We are adopting a very "eager" conversion approach with unicode string: calling smart_unicode() as early as possible. This leads to problem with objects coming from ugettext_lazy() and friends. In the current smart_unicode() implementation, they are converted immediately to unicode strings, which is quite possibly (even likely, in some cases) to be in the wrong locale. If we uncomment the commented-out block at the top of the function, things are a little better with respect to translations, but now the code breaks elsewhere because a lot of things assume smart_unicode() returns something we can directly use as a string.
So I need to remember to fix this.
I tried to implement a kind of smart_unicode_lazy in [4904] on trunk, but it didn't work. One problem is that u''.join(l) doesn't work if l contains lazy objects.
I suspect we need to do a full audit of places where smart_unicode could be being used with lazy translation objects and handle them specially if they are not just done at render-time.
