FreeBSD LGTM

This review was just used for triggering and testing FreeBSD CI; actual change committed via D128084.

regen

XFAIL removal in libcxx/test/libcxx/modules_include.sh.cpp is a separate commit

regen

Seeing an occasional timeout in libcxx/test/std/atomics/atomics.types.operations/atomics.types.operations.wait/atomic_wait_explicit.pass.cpp or other atomic tests

try reverting umtx changes

rebase

remove C++2b and Modular build, which are failing apparently unrelated to my work

It's currently failing in c++2b before getting to my job:

/home/libcxx-builder/.buildkite-agent/builds/0306c2d9b628-1/llvm-project/libcxx-ci/build/generic-cxx2b/include/c++/v1/bitset:396:63: error: shift count >= width of type [-Werror,-Wshift-count-overflow]
        __r |= static_cast<unsigned long long>(__first_[__i]) << (sizeof(__storage_type) * CHAR_BIT);
                                                              ^  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

rebase

regen

regen

adjust xfails

I was away for a bit, and will get back to this. I've pushed a rebased test version to D140050 to check for any new regressions, and if it looks good I will push this.

rebase

LGTM for FreeBSD

In D142422#4080460, @Mordante wrote:

Does this still work on FreeBSD platforms where SIZEOF_LONG != 8 ?

One more FreeBSD libc++ change is in D142422

Can you rebase onto main? I don't think you had D141294 in your tree.

rebase

I've been using D140050 as a scratch-pad to iterate on testing (in an attempt to avoid spamming everyone on the subscriber list here) and will keep it updated regularly until this lands. I did submit many runs with an additional change that removed the other, non-FreeBSD CI targets to speed iteration and avoid unnecessary CI cycles.

rebase

rebase, apply https://reviews.freebsd.org/D38132

Everything should be ready now, test run in progress at https://buildkite.com/llvm-project/libcxx-ci/builds/17740

update to final candidate version

IMO we should fix these issues, but I rather have a CI that tests most of libc++ instead of all of FreeBSD.

rebase after D141294

I applied this change and restored those tests and (in the one run so far) they passed.

FreeBSD CI looks good with the four reviews in the stack
D141294, D142134, D141542, D128084

In D142134#4066210, @Mordante wrote:

Thanks for working on this!

In fact Kostik found this independent of my work on bringing up CI. It's fortuitous timing.

LGTM assuming doc update

rebase

no hard tabs

rebase, add tests back, add kib's atomic patch

This is one of two outstanding issues I need to figure out to get the FreeBSD CI runner going

rebase, restore clang-tidy

LGTM if the runner in D128084 is happy.

• XFAIL adjustments are in D141542
• I've temporarily removed the clang-tidy bits in my local tree as there is an issue that needs to be investigated/resolved
• A number of tests either had inconsistent results or timeouts and I have excluded them temporarily
  • atomic_wait.pass.cpp
  • atomic_wait_explicit.pass.cpp
  • atomic_notify_one.pass.cpp
  • thread.lock.shared.locking/lock.pass.cpp

Update & rebase after committing a number of related changes, and limit this review to the CI portion only

rebase, remove a test that timed out on FreeBSD

rebase

rebase

use // ADDITIONAL_COMPILE_FLAGS: -Wno-macro-redefined

retest flakey tests

test without _LIBCPP_DEPRECATED_ABI_DISABLE_PAIR_TRIVIAL_COPY_CTOR change

rebase, skip all non-FreeBSD for testing

rebase

fixup pipeline

rebase

sorting & curating patches

Also remove unused ABI list from local development environment.

rebase after ABI list change put in D141496

With (quite) a few hacks I finally managed to get a green FreeBSD run
https://buildkite.com/llvm-project/libcxx-ci/builds/17075#01859d7d-2444-4152-a2e6-a4e41c3912fc

tweak symbols again, rebase

rebase and tweak symbols again

Rebase, update symbols

fix python

rebase, tweak symbol list

add xfails to time tests that started failing (again?)