This change caused the Import/namespace/struct-and-var/test.cpp test to fail on ARM due to an extra line with `-CXXRecordDecl being emitted by the compiler that was being matched instead of the intended line. I checked in a fix to tighten up the check a little more so that it gets the correct line in 757bc55. I don't think it should negatively affect the test, but please do review the change.

In D70348#1748580, @craig.topper wrote:

You could probably just define add ORBIS to every line and then rely on the individual feature defines at the end of the line to re-enable the ones that are really supported. I assume the target-cpu is set to btver for ps4?

I tried applying this to our internal sources and it worked, so LGTM as well.

Sorry for breaking this, I thought the change would be trivial, guess it was not. From what I can remember, Peter's change was causing our internal testing to break because we don't build/use LLD for our platform.

LGTM

Is someone working on a fix for this that will go in soon? I am about to revert this change because it is blocking our internal builds as well as breaking both of the public PS4 bots.

In D64931#1647880, @dyung wrote:

Hi, this is causing 3 test failures on the PS4 linux bot. The changes may not have been initially flagged because a different issue was causing a build failure which masked the problem. I have bisected the test failures to this change though.

http://lab.llvm.org:8011/builders/llvm-clang-lld-x86_64-scei-ps4-ubuntu-fast/builds/53916

LLVM :: LTO/Resolution/X86/not-prevailing-weak-aliasee.ll
LLVM :: ThinLTO/X86/printer.ll
lld :: ELF/lto/drop-debug-info.ll

Hi, this is causing 3 test failures on the PS4 linux bot. The changes may not have been initially flagged because a different issue was causing a build failure which masked the problem. I have bisected the test failures to this change though.

In D66682#1644676, @sylvestre.ledru wrote:

@tejohnson Nested if ? :)

CMake Error at projects/compiler-rt/lib/gwp_asan/CMakeLists.txt:108 (if):
  if given arguments:

    "COMPILER_RT_BUILD_LIBFUZZER" "AND" "(" "GNU" "STREQUAL" "Clang" ")" "AND" "CMAKE_CXX_COMPILER_VERSION" "VERSION_GREATER_EQUAL" "6.0"

  Unknown arguments specified

Taking a step back and thinking about this, the compile command uses "-S", but explicitly specifies the output file to be "%t/module.pcm.o" (even though it is generating an assembly file instead of an object file), could we change the test to just check for "-o {{.*}}module.pcm.o "? That would work in all cases I think except for the extreme corner case where that string was part of the build directory name.

It's probably a pathological case, but wouldn't this still fail if the build directory contained ".s" followed by a space in the name? Something like "build-foo.s bar"?

In D65974#1623368, @thakis wrote:

Something like "fix test for out-of-tree targets that don't enable the internal assembler by default by explicitly passing -fintegrated-as everywhere" or something like that.

Actually it seems your original change already implemented that, so I think only the updated test change that I made is needed. Let me know if you disagree with that.

Oh, I just re-read your comments and realized that we should claim the Wa_COMMA and Xassembler arguments, I'll reinstate the change to just claim those.

In D65974#1623140, @thakis wrote:

Thanks! Looks like the change to the cpp file got lost? I only see the test change now.

Reverted changes in Clang.cpp and updated test to expect warning when an external assembler is requested.

I'm not at a Windows machine at the moment, so I can't test your fix, but I'm doubtful that it would actually work.

Adding James and Chris as reviews as it was suggested they might have an interest in this change.

I applied your patch and built it on our internal build bot that failed and these changes worked, so LGTM as well!

In D63600#1554029, @rnk wrote:

This causes the test to fail on Windows:
http://lab.llvm.org:8011/builders/clang-x64-windows-msvc/builds/7712

There seems to be something wrong with the blamelist, so it didn't send email. I see this in the log from the previous build on that bot:

Updating llvm to 363241 at ...

But buildbot thinks the previous build was testing r364000, so it didn't send any notifications.

printf seems like the wrong command to use to test backslash handling, since it interprets them as escapes. %S expands to a path containing backslashes, and the test logs seem consistent with that theory. I'll go ahead and revert this.

In D63105#1544519, @nickdesaulniers wrote:

There are lots of tests under clang/test/ that target arm-- as a -triple. How is it that they don't fail in the same way? %clang_cc1 -triple arm-- is somehow different?

Hi Ziang,

Hi, the test cxx2a-consteval.cpp that you added in this commit is failing on the PS4 Windows bot.

In D61187#1486249, @probinson wrote:

FYI, we had to disable clangd entirely after this patch, getting a weird problem with lit that we haven't figured out yet.
In case anybody else has seen something like this, and knows what's going on, please let us know.

  Running all regression tests
  llvm-lit.py: C:/j/w/opensource/opensource_to_sce_merge__2/o/llvm\utils\lit\lit\llvm\config.py:336: fatal: couldn't find 'clang' program, try setting CLANG in your environment
C:\Program Files (x86)\MSBuild\Microsoft.Cpp\v4.0\V140\Microsoft.CppCommon.targets(171,5): error MSB6006: "cmd.exe" exited with code 2. [C:\j\w\opensource\opensource_to_sce_merge__2\build\check-all.vcxproj]