If you pass gettext.translation a class_ but the mo file you are trying to open already exists in the _translations cache the instance that is returned is an instance of whatever was in the cache and not the class that you requested.

I think that if a class is requested it should return that class. If others agree then I'd be happy to work on a patch.

Thanks for your report. Unfortunately, 2.6 and 3.1 are stable releases, they only get security and documentation fixes. 2.7 is nearly in the same state, since it’s at the release candidate stage. If your bug still applies to 3.2 (branch named “py3k”), please propose a patch. I have added the maintainer of the gettext module and another developer expert in i18n to the nosy list (they are listed in Misc/maintainers.rst). Thanks again!

Any comments on the patch for 3.2?

Relying on str(cls) breaks if two classes have the same name, but I guess this is not a concern outside of the interactive interpreter. Patch looks good to me.

The patch looks fine.

Wouldn't constructing the key as a tuple of (class_, mofile) be much cleaner than making up an artificial key?

This would avoid the issue with different classes having the same name, but otherwise would not make that much of a difference.

Here's a second patch with 

key = (class_, os.path.abspath(mofile)))

Looks good. Another +1 from a core dev and I commit this.

+1

> Unfortunately, 2.6 and 3.1 are stable releases, they only get security > and documentation fixes. 2.7 is nearly in the same state

Well, that's false. 2.7 and 3.1 both receive bug fixes (and this looks like a bug to me :-)).

Which branches should i make patches for then? release27-maint and release31-maint?

Antoine: Thanks for the correction.  I was mistaken about versions some months ago and wrote such misleading comments in a handful of bugs.  It has been corrected in some of them but not all.

v_peter: You don’t have to do anything, it’s the job of the committer to do the backport.  We use a tool (svnmerge) for that, so additional patches wouldn’t help.  If a backport is difficult, for example because of bytes/characters issues in 2.x, the committer may ask for help or feedback.

Sorry for the delay, I’ve been rather taken by the start of year and other bugs.  Be sure your fix will be in the next releases for all three branches.

(I’m not sure my last line is unambiguous.  I meant : Don’t worry, this will get in shortly.)

Committed in r85223, r85224, r85225 (resp. 3.2, 3.1, 2.7).  Thanks again!