Looking at https://bugs.php.net/bug.php?id=65463
My brain is foggy on why we call zend_destroy_modules() before
zend_hash_destroy(GLOBAL_CLASS_TABLE) in zend_shutdown(). Anyone
remember the reason?
I have verified the issue described in 65463. On a SIGTERM to Apache
mid-request, the extension ends up getting unloaded before the global
class table is destroyed causing the segfault described in the bug.
-Rasmus
Rasmus Lerdorf in php.internals (Sat, 17 Aug 2013 07:37:30 -0400):
Looking at https://bugs.php.net/bug.php?id=65463
When I saw the subject, I was reminded of
https://bugs.php.net/bug.php?id=65338
That was another case of unorderly unloading modules. I thought it was a
bug in zend_string.c, but Dmitry solved it by patching opcache.
Disregard this message if it is not related at all.
Jan