llvm dev - Jun 2018 - libFuzzer signal handling tests

Hi,

Now that we include libFuzzer tests into check-all we might want to take another
look at flaky ones.

For us, sigusr.test and merge-sigusr.test tend to fail every now and then.
Those tests send a signal to a fuzzing target which sleeps for 1 second on every
invocation,
and check that the signal is processed and sigaction callback runs within 2
seconds.

In my understanding there’s no guarantee that the signal would be handled within
2 seconds.
I’m marking them as unsupported on Darwin,
but I’m surprised that there aren’t similar issues on Linux.
Could they be rewritten in a more stable way [I’m really not sure how] or
removed altogether?

George

+Matt Morehouse <mascasa at google.com>

On Wed, Jun 27, 2018 at 4:14 PM George Karpenkov <ekarpenkov at apple.com>
wrote:
> Hi,
>
> Now that we include libFuzzer tests into check-all we might want to take
> another look at flaky ones.
>
> For us, sigusr.test and merge-sigusr.test tend to fail every now and then.
> Those tests send a signal to a fuzzing target which sleeps for 1 second on
> every invocation,
> and check that the signal is processed and sigaction callback runs within
> 2 seconds.
>
> In my understanding there’s no guarantee that the signal would be handled
> within 2 seconds.
> I’m marking them as unsupported on Darwin,
> but I’m surprised that there aren’t similar issues on Linux.
> Could they be rewritten in a more stable way [I’m really not sure how] or
> removed altogether?
>
> George
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20180627/8735fe09/attachment.html>