Skip bailout for unwind and graceful exception from zend_throw_exception_internal()

[iluuu1994]

Fixes GH-18619

[bwoebi]

Ummm, I thought calling exit in register_shutdown_function was the approved way of stopping to process shutdown handlers / destructors?

[iluuu1994]

https://www.php.net/manual/en/function.register-shutdown-function.php

If you call exit() within one registered shutdown function, processing will stop completely and no other registered shutdown functions will be called.

You are right. I was not aware of this rule. I presume the same should apply to error handlers?

[bwoebi]

Yes, I'd say so. After all exception handlers are sort of processed as part of the relevant shutdown handlers.

And just for the record, destructors say the same thing:

The destructor will be called even if script execution is stopped using exit(). Calling exit() in a destructor will prevent the remaining shutdown routines from executing.
https://www.php.net/manual/en/language.oop5.decon.php#language.oop5.decon.destructor

[iluuu1994]

Ok, TIL. The test is still wrong, as in that case the shutdown handler shouldn't run. So let me change the patch.

[bwoebi]

@iluuu1994 Oh, the Zend/tests/gh18619.phpt is right. This runs before the shutdown sequence. I thought you meant exit; calls happening in exception handlers executed in response to exceptions thrown in register_shutdown_function.

[iluuu1994]

But then Zend/tests/bug62763.phpt looks wrong too, though. Or does this imply that exit() from shutdown handlers just stops further shutdown handlers, but not destructors? That doesn't make a lot of sense imo.

[bwoebi]

Yes, I agree that this is weird. The wording in the docs sort of support both interpretations:

processing will stop completely [... okay, no destructors either?!] and no other registered shutdown functions will be called [huh, only no other shutdown handlers, but also destructors???].

So, the docs should probably be a bit less ambiguous in either direction and the implementation should follow that.
I think I agree with the interpretation that it also should prevent destructors.
But changing that might be PHP-8.5 stuff rather than 8.3.