[unicode] truncate_html_words doesn't work for non-latin characters

[Ivan Sagalaev <Maniac@…>]

django.utils.text.truncate_html_words (and for this matter, 'truncatwords_html' filter) uses regular expression to find words. The pattern is "[A-Za-z0-9]" that doesn't match non-latin words. It should be replaced with "\w" and compiled with re.UNICODE flag.

Patch follows. There is a small backwards incompatible change of behavior since "\w" matches underscores while "[A-Za-z0-9]" doesn't and thus it now considers "_" as words. I believe this is not very bad since isolated underscores are rare in human-oriented texts.

P.S. This whole thing makes sense only for unicode branch since regular expressions in unicode mode don't work correctly for byte strings anyway.

[Ivan Sagalaev <Maniac@…>]

Patch

[Chris Beaven]

Comment from my original patch regarding unicode: ​http://code.djangoproject.com/ticket/2027#comment:6

Do we need to compile with re.UNICODE?

If you were worried about backwards compatibility, then you could use [^\W_], but I don't really think that's necessary ;P

[Ivan Sagalaev <Maniac@…>]

Do we need to compile with re.UNICODE?

Yes. Because otherwise \w means only [A-Za-z0-9_]. re.UNICODE switches re to use unicode db to get the notion of "letter".

[Chris Beaven]

Oh, I actually missed the fact you were already compiling with re.U. :)

[Malcolm Tredinnick]

(In [5533]) unicode: Fixed #4662 -- Fixed a remaining ASCII assumption in
truncatewords_html(). Thanks, Ivan Sagalaev.