[ndossche]

This properly fixes the entropy computation to take into account function ordering and module versions. This is basically a reintroduction of the previously removed code from GH-5871, but a simpler version. Since this runs once every module startup, the overhead is negligible as it does not influence the request.

I chose this simple approach rather than the suggestion to introduce multiple system entropies based on shm/file-cache.

This PR also simplifies some other parts, since we can get rid of some workarounds.

[Enfiladeguide]

Quoted text written with an LLM and checked by a human developer.

I ran this branch against the FPM setup from #24661. gh17733.phpt passes with the patch and fails without it, but the FPM case is unchanged: two pools under one master, pool a with php_admin_value[disable_functions] = exec, pool b without. t.php (echo var_export([1, 'x' => true], true), "\n";), requested through a and then through b:

pool a: array (   0 => 1,   'x' => true, )
pool b: array(2) {   [0]=>   int(1)   ["x"]=>   bool(true) } bool(true)

The output is the same on PHP-8.4 at 51808ac and on this branch at 18794dc (both built with --disable-all --enable-fpm --enable-opcache, opcache loaded with default settings; newlines shown as spaces).

As far as I can see, the system ID can't catch this case. It is finalized once in the master, in php_module_startup(), while a pool's disable_functions is applied later in each worker (fpm_php_init_child() → zend_disable_functions()). On Linux the shared-memory cache doesn't check the system ID at all.