[libcxx] Run the clang-format and generated-output checks on the "service" queue

[• Quuxplusone]

Summary

As these jobs only run in a couple seconds, and block starting of
other jobs, they can run on the "service" queue which doesn't get
blocked by other long running jobs.

[• Quuxplusone]

Sounds very good to me if it works as advertised. I don't know enough about buildkite to know if it works as advertised, though. :)

(FWIW, I've also occasionally wished that one or two of the builders were pulled up into a new section, e.g. run just the C++03 and C++20 tests and then - wait before running the rest — this would lighten the load due to PRs-that-would-ultimately-fail-anyway by a factor of like 3.5, admittedly at the cost of some latency. Just leaving that here in case it resonates with someone.)

[Mordante]

LGTM, sounds great if we can the results of these test quickly!

In D101437#2722723, @Quuxplusone wrote:

(FWIW, I've also occasionally wished that one or two of the builders were pulled up into a new section, e.g. run just the C++03 and C++20 tests and then - wait before running the rest — this would lighten the load due to PRs-that-would-ultimately-fail-anyway by a factor of like 3.5, admittedly at the cost of some latency. Just leaving that here in case it resonates with someone.)

Sometimes I would like this. On the other hand I feel it's good to get the results of all supported C++ versions directly. I wouldn't object to another section to split the Apple builds in 2 parts. Apple still seems to be the limiting factor for our builds.

[mstorsjo]

Rerunning to hopefully get a clean CI record without spurious unrelated errors.

[mstorsjo]

Surprisingly, the check-generated-output testcase failed this time around, while it didn't yesterday. Maybe it does rely on something that not all workers in the "service" worker pool have installed? @Mordante
do you happen to be able to figure out what went wrong in this CI run?

Then on the other hand, as the whole next block of jobs (C++03..C++20) all run on the "libcxx-builders" queue, I don't think there'll be any practical difference for how quick we get to the final block of jobs with this change...

[• Quuxplusone]

Surprisingly, the check-generated-output testcase failed this time around, while it didn't yesterday. Maybe it does rely on something that not all workers in the "service" worker pool have installed?

Link to logs: https://buildkite.com/llvm-project/libcxx-ci/builds/2879#f4496127-b31e-4e94-b3b8-cec119946838
The raw log output starts like this:

+ grep -q '^--- a' /var/lib/buildkite-agent/builds/llvm-project/build/check-generated-output/generated_output.patch
+ grep -rn '[^ -~]' libcxx/include/
libcxx/include/concepts:1:// -*- C++ -*-
libcxx/include/concepts:2://===-------------------------- concepts ----------------------------------===//
libcxx/include/concepts:4:// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
libcxx/include/concepts:5:// See https://llvm.org/LICENSE.txt for license information.
libcxx/include/concepts:6:// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
libcxx/include/concepts:10:#ifndef _LIBCPP_CONCEPTS
libcxx/include/concepts:11:#define _LIBCPP_CONCEPTS
libcxx/include/concepts:14:    concepts synopsis

...and so on.
This is the same output I get locally from running grep -rn '[ ^-~]' ../libcxx/include/concepts (notice the swapping of ^ and space).
Perhaps this is a known bug?

@mstorsjo, I'm going to commandeer to add grep --version to the script and keep poking buildkite until it fails again; with a --version in hand, we can at least file a StackOverflow question and see if anyone's ever seen this issue before.

[• Quuxplusone]

add grep --version (printf debugging ftw!)

[mstorsjo]

In D101437#2725481, @Quuxplusone wrote:

@mstorsjo, I'm going to commandeer to add grep --version to the script and keep poking buildkite until it fails again; with a --version in hand, we can at least file a StackOverflow question and see if anyone's ever seen this issue before.

Thanks! Please do.

Btw, for quicker CI iteration/testing, you can remove (or comment out) all the other CI jobs in the patch - even though it adds a bit more noise to the patch and mailing list history.

[• Quuxplusone]

grep --version on the normal queue. Shoulda done this the first time.

