1.2.11 regression on Windows / cgi

[nono303]

Hi,
I recently noticed a regression on 1.2.11 when run a Runtime :

Local variables are initialized and locked with the value defined at first execution.

When modifying value in source code, the result output does not change (aka the bug that drives you crazy)
I need to use pskill php-cgi for the value to be updated as what is written in the code.

What I've checked

• php (cli) is OK with or without opcache.enable_cli
• Occurs only with php-cgi
  • with or without opcache.enable
  • with opcache.jit 1255 or off
• 1.2.10 works fine in all situation (cli, cgi, opcache.jit & opcache.enable...)
  Only the local var declared and defined in the returned function called by Runtime->run() are affected.

Find below a standalone example failing on my side

as fcgid http://127.0.0.1/parallel.php

I've checked others var which are ok:

• var & const in included file (cf. BS, $local in parallel_fct.php)
• var & const in bootstrap file (cf. FCT, $local in parallel_bs.php)
• var passed in params & context (cf. $data, $prefix in parallel.php)
  var failing
• parallel.php : $stid, "!bug_2!", "key1", "key2", "key3" (all that is defined & declared between lines 85 & 89)

Hope the description is clear enough, don’t hesitate if not or if you need more test

Own build with php 8.5.3 x64 vs18 TS via httpd mod_fcgid

• parallel.php
• parallel_bs.php
• parallel_fct.php

[nono303]

Catching a 1st chance exception with procdump -ma -e 1 -w php-cgi.exe produce Exception: C0000005.ACCESS_VIOLATION

	[Inline Frame] php8ts.dll!zend_string_hash_val(_zend_string *) Line 132	C
 	[Inline Frame] php8ts.dll!zend_hash_find(const _zend_array *) Line 2683	C
 	[Inline Frame] php8ts.dll!zend_hash_find_ptr(const _zend_array * ht, _zend_string *) Line 882	C
 	php8ts.dll!zend_is_auto_global(_zend_string * name=0x0000000000000000) Line 1987	C
 	[Inline Frame] php_parallel.dll!php_parallel_dependencies_load_globals_vars(const _zend_function *) Line 44	C
 	php_parallel.dll!php_parallel_dependencies_load_globals(const _zend_function * function=0x00000253edf9e6e0) Line 70	C
 	[Inline Frame] php_parallel.dll!php_parallel_dependencies_load(const _zend_function *) Line 138	C
 	[Inline Frame] php_parallel.dll!php_parallel_copy_function(const _zend_function *) Line 977	C
 	[Inline Frame] php_parallel.dll!php_parallel_scheduler_pop(_php_parallel_runtime_t *) Line 367	C
 	php_parallel.dll!php_parallel_thread(void * arg=0x00000253ed485300) Line 602	C
 	pthreadVC3.dll!ptw32_threadStart(void * vthreadParms=0x00000253edf89bc0) Line 233	C
 	[External Code]	

[realFlowControl]

Hey @nono303,
thanks for reaching out, I attended some conferences that's why my reply is late. I'll have a look this week.

[realFlowControl]

Hey @nono303,
I was able to reproduce the bug and created a test already, I am currently looking for fix to keep the feature ;-)

[nono303]

Hi @realFlowControl
Many Thx for your feedback & PR!
Just tried to patch 1.2.11 with PR370 but test failed on my side

FAIL Check closure cache does not return stale cached closure after file change [S:\sdk\src\pecl-parallel\tests\closures\013.phpt]

---- EXPECTED OUTPUT
FIRST
SECOND
---- ACTUAL OUTPUT
FIRST
FIRST
---- FAILED

[realFlowControl]

Yeah, CI in that draft PR is also failing. At least we have a test now 😉
I'll need to put in more thought, will keep you posted