Fix issue with unlink teardown behavior

[codemercenary]

This pull request highlights a problem with the way that unlink teardown occurs, specifically during signal unregistration. Fix this problem.

• Fixes Service crash occurs after sleep/resume #762

[codemercenary]

Any questions on this one @veronicaz41?

[veronicaz41]

Still looking at this. Will ask you on slack if I have question.

[veronicaz41]

Ah, in this case AutoSignalTest.OneRemovesAll fails. But why we want clearall run twice? Would it run once be a preferred behavior?
Am I missing something?

[codemercenary]

That's the image guarantee. If I have signal B, and I assert B, then all handlers on B at the time of the call must be asserted, even if one of the handlers removes another handler.

The reason for this is that I don't want nondeterministic cases where the order where signal handlers are called affects how many signal handlers are called.

[veronicaz41]

Make sense 👍 Ready for review.

[codemercenary]

This solution is correct, excellent work. Can you find a way to implement it that does not require any modification to signal.h? I am trying to prevent that type from getting any more complex, it's already pretty hard to follow as it is.

[veronicaz41]

ok

[codemercenary]

I'm not so sure this is the right answer anymore. Comparing to m_state directly like this is pretty unsafe unless you're doing it in a loop, which we are not.