homepage

Note the following Debug Python session:
>>> s=u"anything"
[8189 refs]
>>> v=s.encode("utf8")
[10967 refs]
>>> v=s.encode("utf8")
[10968 refs]
>>> v=s.encode("utf8")
[10969 refs]
>>> v=s.encode("utf8")
[10970 refs]
Each call to encode is losing a reference. Attaching a
test program that demonstrates this in more detail. 
The output from my test program is:
After 10000 iterations, lost 12850 references
[15227 refs]
and for 100000:
After 100000 iterations, lost 102850 references
[105227 refs]
etc.
As far as I can tell, this appears in all Python 2.x
versions.

Logged In: YES 
user_id=14198
s/decode/encode/ :) Also meant to mention problem not
restricted to UTF8 - changing the encoding in the text file
to anything other than 'ascii' seems to leak in the same way.

Logged In: YES 
user_id=14198
Found it :) Attaching patch.

Logged In: YES 
user_id=14198
Oops - too quick. All calls to _PyCodec_Lookup() leak.

Logged In: YES 
user_id=33168
Patch makes sense to me.
If you add a test, I may be able to catch the problem w/purify
next time I run it (if purify works).

Logged In: NO 
I'm not sure what sort of test you are suggesting I add. I
think the patch is pretty obvious and reasonable, so MAL
should just check it in or assign it back to me <wink>. 
Earlier the better really.

Logged In: YES 
user_id=14198
damn sourceforge - it went to the trouble of asking my email
address when I submitted without being logged in, but it
doesn't seem to have done anything with it - so that was me
just incase you weren't sure :)

Logged In: YES 
user_id=38388
I'll have a look later today.

Logged In: YES 
user_id=33168
Basically the code in the report would be fine.
Purify *should* catch anything which causes the leak.
So:
 s = u'anything'
 assert(s.encode('utf-8') == s.encode('utf-8'))
should work. Perhaps, there is already a test for this?
and purify didn't report leaks.

Logged In: YES 
user_id=14198
A tickle for Marc, assuming his days aren't quite *that*
long <wink>. Just give the OK and I will check it in.

Logged In: YES 
user_id=38388
Perfect. I've marked it as Python 2.2.1 candidate. Please
also mention this in the checkin message.
Thanks. (And sorry for not getting back earlier -- my days
are indeed *very* long ;-)

Logged In: YES 
user_id=14198
Checking in codecs.c;
/cvsroot/python/python/dist/src/Python/codecs.c,v <-- codecs.c
new revision: 2.14; previous revision: 2.13
done