Repository navigation
zend_mm_heap corrupted when serializing/unserializing Enum #181
Description
This one works fine:
enum Random: int
{
case One = 1;
case Two = 2;
case Three = 3;
case Four = 4;
case Five = 5;
}
$value = (object)[
'one' => Random::One
];
$serialized = msgpack_serialize($value);
$unserialized = msgpack_unserialize($serialized);This one seems to fail with zend_mm_heap corrupted after running
enum Random: int
{
case One = 1;
case Two = 2;
case Three = 3;
case Four = 4;
case Five = 5;
}
$value = (object)[
'one' => Random::One
];
$serialized = msgpack_serialize($value);
$unserialized = msgpack_unserialize($serialized);
$serialized2 = msgpack_serialize($value);
$unserialized = msgpack_unserialize($serialized);Using the object methods also fails:
$value = (object)[
'one' => Random::One
];
$packer = new \MessagePack(true);
$unpacker = new \MessagePackUnpacker(true);
$serialized = $packer->pack($value);
$unpacker->feed($serialized);
$unpacker->execute();
$unserialized = $unpacker->data();
$unpacker->reset();
$serialized2 = $packer->pack($value);
$unpacker->feed($serialized);
$unpacker->execute();
$unserialized2 = $unpacker->data();
$unpacker->reset();Errors in both
PHP 8.4.14 (cli) (built: Oct 24 2025 19:07:34) (NTS)
Copyright (c) The PHP Group
Built by https://github.com/docker-library/php
Zend Engine v4.4.14, Copyright (c) Zend Technologies
with Zend OPcache v8.4.14, Copyright (c), by Zend Technologies
with Xdebug v3.4.7, Copyright (c) 2002-2025, by Derick Rethans
and
PHP 8.4.16 (cli) (built: Dec 16 2025 16:03:34) (NTS)
Copyright (c) The PHP Group
Built by Homebrew
Zend Engine v4.4.16, Copyright (c) Zend Technologies
with Xdebug v3.5.0, Copyright (c) 2002-2025, by Derick Rethans
with Zend OPcache v8.4.16, Copyright (c), by Zend Technologies
Activity
earthiverse commentedon Jan 8, 2026
earthiverse@ca6d866 made a test to reproduce it, seems more likely to fail when you include the Enum from another file
php-8.4-debug-nozts
php-8.4-nodebug-nozts
Edit: Tested a fix generated by AI earthiverse@671217f
It fixed the test case, but I have no idea what it's doing, and I don't want to burden anyone with slop, so I'm just editing my previous comment here.
cracksalad commentedon Apr 11, 2026
@earthiverse could you please try and build #184. It could help with your issue as well, but I am not sure about that.
earthiverse commentedon Apr 13, 2026
Seems like it could be different issues...
Rebased the test I made on your master branch, but my test failed.
https://github.com/earthiverse/msgpack-php/actions/runs/24319495765
Rebased your test on the fix that fixed my test, but your test failed.
https://github.com/earthiverse/msgpack-php/actions/runs/24319566957
cracksalad commentedon Apr 13, 2026
Would have been too easy, I guess. Thanks for testing though.
The your-tests-my-fix variant's test results look all over the place. I mean 8.1 and 8.4 are mostly satisfied while 8.2 and 8.3 are not. At least the php-8.4-debug-nozts and php-8.4-nodebug-zts tests seem to be fixed compared to the your-tests-no-fix variant. Sadly I can not see the test result details this way. So I will integrate your tests into my version and maybe I am able to extend my fix.
cracksalad commentedon Apr 13, 2026
There you go. After hours of looking at the location valgrind points to during serialize, I came to the conclusion that the root cause is in unserialize as you already suggested. At least, I found a cleaner and simpler solution, which does exactly the same as your
GC_ADDREF(...)line, but hidden 😁I also integrated your test and CI ran successfully.