[Mordante]

In D101437#2725330, @mstorsjo wrote:

Surprisingly, the check-generated-output testcase failed this time around, while it didn't yesterday. Maybe it does rely on something that not all workers in the "service" worker pool have installed? @Mordante
do you happen to be able to figure out what went wrong in this CI run?

This looks weird. I've never seen anything like this before.
Thanks Arthur for looking into it!

[• Quuxplusone]

https://stackoverflow.com/questions/67320156/misbehavior-of-gnu-grep-when-grepping-for-ignores-spaces/

StackOverflow says "check your locale" — it's possible that we're grepping in a locale where in the locale's collation order e.g. the letter a sorts first, then space, then ~, so in that collation order it is true that a matches [^ -~]. Using the ASCII locale LANG=C should fix that.

If it does fix it, imma land this.

[• Quuxplusone]

Try setting LANG=C a different way, at the top of this script.
Roll in some <iterator> cleanup just to get testing on it.

[• Quuxplusone]

Try some more printf debugging. I don't suppose there's a way to ssh into this machine...? :P

[• Quuxplusone]

continue printf-debugging

[• Quuxplusone]

Turns out that the "service" machine had LC_ALL=en_US.UTF-8 and LC_COLLATE=en_US.UTF-8, even though it had LANG=.
Let's try unsetting those two env vars.

[ldionne]

The service queue isn't meant for that. It's meant for BuildKite infrastructure jobs only. If we want to reduce the latency for getting the result of those jobs, we can investigate other options like agent priorities or setting up a special queue in amongst the libc++ agents themselves.

Also, please ping me on those infrastructure related reviews, I never got a chance to see this before it got shipped.

[• Quuxplusone]

@ldionne: Sorry; okay.
I'm going to recommit the LC_COLLATE part of this patch for tech debt's sake, and then put the rest out of my mind. But I do think we (by which I mean you and @mstorjo) should look into setting up a special queue, because this really did seem to speed up the builds.

[ldionne]

In D101437#2729911, @Quuxplusone wrote:

@ldionne: Sorry; okay.
I'm going to recommit the LC_COLLATE part of this patch for tech debt's sake, and then put the rest out of my mind. But I do think we (by which I mean you and @mstorjo) should look into setting up a special queue, because this really did seem to speed up the builds.

Yeah, no worries.

libcxx/include/iterator

 template <class T, size_t N> constexpr T* data(T (&array)[N]) noexcept;           // C++17
 template <class E> constexpr const E* data(initializer_list<E> il) noexcept;      // C++17
 
 }  // std
 
 */
 
 #include <__config>
-#include <iosfwd> // for forward declarations of vector and string.
+#include <__debug>
 #include <__functional_base>
-#include <type_traits>
-#include <compare>
-#include <concepts> // Mandated by the Standard.
-#include <cstddef>
-#include <initializer_list>
 #include <__iterator/concepts.h>
 #include <__iterator/incrementable_traits.h>
 #include <__iterator/iter_move.h>
 #include <__iterator/iterator_traits.h>
 #include <__iterator/readable_traits.h>
-#include <__iterator/concepts.h>
 #include <__memory/addressof.h>
 #include <__memory/pointer_traits.h>
+#include <compare>
+#include <concepts> // Mandated by the Standard.
+#include <cstddef>
+#include <initializer_list>
+#include <iosfwd> // for forward declarations of vector and string
+#include <type_traits>
 #include <version>
 
-#include <__debug>
-
 #if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
 #pragma GCC system_header
 #endif
 
