URLValidator does not accept urls with usernames or passwords in them

[marshall@…]

In [1]: URLValidator()('https://user:pass@from django.core.validators import URLValidator

In [2]: from django.core.validators importURLValidator()('https://user:pass@domain.com')
---------------------------------------------------------------------------
ValidationError                           Traceback (most recent call last)
<ipython-input-2-3adf7dbac94c> in <module>()
----> 1 URLValidator()('https://user:pass@domain.com')

/Users/marshall/.virtualenvs/django/lib/python2.7/site-packages/django/core/validators.pyc in __call__(self, value)
     72                     raise e
     73                 url = urlparse.urlunsplit((scheme, netloc, path, query, fragment))
---> 74                 super(URLValidator, self).__call__(url)
     75             else:
     76                 raise

/Users/marshall/.virtualenvs/django/lib/python2.7/site-packages/django/core/validators.pyc in __call__(self, value)
     42         """
     43         if not self.regex.search(smart_unicode(value)):
---> 44             raise ValidationError(self.message, code=self.code)
     45 
     46 class URLValidator(RegexValidator):

ValidationError: [u'Enter a valid value.']

[marshall@…]

Sorry, readline munged the display, what you should see for the first two lines are:

In [1]: from django.core.validators import URLValidator

In [2]: URLValidator()('https://user:pass@domain.com')

[Preston Holmes]

​http://url.spec.whatwg.org/#concept-url-userinfo

[matiasb]

Proposed update: ​https://github.com/django/django/pull/998
Tests passing with sqlite.

[Florian Apolloner]

See my comments on the pull request (​https://github.com/django/django/pull/998#issuecomment-18097508)

[Aymeric Augustin]

#21532 was a duplicate.

[Danilo Bargen]

These URLs could be used for a thorough test case: ​http://mathiasbynens.be/demo/url-regex

[Danilo Bargen]

Another attempt at fixing this issue (including an extended test suite): ​https://github.com/django/django/pull/2873

[Tim Graham]

Florian and I (and some other core devs I think) are wary of continuing to expand the regex for every use case out there. The validator allows specifying a custom regex and I think we should consider discontinuing "enhancements" to the core regex and let users bring their own as needed. If we go this route, we should beef up the documentation about what is and isn't supported by default.

[Danilo Bargen]

If that is the plan, the documentation really does need an update. The validator claims to validate URLs. Most people who use the URL Validator assume it will validate URLs. But in the current state it is broken and does not. It would be fine if it would consider some invalid URL edge cases valid (meaning it's not too strict), but if it returns a ValidationError for perfectly valid URLs, that's a clear bug to me.

Even if you look online for other ways to validate URLs in Python, you get pointed to the Django implementation: ​http://stackoverflow.com/q/7160737/284318

I think this is a bug that should be addressed somehow. Either by a fix (possibly with a different implementation than the approach I took) or by a deprecation of the validator.

[Claude Paroz]

I'm on Danilo's side on this issue, I don't see why we wouldn't improve the regex if we have a good patch with solid tests.

[Collin Anderson]

Are there possibly any security issues with including a username in the url?

[Danilo Bargen]

@collinanderson I don't think so. The job of the URLValidator is to test whether an URL is valid according to some RFCs, not to decide whether a specific URL is a good idea in a specific case :)

It could have been a security issue if verify_exists were still available (e.g. because the credentials would show up in network traffic), but as this functionality has been removed in 1.5 that's not a concern anymore.

[Tim Graham]

The proposed change appears vulnerable to the "catastrophic backtracking situation" described in 9f8287a3f145fe5cbe71ef5573ea8898404727ad as the test added there now hangs with the proposed changes. This is one reason I'm reluctant to add regex complexity.

[Danilo Bargen]

I think I fixed it: ​https://github.com/django/django/pull/2873#issuecomment-52542276

[Tim Graham]

Some tests on Python 3 are failing.

[Danilo Bargen]

Pull request has been updated. All tests now pass on all supported Python versions.

​https://github.com/django/django/pull/2873

[Loic Bistuer]

Left some comments on the PR.

[Danilo Bargen]

I left some comments as well.

Any opinions on the proposed RegEx simplifications?

[Tim Graham <timograham@…>]

In 2e65d56156b622e2393dee1af66e9c799a51924f:

Fixed #20003 -- Improved and extended URLValidator

This adds support for authentication data (user:password) in URLs,
IPv6 addresses, and unicode domains.

The test suite has been improved by adding test URLs from
​http://mathiasbynens.be/demo/url-regex (with a few adjustments,
like allowing local and reserved IPs).

The previous URL validation regex failed this test suite on 13
occasions, the validator was updated based on
​https://gist.github.com/dperini/729294.