Memory leak after calling exit() when Zend MM is turned off?

When the Zend memory manager is turned on, everything is fine. Is this
behavior intended?

We do have some intentional leaks where we rely on the pool allocator to
wipe things at the end of a request. So I am less concerned about things
you see at request termination than I would be if you found one mid-request.

-Rasmus

Hi internals!

I am currently developing a Zend extension. Because the Zend MM leak reports are not really useful sometimes, I switched the memory manager off with the environment variable USE_ZEND_ALLOC=0, so that I can use valgrind.

Right now I have found some kind of memory leak. You can use every script which has an exit()-call to reproduce this leak. For example:

<?php exit("Memory leak example.\n"); ?>

If you run this PHP script with valgrind, it will report some ugly memory leak:

valgrind version: valgrind-3.6.0.SVN-Debian
PHP Version: 5.4.10 (debug)
Configure options: --disable-all --disable-cgi --enable-debug

Command: export USE_ZEND_ALLOC=0 && valgrind --leak-check=full php-5.4.10-debug test.php

Report:
==5693== HEAP SUMMARY:
==5693== in use at exit: 420 bytes in 4 blocks
==5693== total heap usage: 8,355 allocs, 8,351 frees, 2,503,705 bytes allocated
==5693==
==5693== 420 (240 direct, 180 indirect) bytes in 1 blocks are definitely lost in loss record 4 of
==5693== at 0x4C244E8: malloc (vg_replace_malloc.c:236)
==5693== by 0x5E08C4: _emalloc (zend_alloc.c:2423)
==5693== by 0x5C11C3: compile_file (zend_language_scanner.l:551)
==5693== by 0x617311: zend_execute_scripts (zend.c:1301)
==5693== by 0x58A22F: php_execute_script (main.c:2482)
==5693== by 0x768967: do_cli (php_cli.c:988)
==5693== by 0x76987C: main (php_cli.c:1364)
==5693==
==5693== LEAK SUMMARY:
==5693== definitely lost: 240 bytes in 1 blocks
==5693== indirectly lost: 180 bytes in 3 blocks
==5693== possibly lost: 0 bytes in 0 blocks
==5693== still reachable: 0 bytes in 0 blocks
==5693== suppressed: 0 bytes in 0 blocks

When the Zend memory manager is turned on, everything is fine. Is this behavior intended?

Yes. The implementation of exit() will not necessarily free all
allocated blocks. It therefore intentionally suppresses errors that
might come from the memory manager regarding leaks - but if you turn
off the memory manager and use valgrind - you'd see them. It's not a
problem unless you actually use PHP without the memory manager in
production, which of course you should not.

Zeev