Ignore fragment url()s

[cyberdelia]

A library I'm using has the following URI value:

behavior: url(#default#VML);

This breaks collectstatic, as it looks for a file of that name. I imagine it would be harmless to ignore any url that is entirely a fragment (ie any URL starting with a #), as it wouldn't involve any remote requests.

Although #17 looks to offer a way to ignore such errors, I think this case should be ignored for everyone as it's obvious that no file can be ever be served with such a name.

[jezdez]

Sounds like a plan.

[cyberdelia]

Also this urls are failing :

url(web-font.eot?#iefix)

And SVG id in url :

url(web-font.svg#webfontIyfZbseF)

And also try to rewrite data-uri :

url(data:font/woff;charset=utf-8;base64,d09GRgABAAAAADJoAA0AAAAAR2QAAQAAAAAAAAAAAAA)

And it's a common pratice used for cross-platform web-fonts support.

It would be nice to not ignore them (except for data-uri), since they are font or others files behind them.

[jezdez]

Okay, sounds like a plan then:

• ignore paths that are only fragments or start with data:
• re-attach the fragments for files that have them

[willhardy]

I tried to write some tests for this, but the presence of the #ed entries broke every other test. Does it need its own test environment or is it ok like this (because when this issue is fixed it won't break the other tests)?

[cyberdelia]

This might no be a good thing, but I see no use case for url() and a directory.

[cyberdelia]

I've been giving a try to this issue myself, it's not perfect but more acceptable than raising an error :|

hub has made this issue an pull request, I hope you don't mind, I will try to reverse that if it's the case.

[cyberdelia]

Great! Thanks.