[PATCH] D95159: [ASTReader] Allow controlling separately whether validation should be disabled for a PCH vs a module file

Thu Jul 22 13:48:55 PDT 2021

mibintc added a subscriber: MaskRay.
mibintc added a comment.

More info about the test failure I'm seeing for clang/test/Index/preamble-reparse-changed-module.m

It appears it might be a non-deterministic failure, perhaps some test dependencies are missing?

Yesterday before I pushed my clang patch, I saw it fail when I used "ninja check-clang".  I quickly executed each step used by this test (llvm-lit -v ; then save all the RUN commands to a file ; then "source" the file -- nothing was printed to stdout and also the return code was 0).  I thought the fail report wasn't real at this point and I pushed the patch.  This was a Debug build.

Then I saw a bot failure, @MaskRay pinged my clang patch and said my patch had caused this test to fail.  He told me there was an assertion fail.

Then I created a new build area, configured with Debug and "enable llvm assertions", and "ninja check-clang".  This time 0 tests reported as fail.

Then I went back into the build-debug build where "ninja check-clang" had reported the fail and used gdb to debug.  In the course of debugging gdb told me that the build was out-of-date w.r.t. the source file, so I exitted gdb, "ninja clean" "ninja check-clang".  But now no tests are failing.  I don't know why gdb told me the build was out of date, I believe I did not change any of the files but I'm not 100% certain.

I removed the patch and created another build area, and rebuilt with Debug and asserts, everything is clean there too.

I captured the stdout from executing c-index-test, there is a difference between patched and unpatched. The __FLT_EVAL_METHOD__ line is not present in the patched, this is expected.

Can you suggest how to find the problem?  Thanks a lot.

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D95159/new/

https://reviews.llvm.org/D95159