proc_open inaccessible cwd behavior change in 8.3

[divinity76]

Description

As of PHP8.3, the following code:

<?php
$descriptorspec = array();
$cwd = '/inaccessible/dir/';
$process = proc_open('/usr/bin/env php -r ""', $descriptorspec, $pipes, $cwd);
proc_close($process);

Resulted in this output:

PHP Warning:  proc_open(): posix_spawn() failed: No such file or directory in /home/hans/test/test.php on line 4
PHP Fatal error:  Uncaught TypeError: proc_close(): Argument #1 ($process) must be of type resource, false given in /home/hans/test/test.php:5
Stack trace:
#0 /home/hans/test/test.php(5): proc_close()
#1 {main}
  thrown in /home/hans/test/test.php on line 5

But in PHP<8.3, it results in no output whatsoever, and proc_open succeeds.

This behavior change breaks real code in the wild: chrome-php/wrench#20

While I think the behavior change is good, I also think a E_DEPRECATED period is warranted, something like

Requesting an inaccessible working dir ($cwd) has been deprecated and will stop working in a future version of PHP

• Also for some unknown reason, when I test this locally, i get the posix_spawn() error, but when I test on Github CI, i don't get the posix_spawn() error (but proc_open() still fails.. silently.)

PHP Version

PHP8.3.0

Operating System

Ubuntu (both 22.04 and 24.04-beta)

[bwoebi]

I suppose now it's a bit too late to change this as code likely will have to at least special case the behaviours of early PHP 8.3 versions...

[iluuu1994]

I suppose now it's a bit too late to change this as code likely will have to at least special case the behaviours of early PHP 8.3 versions...

We usually still fix such issues in patches, unless the issue is multiple years old.

This seems to be cause by behavioral differences between fork+exec vs posix_spawn (which is used since 8.3). To be more specific, fork+exec ignores the cwd if the provided path does not have the x chmod flag, while posix_spawn errors. We could avoid calling posix_spawn_file_actions_addchdir_np when the path is not executable. I'll create a PR shortly.

[bukka]

No this should not be fixed - we just need to update UPGRADING I guess. I think Windows will fail as well (might need to try) so we are actually making it more consistent.

[bukka]

Ah actually Win changes it explicitly.

[bukka]

So I guess we could maybe consider this as a bug.

@crrodriguez What do you think?

[bukka]

Ah no it's just for missing cwd. So it seems to me that Win would fail as well. I need to try it. The point is that I would consider it as a bug only if it becomes inconsistent. But if it becomes consistent with Win, then it's not a bug but just a missing UPGRADING note. In the same way as #12589

[bukka]

Just tried this on Win and it failed so not a bug. This should be just a note to the UPGRADING as we did for #12589