Add support for relative path redirects to the test Client

[master]

Consider this definition, in a reusable app:

url(r'^inbox/$', InboxView.as_view(), name='inbox'),
url(r'^$', RedirectView.as_view(url='inbox/')),

And

urlpatterns = [
    url(r'^messages/', include((app_name_patterns, 'app_name'), namespace='app_name')), 
]

This was working on 1.8, thanks to http.fix_location_header, which converts the url to something like http://testserver/messages/inbox/.

1.9 introduced the "HTTP redirects no longer forced to absolute URIs" change and the fix has disappeared, the reason being "... allows relative URIs in Location, recognizing the actual practice of user agents, almost all of which support them.".
Unfortunately, this is not fully the case of test.Client. It doesn't support relative-path reference - not beginning with a slash character, but only absolute-path reference - beginning with a single slash character (​ref). A GET to inbox/ leads to 404.

I'm using a workaround with ... url=reverse_lazy('app_name:inbox') ..., referred by a TestCase.urls attribute, to produce /messages/inbox/, but I'm not happy with this hardcoded namespace.

[Tim Graham]

I think we should try to fix it in 1.9 given the regression nature of the report. A test is attached.

[Tim Graham]

Does ​this patch fix your issue? If not, could you provide a sample of your test code so we understand exactly what you're doing? Thanks.

[master]

Replying to timgraham

The patch does fix my issue but a simple concatenation doesn't cover all cases.
Suppose your test scenario as: url(r'^accounts/optional_extra$', RedirectView.as_view(url='login/')),
with: response = self.client.get('/accounts/optional_extra')

I think a more appropriate code is:

            # Prepend the request path to handle relative-path redirects.
            if not path.startswith('/'):
                url = urljoin(response.request['PATH_INFO'], url)
                path = urljoin(response.request['PATH_INFO'], path)

[Tim Graham]

Thanks for the feedback. I updated the pull request for your suggestion and added some release notes for 1.9.6.

[master]

Replying to timgraham

Typo in 1.9.6.txt

[Tim Graham <timograham@…>]

In d2569f89:

Fixed #26428 -- Added support for relative path redirects in assertRedirects().

Thanks Trac alias master for the report and review.

[Tim Graham <timograham@…>]

In 192b0659:

[1.9.x] Fixed #26428 -- Added support for relative path redirects in assertRedirects().

Thanks Trac alias master for the report and review.

Backport of d2569f89f28883d07ede0e44a0a69ae678d3b35f from master

[Ian Tyrrell]

Also fixes Client.get()

[Ian Tyrrell]

Though the fix solves the original problem i.e. assertRedirects(), the summary of the ticket suggests broader support.

I've attached a patch based on the original fix, that also fixes Client.get() / Client.post() etc. when follow=True.

[Tim Graham <timograham@…>]

In 2f698cd:

Refs #26428 -- Added support for relative path redirects to the test client.

Thanks iktyrrell for the patch.

[Tim Graham <timograham@…>]

In d2338ba:

[1.9.x] Refs #26428 -- Added support for relative path redirects to the test client.

Thanks iktyrrell for the patch.

Backport of 2f698cd9916cb3d64932248c7114049e02b74fb3 from master

[Jacob]

This issue was not fully addressed. There are still a couple problems:
1) It assumes that a relative redirect does not start with '/'. But what if you redirect relative to host with a '/', as in redirecting to '/subpath' instead of '​https://example.com/subpath'?
2) It does not set the secure kwarg on the client.get call based on the original host scheme, because urlsplit is called only on the relative path. It seems that urlsplit should be called after building an absolute url.

Should this ticket be re-opened, or a separate ticket created?

[Tim Graham]

New ticket, please, as the fix for this one has been released for years.