Just noticed some segfaults during shutdown of dl()'d modules from the
CLI; this is what I see running under valgrind:
==19025== discard syms in
/usr/local/php5/lib/php/extensions/no-debug-non-zts-20040718/pdo_sqlite.so
due to munmap()
==19025== discard syms in /usr/local/lib/libsqlite3.so.0.8.6 due to munmap()
==19025== discard syms in
/usr/local/php5/lib/php/extensions/no-debug-non-zts-20040718/pdo.so due
to munmap()
==19025== Invalid read of size 4
==19025== at 0x8184ABC: _zval_dtor (zend_variables.c:55)
==19025== by 0x817C2E0: _zval_ptr_dtor (zend_execute_API.c:393)
==19025== by 0x818BE8B: zend_hash_apply_deleter (zend_hash.c:574)
==19025== by 0x818BF19: zend_hash_graceful_reverse_destroy
(zend_hash.c:640)
==19025== Address 0x4239AA24 is not stack'd, malloc'd or free'd
zsh: 19025 segmentation fault valgrind /usr/local/php5/bin/php ./sqlite.php
This looks to be our old friend
Mr.-access-pointers-and-functions-after-we-dlclosed-the-module. Sure
enough, removing the dlclose() call fixes the segfault.
This smells like a shutdown order problem; the script creates two
objects, and if I unset them before the script end there is no segfault.
--Wez.