Re: removing APC memleak

Gopal V wrote: > Gopal V wrote: >> Gregory Beaver wrote: >> >>> the class default_properties hash table definitions on shutdown in the >>> garbage collection code, and no idea why, backtrace is useless to me, >>> but you smarter folks may be able to do something with it. This is >>> with >>> latest 3.1-dev of APC and PHP_5_3 with pecl/phar CVS, apc and phar as >>> shared modules >> >> Symptom of a much larger problem, IMHO. > > The larger problem aside, I'd like this patch to be tested (on APC_3_0). > > http://t3.dotgnu.info/code/zval_gc.patch > > Is a stupid, idiotic band-aid ... but knowing if that works or not > for your case will make all the difference :) > > Seemed to work for a simple example, but I plan to split out the > apc copy-in & copy-out functions in the HEAD branch - which was > initially thought of to handle stdClass efficiently (in-wards is > a hashtable copy not serialize, outwards is an stdClass ctor). > > But that'll have to wait for a better time than today. *Great news - the patch works on APC_3_0! I also tried it on HEAD (the patch applied just fine with offsets), and that didn't fix the issue, but that was only my dumbness, I didn't read the (on APC_3_0) until I was literally about to click "send." Also good news - I was able to reproduce the exact segfault with HEAD using just plain phpMyAdmin 2.11.3-english - no phar ext and phar disabled in php.ini, so that should make testing for this when working on HEAD easier than doing some phar stuff. Greg *