 _LIBCPP_BEGIN_NAMESPACE_STD
 
 template<class _Category, class _Tp, class _Distance = ptrdiff_t,
          class _Pointer = _Tp*, class _Reference = _Tp&>
#else
     typename __wrap_iter<_Iter1>::difference_type
     operator-(const __wrap_iter<_Iter1>&, const __wrap_iter<_Iter2>&) _NOEXCEPT;
 #endif
 
     template <class _Iter1>
     _LIBCPP_CONSTEXPR_IF_NODEBUG friend
     __wrap_iter<_Iter1>
     operator+(typename __wrap_iter<_Iter1>::difference_type, __wrap_iter<_Iter1>) _NOEXCEPT;
-
-    template <class _Ip, class _Op> friend _LIBCPP_CONSTEXPR_AFTER_CXX17 _Op copy(_Ip, _Ip, _Op);
-    template <class _B1, class _B2> friend _LIBCPP_CONSTEXPR_AFTER_CXX17 _B2 copy_backward(_B1, _B1, _B2);
-    template <class _Ip, class _Op> friend _LIBCPP_CONSTEXPR_AFTER_CXX17 _Op move(_Ip, _Ip, _Op);
-    template <class _B1, class _B2> friend _LIBCPP_CONSTEXPR_AFTER_CXX17 _B2 move_backward(_B1, _B1, _B2);
 };
 
 #if _LIBCPP_STD_VER <= 17
 template <class _It>
 struct __is_cpp17_contiguous_iterator<__wrap_iter<_It> > : __is_cpp17_contiguous_iterator<_It> {};
 #endif
 
 template <class _Iter>

libcxx/utils/ci/buildkite-pipeline.yml

steps:
   # Light pre-commit tests for things like formatting or when people forget
   # to update generated files.
   #
   - label: "Format"
     command: "libcxx/utils/ci/run-buildbot check-format"
     artifact_paths:
       - "**/clang-format.patch"
     agents:
-      queue: "libcxx-builders"
+      queue: "service"
     retry:
       automatic:
         - exit_status: -1  # Agent was lost
           limit: 2
     soft_fail:
         - exit_status: 1
 
   - label: "Generated output"
     command: "libcxx/utils/ci/run-buildbot check-generated-output"
     artifact_paths:
       - "**/generated_output.patch"
     agents:
-      queue: "libcxx-builders"
+      queue: "service"
     retry:
       automatic:
         - exit_status: -1  # Agent was lost
           limit: 2
 
   #
   # General testing with the default configuration, under all the supported
   # Standard modes, with Clang and GCC. This catches most issues upfront.

libcxx/utils/ci/run-buildbot

 #!/usr/bin/env bash
 #===----------------------------------------------------------------------===##
 #
 # Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 # See https://llvm.org/LICENSE.txt for license information.
 # SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 #
 #===----------------------------------------------------------------------===##
 
 set -ex
 set -o pipefail
+export LANG=C  # for grep
 
 PROGNAME="$(basename "${0}")"
 
 function usage() {
 cat <<EOF
 Usage:
 ${PROGNAME} [options] <BUILDER>
 
check-generated-output)
     echo "+++ Checking the output of the generator scripts"
     mkdir -p ${BUILD_DIR}
     # Reject patches that don't update the generated output correctly.
     python3 libcxx/utils/generate_feature_test_macro_components.py
     python3 libcxx/utils/generate_header_inclusion_tests.py
     python3 libcxx/utils/generate_header_tests.py
     git diff | tee ${BUILD_DIR}/generated_output.patch
     # Check if the diffs are empty, fail otherwise.
+    echo "hello world" > foo.txt
+    echo "héllo world" >> foo.txt
+    echo "GOODBYE" >> foo.txt
+    grep -rn '[^ -~]' foo.txt
+    LANG=C grep -rn '[^ -~]' foo.txt
+    LANG=C.UTF-8 grep -rn '[^ -~]' foo.txt
     ! grep -q '^--- a' ${BUILD_DIR}/generated_output.patch || false
     # Reject patches that introduce non-ASCII characters or hard tabs.
+    # Depends on LANG=C set at the top of this script.
     ! grep -rn '[^ -~]' libcxx/include/ || false
     # Check that no dependency cycles have been introduced.
     python3 libcxx/utils/graph_header_deps.py >/dev/null
 ;;
 generic-cxx03)
     export CC=clang
     export CXX=clang++
     clean