[Image-SIG] Converting images to palette mode

[Evgeny Salmin]

Hello,

I made some basic checks in this problem (quantize error) and have following
results now:

I've added gc debugging to a script, tried to force collection, but at the
time when quantize error happens gc says that there is no leaks.
So I've decided to check C implementation of quantize() function and wrote
small utility that uses C part of PIL for image quantization. Library is too
old, so there is no leaks in quantize() of course :)
Debugging of script shows that after processing some files malloc returns 0
(usually it in Quant.c : 1265). Sometimes I get python MemoryError exception
after receiving "quantization error", sometimes not. As for memory
allocation process it looks like: 7 MB (at start) -->  processing --> 50 MB
--> seems gc begins working here --> 7 MB -> repeat N times -> quantization
error.
Now I have some assumptions about it:

a) Possible memory leak somewhere in split/eval/paste functions. I can check
them, but I don't think it will give any results.
b) Leak somewhere in a glue code. Cycle references/memmove ?

I have similiar subjects at other mailing lists, maybe they will give some
ideas:
http://mail.python.org/pipermail/image-sig/2007-December/004720.html
http://mail.python.org/pipermail/image-sig/1999-April/000713.html
http://www.daniweb.com/forums/thread133133.html

How will be better to track an allocation/deallocation of Imaging objects?
Anyone have C samples of using C part library of PIL? I'd like to rewrite
python script that causes an error in C to make sure that there is no leaks
at C part, but have to work with png loading/decoding manually, so an
example will be great and help to save time which I spend on examination C
API.

Any ideas?


Regards, Evgeny.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/image-sig/attachments/20090201/fac02e57/attachment.htm>