reverse_lazy representation

[Stanislas Guerra]

Hi devs,

This is related to #18776 and #24097 I guess.

Since Django-1.8:

my_url = "%s?%s" % (reverse_lazy("foo:bar"), params)

Gives something like:

"<django.utils.functional.__proxy__ object at 0x103563850>?param1=val1&param2=val2"

Prior to 1.8, you'd get a "valid" result.

A workaround is to force the evaluation with an explicit unicode(my_lazy_expression) call.

Is it a bug?
If not, is it worth a note in the release notes?

[Claude Paroz]

It would be nice to git-bisect the code to find the culprit commit.

[Tomáš Ehrlich]

Related PR: ​https://github.com/django/django/pull/3239
Commit: ​https://github.com/django/django/commit/c1ef234e310fb984ced37eac8c8b90f5b6155e5f

The problem is when running Python 2.x and it really works in Django 1.7.x.

proxy object for reverse_lazy = lazy(reverse, six.text_type) has defined either __str__ method (in Python 3) or __unicode__ method (in Python 2).

However, when you try to put reverse_lazy into string as described above, the __str__ method is always called (no matter on Python version).

The related part from django.utils.functional is:

            cls._delegate_text = six.text_type in resultclasses
            if cls._delegate_text:
                if six.PY3:
                    cls.__str__ = cls.__text_cast
                else:
                    cls.__unicode__ = cls.__text_cast

[Aymeric Augustin]

I don't think it's a good idea to have __str__ return a unicode object instead of a str. This will lead to confusion.

[Tomáš Ehrlich]

Well, handling of unicode in Python 2 is a little bit confusing to be honest.

The problem is: %s requires __str__ method defined on object and there's no placeholder which would call __unicode__. In Python 3, the %s placeholder requires also __str__, not __bytes__ as one would expect. So, we need to handle, that in Python 2, objects are converted to non-unicode strings (str), while in Python 3 to unicode strings (again str).

Right now, the proxy objects does the *right* thing -- converts objects to unicode strings in both versions. This solutions is logical and seems to be correct, but unfortunately it doesn't work.

We could add obj.__str__ = lambda s: str(s.__unicode__()), but this is exactly what I wanted to avoid because converting unicode to str "just-like-that" can cause lots of exceptions...

[Claude Paroz]

Replying to tricoder42:
...

We could add obj.__str__ = lambda s: str(s.__unicode__()), but this is exactly what I wanted to avoid because converting unicode to str "just-like-that" can cause lots of exceptions...

Maybe something like obj.__str__ = lambda s: s.__unicode__().encode('utf-8')?

[Tomáš Ehrlich]

This looks good

[Claude Paroz <claude@…>]

In cd0ceaa102c1e12b016432db4d180bee7906131f:

Fixed #24252 -- Forced lazy str to utf-8 on Python 2

Thanks Stanislas Guerra for the report and Tomas Ehrlich for
the review.

[Claude Paroz <claude@…>]

In ea3e40c278eb27fb3c8362d9e8cd67f957bf4f57:

[1.8.x] Fixed #24252 -- Forced lazy str to utf-8 on Python 2

Thanks Stanislas Guerra for the report and Tomas Ehrlich for
the review.
Backport of cd0ceaa102 from master.