Python xgettext merges gettext and ngettext strings, breaking translation lookup

Question

Running:

> xgettext test.py -o out.pot

On the Python snippet test.py:

def main(num):
    gettext("TEST")
    ngettext(num, "TEST", "TESTS")

Produces a pot file with the following row (translated strings are in the po file):

#: test.py:3 test.py:4
msgid "TEST"
msgid_plural "TESTS"
msgstr[0] "TEST-SINGLE"
msgstr[1] "TEST-PLURAL"

After turning this into a po file and then a mo file. I cannot get translations for gettest("TEST") calls.

> ngettext("TEST", "TESTS", 1)
> TEST-SINGLE
> gettext("TEST")
> TEST

I am using the standard gettext package for Python. I am not sure if these merging behaviour is expected, but it seems to destroy the ability to look up translations for for non pluralized strings. Is there a way to avoid this?

I was thinking of hacking up a fallback for gettext, to try a ngettext call if the first one fails. That seems very hacky though.

Answer 1

The problem appears to stem from the way that the gettext package looks up translations. For gettext and ugettext calls, it simply looks inside the catalog for _catalog['TEST'], and does not search for _catalog[("TEST", 0)].

I do not believe this to be the correct behaviour since xgettext decides to merge the two strings, but I cannot find anything in the documentation to prove one way or the other.

To solve this, I am monkey patching in two replacement methods for gettext and ugettext, that will fallback on a (message, 0) catalog lookup if the simple lookup fails.

Answer 2

The two entries are exactly the same ie singular entry is the same as your plural one. To demonstrate:

msgid "Test"
msgstr "Toets"

msgid "Test"
msgid_plural "Tests"
msgstr[0] "Toets"
msgstr[1] "Toetse"

Then use msgfmt to compile it:

msgfmt test.po 
bob.po:4: duplicate message definition...
bob.po:2: ...this is the location of the first definition
msgfmt: found 1 fatal error

Gettext uses the msgid as the key for your string so it will see these singular and plural strings are duplicates.

Answer 3

I came across the same problem in a recent project. In my opinion, this might be considered as a bug in the python gettext module. And others share this opinion since there is an open issue for this on Python bugtracker, since 2013... I proposed a patch but it may take a while until it is included in a release.

Using ngettext("TEST", "TESTS", 1) instead of ngettext("TEST") is a pretty simple workaroud of course. Not completely satisfactory, but it works...