test leak fixes

[beniwohli]

this fixes all the test leaks that are currently appearing

WSGIRequest has no attribute current_page

This can be solved by setting the current_page attribute on the fake request

NonRootCase URL reversal

as @yakky pointed out in #1335, it seems to be necessary to call client.get() or client.post() before doing URL-reversal. While I'm not 100% sure why this is necessary, I guess that django only sets up the whole URL-reversal infrastructure when the client is used the first time.

[travisbot]

This pull request passes (merged f0911d7 into 33a994c).

[ojii]

to me this looks like covering up the real issue. which is that request.current_page apparently leaked in some test to another test. Shouldn't that be fixed instead?

[beniwohli]

I don't think that the change per se is a bad one (it makes the fake request look more like a real one). But I'll admit that it may cover the real problem. Let's try to fix that first, then merge this

[yakky]

I tried runtests-isolated.py from #1336 with beniwohli/django-cms@f0911d7 with success.
Still don't get why, but it looks the only sensible solution ATM.

[yakky]

As for the current_page issue, maybe the cause is exactly the same.
I concentrated on FixturesMenuTests: test_basic_cms_menu() calls self.client.get() (https://github.com/divio/django-cms/blob/develop/cms/tests/menu.py#L86) which calls the middlewares, including CurrentPageMiddleware which inserts the current page in the request.
I bruteforced a fix inserting a self.client.get("/") in get_request() (https://github.com/divio/django-cms/blob/develop/cms/test_utils/testcases.py#L256) this is obviously not a fix, but makes the failing tests pass (other tests which check the number of the queries or cache content obviously fails due to the different environment).

[ojii]

real bug for nonroot tests is that get_pages_root returns '/' instead of '/content/' before the first request.

[yakky]

Yes but actually that happens only if no client.get() has been issued before (in any previous test)
We discovered that while executing that on test at a time.
As reported by me (in #1335) and @piquadrat here is not easy to figure out why does this happens (whether a new_reverse issue, or a too lazy loading of cms urls by django or an incomplete request object)

[ojii]

See https://github.com/ojii/django-cms/tree/get-pages-root-problem for a test that shows get_pages_root seems to be the culprit

Run python runtests.py NonRootCase.test_get_page_root.

[ojii]

fixed in 998bceb

[coveralls]

Changes Unknown when pulling f0911d7 on piquadrat:feature/test-leak-fixes into * on divio:develop*.