[patch] Fix BZ#20544 (assert function passed to atexit/on_exit/__cxa_atexit != NULL)

[Florian Weimer]

* Adhemerval Zanella:

>> Yes, but all glibc targets currently behave the same.  I don't expect
>> this to change because the a lot of things are hard-coded (see the
>> JRuby/JVM discussion on reporting signal termination).
>
> Do you have a link for the discussion?

I think this it:

<http://mail.openjdk.java.net/pipermail/hotspot-runtime-dev/2018-June/028864.html>

In particular this:

<http://mail.openjdk.java.net/pipermail/hotspot-runtime-dev/2018-June/028885.html>

Which mentions:

  Shutdown.exit(sig.getNumber() + 0200);

>> We should probably add macros to go into the other direction to
>> <sys/wait.h>, given that the layout is unlikely to change, ever.
>
> Is is something usually required and/or reimplemented elsewhere?

I think it would be useful for writing tests at the very least.

Thanks,
Florian