Noticed this when I tried committing r342233. A build bot failed (http://lab.llvm.org:8011/builders/clang-with-thin-lto-ubuntu/builds/12667) when running lld's "ELF/lto/cache.ll". I reverted the change, although looking at the code, I can't see any way my change could have interacted with it. The buildbot went green.
However, I then noticed that a practically identical test failed in the same manner a bit later, some time after my revert landed. See http://lab.llvm.org:8011/builders/clang-with-lto-ubuntu/builds/10313, where "tools/gold/X86/cache.ll" failed. The test passed on the next run of the buildbot, with no related changes.
Extended Description
Noticed this when I tried committing r342233. A build bot failed (http://lab.llvm.org:8011/builders/clang-with-thin-lto-ubuntu/builds/12667) when running lld's "ELF/lto/cache.ll". I reverted the change, although looking at the code, I can't see any way my change could have interacted with it. The buildbot went green.
However, I then noticed that a practically identical test failed in the same manner a bit later, some time after my revert landed. See http://lab.llvm.org:8011/builders/clang-with-lto-ubuntu/builds/10313, where "tools/gold/X86/cache.ll" failed. The test passed on the next run of the buildbot, with no related changes.
And just now the LLD test failed again: http://lab.llvm.org:8011/builders/clang-with-lto-ubuntu/builds/10315.
I'm guessing the root cause of these test failures is the same.