[PATCH] D115232: [clangd] Indexing of standard library

[Sam McCall via Phabricator via cfe-commits]

sammccall added a comment.

I managed to get a stack trace from a bot <https://lab.llvm.org/buildbot/#/builders/123/builds/10796> (by leaving the broken commit up for longer this time).

In D115232#3522598 <https://reviews.llvm.org/D115232#3522598>, @thakis wrote:

> In D115232#3522571 <https://reviews.llvm.org/D115232#3522571>, @sammccall wrote:
>
>> In D115232#3522514 <https://reviews.llvm.org/D115232#3522514>, @thakis wrote:
>>
>>> In D115232#3520461 <https://reviews.llvm.org/D115232#3520461>, @sammccall wrote:
>>>
>>>> Hmm, the test keeps crashing on the GN bot: http://45.33.8.238/win/58316/step_9.txt
>>>> Unfortunately the stacktrace is not symbolized, and I'm not seeing this elsewhere (e.g. premerge bot).
>>>>
>>>> @thakis, any idea why unittests no longer manage to symbolize stack traces on crash on the windows bot? I believe this used to work...
>>>
>>> I do not know. Maybe related to the "run many unit tests in a single process" lit change from a month ago?
>>
>> I suspected that, and verified locally that:
>>
>> - llvm-symbolizer on PATH still works
>> - LLVM_SYMBOLIZER_PATH env variable didn't work, but I fixed it in 1236b66a98197109ed40141329d6056dfbe25967 <https://reviews.llvm.org/rG1236b66a98197109ed40141329d6056dfbe25967> along with this reland, still no dice.
>
> (My bot neither has llvm-symbolizer on path, nor sets LLVM_SYMBOLIZER_PATH fwiw.)

Is this something you could add? I'd much rather revert quickly when after seeing a problem than wait for the slower official bots to catch it, but the stacktrace is pretty key.

>>> Anyways, looks like this relanded and broke tests yet again. Maybe find a win box before relanding the next time?
>>
>> I found one, but the test doesn't crash (nor on the premerge bots).
>>
>> I'll revert again, but I have no idea how to proceed. Only this bot and `llvm-avr-linux` show the failure, and neither of them have a working symbolizer.
>
> I wouldn't be super surprised if this is related to windows and delayed template parsing.

It seems possible. The other other failure I'd seen was the `avr-linux` bot (not windows).
However it definitely passes in some configurations: both the premerge tests <https://buildkite.com/llvm-project/premerge-checks/builds/93301> and my local build.

> I haven't found any bots on llvm's official waterfall that run ClangdTests.exe – maybe that's why there aren't more bots finding this?

clang-x64-windows-msvc does, but it's slow.

> I'd recommend building and running the test on a win box and see if it repros locally.

Again, I have done this, and cannot reproduce (on 19.29.30038.1).

Now I have some idea where the crash is I can try some blind fixes, though.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D115232/new/

https://reviews.llvm.org/D115232