don't throw EOFError from sleep

[stevengj]

This prevents sleep from throwing an EOFError under rare conditions (with very short timer durations) where the Timer object is already closed before it can be signaled, which seems like a bug.

See the discourse discussion.

[vtjnash]

This is not the correct fix, but it seems like it should work around the bug in most cases

[oscardssmith]

why isn't this a correct fix?

[stevengj]

It seems like a better fix would make wait(Timer(sec)) safe, somehow.

Timer is an exported API, so ideally calling wait on a newly created Timer should not be a dangerous pattern.

[vtjnash]

Yes, that is what is wrong about this fix. The wait tries to do some fast paths before locking, but it doesn't implement those correctly, and it doesn't re-check if it is .set after !isopen

[A1-Mathematik]

julia/base/asyncevent.jl

Line 159 in 222231f

_trywait(t) || throw(EOFError())

Why is there an error throw anyways?

[lgeissbauer-btig]

@vtjnash and @stevengj I went through the code a couple of times but don't feel confident to fix this. Can you help me a little to fix it proper

[lgeissbauer-btig]

This should fix it #54955

[lgeissbauer-btig]

This should be closed because #54955 is merge