uopz.exit=0 and never return type

[vjik]

PHP code:

function x(): never { exit(0); };
x();

Throws error:

Fatal error: Uncaught TypeError: x(): never-returning function must not implicitly return in Command line code:1
Stack trace:
#0 Command line code(1): x()
#1 {main}
  thrown in Command line code on line 1

[blueblakk]

I'm getting a similar error. I cannot use PHPUnit because of this.

Only after I remove uopz from my docker environment does it work properly.

This was tested with PHP 8.2.8 in a Docker environment on Docker Engine v24.0.2, Windows 10.

I don't seem to have this issue running Docker on MacOS Ventura.

[kornrunner]

This is actually present since PHP 8.1 if you have configured uopz.exit=0.

I've submitted a PR which should sort it, but in case it doesn't get merged you could either set uopz.exit=1 or use my fork. Hope this helps, thanks.

[cmb69]

This is a somewhat philosophical question: if exit does not stop program execution, is the never return type actually appropriate?

[kornrunner]

uopz_allow_exit was initially implemented to allow testing code that does exit() (otherwise hitting exit terminates test execution as well).

On the other hand, never return type (which appeared later) is there to let people know that script will exit after invocation, so - from that point it's appropriate and useful.

But again, while testing - if we disable exit in order to test code that might exit- we'll end up with errors if method also has (appropriate/expected) never return type, so ... thus this issue and my PR.

Perhaps another solution would be to use uopz to modify return types of specific methods / functions. I'm willing to update the PR if there are any issues with it. Thanks!

[cmb69]

I understand what this ticket is about, but I'm still not sure if we should try to "fix" the issue, for the following reasons:

• disabling ZEND_EXIT might not be the best idea anyway (see e.g. Print a warning when exit is ignored #174)
• can we be sure that the ZEND_VERIFY_NEVER_TYPE opcode is always emitted?
• could there be issues with the likely passing https://wiki.php.net/rfc/exit-as-function?
• and maybe more

My point is that I don't like to add yet another feature, that might break in the future (or might even not work under all conditions right now), especially if that feature might not be that useful. I mean, if uopz.exit=1 solves the problem, there is no issue.

Note that I'm not strictly against this change.

@remicollet, what do you think?