[jorgsowa]

The test checks the session lock with flock($fp, LOCK_EX | LOCK_NB) on a file opened with fopen($file, 'r'). On platforms without flock(), such as Solaris, PHP falls back to fcntl() locks. fcntl() refuses an exclusive lock on a read-only file, so the check returns false even though the lock was released. Its flock() goes through the fcntl() wrapper in ext/standard/flock_compat.c, which is why the test fails.

Should fix Solaris issue in nigthly: https://github.com/php/php-src/actions/runs/38990675359/job/117272412935

[devnexen]

but then this test will always "succeed" on solaris, flock is emulated on solaris and is in the same process as session_write_close. Probably best to skip solaris instead.

[jorgsowa]

But it's not the point of the test. It's testing ext/session, not the fopen or flock. Those handlers here are for manipulating session file.

[devnexen]

can you make rerun the CI for solaris ? it was skipped

[jorgsowa]

The fixed test doesn't fail anymore. There are still two failures related to proc_open().

[devnexen]

The fixed test doesn't fail anymore. There are still two failures related to proc_open().

likely because.

[jorgsowa]

That's true, I did not cover them. I checked only ext/session tests.

[iliaal]

This is covered by 96cbae8, which skips the test on Solaris on PHP-8.6 and master as devnexen suggested: the fcntl() emulation can't see a lock held by the same process, with 'r' or 'r+'. I think this can be closed.

[jorgsowa]

That's not true. The test passes on Solaris with my fix. The test shouldn't be skipped.

[iliaal]

It passes, but it would also pass if the save handler never released the lock: on Solaris both locks are fcntl() locks owned by the same process, so the probe can't conflict with it. Only a probe from another process can see the lock, which is why it's skipped there.