Run git-clang-format also on libcxxabi/.
I removed the unnecessary check for > C++11 in make_implicit, but that doesn't change a lot as it's used anyway only in C++20.
Anyway, I've just seen that P0935 has been adopted as a DR applied retroactively to previous standards down to C++11. My only source tho is https://en.cppreference.com/w/cpp/numeric/random/uniform_real_distribution/uniform_real_distribution.
Should I apply it to pre-C++20 then?
- Update status page.
- Generalize tests. Clean up.
- Remove requirement for > C++11 in make_implicit.h.
Tue, Dec 1
- Fix [rand] in C++03 mode.
- Use typedef for C++03 compatibility.
- [queue] [priority.queue]
One more question, lots of stuff in <random> tests are also run in c++03. Is it expected? Should I add guards to my additions (#if TEST_STD_VER >= 11)?
- Implement: [rand.device] [rand.eng] [rand.dist.uni.int] [rand.dist.uni.real]
Mon, Nov 30
- Fix macOS ABI list.
@ldionne, I've modified the run-buildbot script (and added *.abilist to artifact_paths in buildkite-pipeline.yml) so as to run generate-cxx-abilist if check-cxx-abilist failed.
The build will fail anyway but the regenerated ABI lists will be present among the artifacts.
That's for people, like me, that don't have access to one of the platforms on which we check ABI lists.
If you prefer, I'll spin off another patch to keep things apart.
- Fix linux ABI list (again).
- Add forgotten artifact_paths.
This patch updates the notes about how to contribute to libc++ adding some missing elements (found in https://reviews.llvm.org/D92240).
- Fix linux ABI list.
- Re-generate ABI list (after a failed check) and get them as artifacts, so as to ease the development for contributors not having access to necessary platform.
- Get only clang-format.patch artifact.
@ldionne, I'll need your help to set this up. Agents need to have git-clang-format installed. I haven't tried but I guess I can't apt install clang-format in the buildkite script and it doesn't sound like a great idea anyway.
And, wouldn't it be possible to reuse the existing machinery in llvm/clang premerge checks that runs git-clang-format? I am afraid it can't be reused because it uses different CI pipeline and different agents...
Sun, Nov 29
- Bash: set -o pipefail to fail on missing git-clang-format.
It would be great if you rebased this patch to trigger current CI.
Sat, Nov 28
Could you please update the status page together with the corresponding note (https://libcxx.llvm.org/docs/Cxx2aStatus.html#note-p0202)?
- Create BUILD_DIR.
- Use tee instead of redirection.
- Wait after Format step.
- [abi] Update ABI list for linux (and changelog).
- Rebase to include changes from D92194
I modified headers in include/ just to test the CI.
Fix bad arc diff base.
- Break format for testing.
- Add buildkite job for checking formatting.
I'll see what I can do!
Fri, Nov 27
I agree with @miscco, the advantages of consistent formatting outweigh the inconveniences caused by clang-format's shortcomings.
Comments like "clang-format on/off" are one of the solutions to the problem. Another being to force line breaks with empty trailing comments.
To fix Apple back-deployment build, you'll need to mark your test with sth like:
// This test requires the dylib support introduced in D92214. // XFAIL: with_system_cxx_lib=macosx10.15 // XFAIL: with_system_cxx_lib=macosx10.14 // XFAIL: with_system_cxx_lib=macosx10.13 // XFAIL: with_system_cxx_lib=macosx10.12 // XFAIL: with_system_cxx_lib=macosx10.11 // XFAIL: with_system_cxx_lib=macosx10.10 // XFAIL: with_system_cxx_lib=macosx10.9
because this build uses a released dylibs for the given version of macOS (here 10.9) and does not use whatever changes you put in src/.
So, in your case, the dtor nor the vtable of format_error is just not in the libc++.dylib that the tests use.
Hence the error message: dyld: Symbol not found: __ZTVNSt3__112format_errorE.
FYI, https://clang.llvm.org/docs/ClangFormattedStatus.html shows the status of LLVM parts being clang-formatted.
Ok. Will wait for your patch. I just added XFAIL for macosx dylibs.
- Mark test as XFAIL on macosx10.*.
Thu, Nov 26
Ok, thanks. I still need to update ABI list and fix macOS backdeployment.
- Fix extra allocation for additional facets.
I don't know either what updates the doc on the site. Anyway, I just thought that could be nice to have it, but anyone that does substantial changes to docs would anyway run make docs-libcxx-html.