Re: Unwarranted assumption in tst-waitid, or a kernel bug?

[Roland McGrath <roland at redhat dot com>]

> It dequeues the signal, this means that signal handler won't be called.
> But probably this is OK for this testcase, sigchld() does nothing
> interesting and check_sigchld() can save sigchld_info itself.

Yes, I think that is sufficient.  I guess there might be some very old
kernels that don't support sigwait, but those don't have waitid either, so
are getting glibc's poor man's waitid using wait4, where some other things
that test does might not work either.

But the test does build on configurations without SA_SIGINFO and WNOWAIT,
and those would almost certainly not have sigwait either.  Some gratuitous
sleeping and so forth is not the worst thing ever in a glibc test.

It was me who wrote this test (and the waitid kernel code), but long enough
ago that I don't feel especially authoritative about this stuff any more.


Thanks,
Roland