[Buildbot-devel] Step fails only under Buildbot

[Bob Hood]

On 3/8/2014 6:23 AM, Dustin J. Mitchell wrote:
>
> Signal 11 can be really tricky to track down, too.  It might depend on
> something as trivial as the length of an unrelated string that shifts the
> memory allocation pattern slightly.
>
> If you can reliably reproduce it within the Buildbot environment, then 
> reproducing outside that environment may not provide you with much more
> information.  Using debugging mallocs or adding debugging information as
> others have suggested may be a better approach.  The risk is that even these
> changes will make the failure go away.
>
> If that happens, fuzzing is also an option - running repeatedly with
> randomized inputs until failure.
>

Hi, Dustin.

I did as Stephen suggested, and added the GuardMalloc library to the
execution.  It produced no actual diagnostic output (beyond the header that
indicated its presence), but the step actually completed successfully with its
inclusion.

It does indeed "feel" like a pointer issue, however, as long as GuardMalloc
gets me past the step, I can let the rest of the build complete while we try
to track down the actual issue.

Let me extend my thanks to everybody for their input on this problem.