This could be a nice incentive to report analyzer regressions to the LLVM folks.

dpranke: Is there a way we can turn the bot red, short of returning an error code from a build task & stopping the entire build? I'd like to avoid that.

For the `compile` step, currently the only mechanism we have to tell the build that something passed or failed is to look at the return code.

If you wanted to do something else (like fail based on the contents of stdout), you'd need to write something that wrapped the invocation of ninja, parsed it, and then returned a different exit code. 

Somewhat coincidentally, we're actually working on a wrapper that does that but for different reasons, so I don't know if that would be useful here.

Alternatively, you could periodically fetch the logs for those steps afterwards and scan them for errors.

A third option would be to change the clang invocation to invoke a wrapper that captured the output and failed the individual compile command. I'm not actually sure why the crashes you're seeing now aren't already failing the compile?

The wrapper script does a regular compile regardless of the analysis output, since as you can see, the analyzer is a bit more fragile/regression-prone. Only the return code from the second compile step is propagated.

Issue has not been modified or commented on in the last 365 days, please re-open or file a new bug if this is still an issue.

For more details visit https://www.chromium.org/issue-tracking/autotriage - Your friendly Sheriffbot