RFC: Allow calls to exit from within exit hooks

[DilumAluthge]

Fixes #31743

This is an alternative to pull request #32253. I personally prefer the approach in #32253.

This pull request allows calls to exit(n) from within exit hooks registered with Base.atexit.

From the updated docstring for Base.atexit:

• Exit hooks are allowed to call exit(n), in which case Julia will exit with
  exit code n (instead of the original exit code). If more than one exit hook
  calls exit(n), then Julia will exit with the exit code corresponding to the
  last called exit hook that calls exit(n). (Because exit hooks are called in
  LIFO order, "last called" is equivalent to "first registered".)

I have included some tests. They are located in the test/atexit.jl file.

[DilumAluthge]

One of the consequences of this pull request is that this code:

atexit(() -> exit(0))
atexit(() -> exit(5))
atexit(() -> exit(10))
exit(15)

will cause Julia to exit with exit code 0.

I'm not sure that this behavior makes sense. Is this what we want?

[DilumAluthge]

Closing in favor of #32253

[vtjnash]

FWIW, I think we should do both

[DilumAluthge]

Reopened!

[vtjnash]

There seems like a lot of tests here. How long do these take? It doesn't seem like it should need that many to fully cover the cases.

[JeffBezanson]

I like this change; it's very elegant. For reference, posix seems to allow atexit handlers to call _exit(n), which causes the process to exit immediately without running further handlers. Calling exit is undefined. So we are doing something else, but that might be ok.

[DilumAluthge]

There seems like a lot of tests here. How long do these take? It doesn't seem like it should need that many to fully cover the cases.

I've deleted a bunch of tests. There are relatively few tests now, but I think we still cover all the cases.

[DilumAluthge]

Bump @vtjnash Take a look and let me know if I’ve addressed your concerns with regards to the number of tests.

[vtjnash]

Yep. Could I ask you to rebase and push an updated copy? We had some CI issues on this run (from stuff on master at the time), and I'd like to check that it's green before merging.

[DilumAluthge]

Could I ask you to rebase and push an updated copy?

Done!

[vtjnash]

Would you mind making a quick followup PR to make these silent during a normal test run?

Suggested change

	atexit(() -> println("No error"))
	atexit(() -> "No error")

[DilumAluthge]

Yeah!