homepage

The gettext.py parser used by django (lib/python2.7/gettext.py),
GNUTranslations._parse(), around line 313 does not use clean values for k,v on each
iteration ("for item in tmsg.splitlines():")
To reproduce the problem (see traceback, below), try parsing a .PO file containing two headers like this, with a comment header immediately following a plurals header. This example was created by calling msgcat to combine several .po files into a single .po file. Msgcat inserted the comment line.
"Plural-Forms: nplurals=2; plural=(n != 1);\n"
"#-#-#-#-# messages.po (EdX Studio) #-#-#-#-#\n"
Parsing the first header binds the inner loop variables:
 k= plural-forms v= ['nplurals=2', ' plural=(n != 1)', '']
Parsing the second header leaves k,v untouched, which then causes an improper
attempt to parse (since it's a comment, no further parsing of k,v should occur)
 v = v.split(';')
Bug workaround: I use polib to read and immediately save the file. This reorders the metadata to avoid presenting the parser with something that will break it.
Recommended bug fix: on each iteration over tmsg.splitlines, reset the values of k,v = (None, None)
--------------------
Traceback:
File "/Users/sstrassmann/src/mitx_all/python/lib/python2.7/site-packages/django/core/handlers/base.py" in get_response
89. response = middleware_method(request)
File "/Users/sstrassmann/src/mitx_all/python/lib/python2.7/site-packages/django/middleware/locale.py" in process_request
24. translation.activate(language)
File "/Users/sstrassmann/src/mitx_all/python/lib/python2.7/site-packages/django/utils/translation/__init__.py" in activate
105. return _trans.activate(language)
File "/Users/sstrassmann/src/mitx_all/python/lib/python2.7/site-packages/django/utils/translation/trans_real.py" in activate
201. _active.value = translation(language)
File "/Users/sstrassmann/src/mitx_all/python/lib/python2.7/site-packages/django/utils/translation/trans_real.py" in translation
191. current_translation = _fetch(language, fallback=default_translation)
File "/Users/sstrassmann/src/mitx_all/python/lib/python2.7/site-packages/django/utils/translation/trans_real.py" in _fetch
180. res = _merge(localepath)
File "/Users/sstrassmann/src/mitx_all/python/lib/python2.7/site-packages/django/utils/translation/trans_real.py" in _merge
156. t = _translation(path)
File "/Users/sstrassmann/src/mitx_all/python/lib/python2.7/site-packages/django/utils/translation/trans_real.py" in _translation
138. t = gettext_module.translation('django', path, [loc], DjangoTranslation)
File "/System/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/gettext.py" in translation
480. t = _translations.setdefault(key, class_(fp))
File "/Users/sstrassmann/src/mitx_all/python/lib/python2.7/site-packages/django/utils/translation/trans_real.py" in __init__
76. gettext_module.GNUTranslations.__init__(self, *args, **kw)
File "/System/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/gettext.py" in __init__
180. self._parse(fp)
File "/System/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/gettext.py" in _parse
315. v = v.split(';')
Exception Type: AttributeError at /
Exception Value: 'list' object has no attribute 'split'

Does this bear any relationship to issue 1475523? (And yes, I know it is...sad...that that issue hasn't been fixed yet.)

There seem to be several bugs involving this particular inner loop in gettext._parse(), but I don't think they're equivalent.
The present bug (issue17898) is that parsing a plural header breaks the following header when it happens to be a comment.
issue1475523 seems to involve multi-line handling
issue12425 seems to involve breaking when the plural-forms value is empty.
Perhaps a useful design pattern to follow for code which executes this inner loop would be to have some initialization and loop invariants which are asserted true on each iteration. For example, properly initializing k and v on each iteration.

Since the other two listed inner-loop issues apply to 3.x, I would guess that this does also. Steve, can you test with 3.3? And provide a *minimal* test case?

Sorry, I haven't installed python 3.*, I just have default Mac OS python 2.7.
Here's a minimal test case. Tar expands to
 file structure: 
 ./test.py 
 ./en/LC_MESSAGES/messages.po 
$ ./test.py 
Traceback (most recent call last):
 File "./test.py", line 28, in <module>
 test()
 File "./test.py", line 23, in test
 gettext.install('messages', localedir=LOCALEDIR)
 File "/System/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/gettext.py", line 494, in install
 t = translation(domain, localedir, fallback=True, codeset=codeset)
 File "/System/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/gettext.py", line 479, in translat\
ion
 t = _translations.setdefault(key, class_(fp))
 File "/System/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/gettext.py", line 180, in __init__
 self._parse(fp)
 File "/System/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/gettext.py", line 314, in _parse
 v = v.split(';')
AttributeError: 'list' object has no attribute 'split'

Proposed patch against 3.5.

Adding a link to a bitbucket repo.

I would apply this change to 3.4 and 3.5. Should I also backport it to 2.7? I think the same bug applies there, though I haven't verified this or tried my patch.

LGTM. I think it should be backported to 2.7.

New changeset c3d269c01671 by Andrew Kuchling in branch '2.7':
#17898: reset k and v so that the loop doesn't use an old value
https://hg.python.org/cpython/rev/c3d269c01671 

New changeset 54df02192bfc by Andrew Kuchling in branch '3.4':
#17898: reset k and v so that the loop doesn't use an old value
https://hg.python.org/cpython/rev/54df02192bfc