Refs #2068: Do not reinstantiate staticfiles storage classes

[matthiask]

Description

This change might address an issue where reinstantiating the staticfiles class inside our staticfiles panel would lose some attributes of the storage instance.

Fixes #2068

Checklist:

• I have added the relevant tests for this change.
• I have added an item to the Pending section of docs/changes.rst.

[tim-schilling]

@matthiask would it be possible to wrap the url method on the storage class without having to create a new class?

[matthiask]

@tim-schilling Yes, I think we could directly monkey-patch the url method on the static files storage instance. Do you think that would be preferrable to the Mixin injection? I can see how a straightforward monkeypatch might be easier to understand.

[tim-schilling]

I think either case works well. I haven't manipulated __mro__ often, but if it works I'm good with it.

[matthiask]

I have done too many hacks with __bases__, and monkey patching the method seems to be a better way to go.

If we want a quick bugfix release for #2068 I can finish this, but maybe this would also be a good fit for our Djangonauts? Especially if we can wait a few days or weeks for a fix.

[tim-schilling]

I'm good with waiting.

[dr-rompecabezas]

I assigned #2068 to myself, thinking all this PR needed was tests, but now I see you were considering a different approach (monkey patching). I'm trying to understand the proposal to see if I can help.

[matthiask]

So, monkeypatching would mean something like this:

def patched_url(self, path):
    url = orig_url(path)
    ... our own customizations
    return url

orig_url = storage.staticfiles_storage.url
storage.staticfiles_storage.url = patched_url

It's a less tricky way to achieve basically the same thing my MRO hack did. I'm not immediately aware of any downsides either way. Also, runtime patching is frowned upon, but we have to do something ugly to add our instrumentation.

[tim-schilling]

I suspect the MRO hack may be less intrusive to other devs. If we monkey patch things, we'd want to consider using functools.wraps or something similar to keep the method interface the same.

[dr-rompecabezas]

Superseded by #2097