Allow to not close stream on rscr dtor in php cli sapi

[bukka]

This change allows not closing stream handle when resource dtor is being done. Fixes#8827

[cmb69]

Ah, nice, thank you! I assume it is best to apply this to "master" only, and revert the changes to PHP-8.0 and PHP-8.1 altogether.

[bukka]

@cmb69 yes agreed this should go to master only and we should revert the previous changes in 8.0 and 8.1 before the rc is tagged. I will need to do a bit more testing first and look to the possible edge cases but should be hopefully ready in the next few weeks.

[arnaud-lb]

Thank you @bukka@cmb69

Should we also add some tests for this ? #8827 includes one that checks that STDERR can be closed

[moonlitbugs]

@arnaud-lb , The test cited in #8827 only checks whether STDERR can be closed from within PHP. The STDERR global shows itself closed. But that is not the problem. The problem is that, when STDERR is closed, file descriptor 2 is open. Ditto STDIN/STDOUT.

As I noted in #8827, php-8.1.8 leaves /dev/fd/{0,1,2} open post-fclose(), despite the NEWS claiming otherwise.

[fds242]

@moonlitbugs

The test cited in #8827 only checks whether STDERR can be closed from within PHP.

I fail to see what your issue is with the simple test I provided, since it continues to demonstrate the issue's existence in PHP 8.1.8, just as well as it did in 8.1.7, and also correctly pointing out every version which functioned correctly in the past. It cares not what the fclose() call returns, and instead tests if it php://stderr can be immediately reopened after the close attempt. Which it can, in the broken versions.
In any case, you're welcome to provide a better test.

I chose to test STDERR since what else could be reasonably tested in both a .phpt test and on ev4l.org? You don't control stdin; and stdout should not be messed with, since if you are able to successfully close it, you have also lost the ability for the test to output any results. As the bug has always equally affected all three std handles by design, seeing if the issue exists for stderr ought to be enough in this case.

[moonlitbugs]

@moonlitbugs

I fail to see what your issue is with the simple test I provided, since it continues to demonstrate the issue's existence in PHP 8.1.8, just as well as it did in 8.1.7

Ah, then I spoke too soon, my bad. I wrote that, as I remember that closing STDERR succeeds, var_dump() confirms, etc., yet the file descriptor itself is still open. Attempts at reference would return false, or throw an error. I was thinking that an external check of the file descriptor (system("lsof ...")) would be necessary. But I can see now that fopen("php://stderr") could work.

[bukka]

I just realised that I haven't pushed that test so pushed now. It's still not completely ready as I need to check some edge cases.