urlparse do not support reverse_lazy as url arg

[Claude Paroz]

With Python 2, we could pass a reverse_lazy result (__proxy__ instance) to urlparse and the string conversion was done without problem.

With Python 3, this is failing, because urlparse is feeding url argument to a _coerce_args function which basically run:
if not isinstance(url, str): str.decode(encoding, errors) .
And this is failing with an AttributeError: '__proxy__' object has no attribute 'decode'.

I think it is valuable not to have to explicitely force_text(url) each time we want to pass it to urlparse.

[Claude Paroz]

Here is some experimental code that is specifically adding the decode attribute to __proxy__ instances resulting from reverse_lazy.
(Warning: this might be a horrible hack, don't yell at me!)

from functools import wraps

def reverse_lazy(*args, **kwargs):
    @wraps(reverse)
    def wrapper(*args, **kw):
        proxy = lazy(reverse, str)(*args, **kw)
        # Needed in Python 3 to pass the _coerce_args function in urlparse
        proxy.decode = lambda x, y: str(x)
        return proxy
    return wrapper(*args, **kwargs)

[Claude Paroz]

Note also that the failure on Python 3 can be reproduced with the test urlpatterns_reverse.ReverseLazyTest.test_user_permission_with_lazy_reverse

[Aymeric Augustin]

Since reverse_lazy is just defined at reverse_lazy = lazy(reverse, str) I strongly believe this should be fixed in lazy, not only in reverse_lazy.

Otherwise any other use of lazy would still be vulnerable to the same problem.

[Claude Paroz]

Mmmh... it's even worse than that, because even if we solve the issue at lazy level, the _coerce_args function will still break with a TypeError("Cannot mix str and non-str arguments") if we pass a non-empty scheme argument. But who wrote this _coerce_args function!!

So now either we find a way to make isinstance(<__proxy__ instance>, str) == True, or we have to find some workaround (custom urlparse in utils.http, document the limitation, etc.)

[anonymous]

An easy solution was : ​https://github.com/pelletier/django/commit/22987ddf66ed2af39b8df920e3e8501af8782819
But force_text is also good.

[Aymeric Augustin]

I commented on the pull request for that patch: ​https://github.com/django/django/pull/281

[Aymeric Augustin]

Given Claude's decent but unsuccessful efforts toward a "correct" fix, I think we should just go for the simplest solution and just break the lazy layer before calling urlparse.

If the same problem crops up somewhere else, we can always revisit this decision...

[Aymeric Augustin <aymeric.augustin@…>]

In [de3ad8bb2d14ccf878121b96e6e0359dd8b1f9d5]:

[py3] Avoided passing a lazy string to urlparse.

This causes an exception under Python 3.

Fixed #18776.

[Peter Schmidt]

Re: https://code.djangoproject.com/ticket/18776#comment:4

In case anyone is interested (came up through reviewing this from #24097), the original author can be found here: ​https://hg.python.org/cpython/diff/e44410e5928e/Lib/urllib/parse.py

[bugmenot]

This bug has regressed into Django 2.

[Claude Paroz]

If you found a regression, please open a new ticket and show us some code which reproduces the failure.