GC Problems...

Thu Sep 9 23:51:00 GMT 2004

Hi Simon,
>> Let me know tomorrow how you fared and I'll submit a format patch.
>>>I discovered a problem with my test last night that doesn't make the results 
>as good, although I think there better than before (certainly no worse).
>>...
>>Re-enabling the use of GlobalAlloc results in a slow leak (from my simple 
>test) that means the heap increases quite slowly over time (taking 4k jumps I 
>believe).
>>Damn... I though you'd got it...

Keep in mind that the PR that my patch is intended to fix
(http://gcc.gnu.org/PR14751) relates to the non-release of several
Win32 synchronization objects needed for thread-related bookkeeping.
I mentioned before that any memory-related issues this patch would
fix would be incidental. If the problems are GC-related, my patch won't
fix them. This being said, I think the patch is necessary because based
on my tests, it does end up freeing the synchronization objects that
weren't previously being freed.
If you go the the above PR, you should see a procedure Rutger outlines
for determining the number of handles a process is currently using
via the Win32 task manager. I'd be curious to know if with my patch,
your handle usage is increasing along with memory usage.
-- Mohan
http://www.thisiscool.com/
http://www.animalsong.org/