utils: improve extract_mentioned_users usefulness

Previously, extract_mentioned_users performed a regex search, converted
to a set to remove duplicates, converted that set back into a list, and
sorted the list on username, ignoring case. Every single caller of the
function then promptly took the carefully pruned and sorted result and
put it back into a set, rendering the entire exercise pointless.

In addition, every caller also resolved the usernames to database User
objects.

In this changeset, extract_mentioned_users is changed to return a set
of database User objects directly. A new extract_mentioned_usernames
function, which does only username extraction, is kept for the sole
purpose of providing a more testable interface (no database needed).

Bonus feature: The new extract_mentioned_users will prune non-existent
users, as well as the DEFAULT user. This means it is no longer possible
to @mention (and send notifications to) the DEFAULT user.

auth: prevent misuse of PermFunction in bool context

Evaluating a PermFunction as a boolean, rather than calling it, is
almost certainly an error. If not, "pf is not None" can be used.

notifications: untangle notification access check

This removes a broken permission check when viewing notifications (the
HasRepoPermissionAny object was created, but never actually called with
a repo_name argument as required). It would be non-trivial to actually
implement the check, as notifications don't track their repository
relationship explicitly, and even then, it's unclear why it would
make sense to allow a repository admin to see notifications to
other users.

It was never a vulnerability, due to a subsequent (and much stricter)
ownership check, which remains but has been untangled for readability.
In short, this changeset is a pure refactoring, except that specifying
a non-existent notification ID will now produce error 404, not 403.

setup.py: enable pytest-sugar plugin

pytest-sugar enables a nice visual representation of test progress and
results.
Minimum version is set to 0.7.0 as it fixes a problem where skipped tests
would not be shown in the test summary.
The upper version 1.0.0 is arbitrarily chosen and could be increased as seen
fit.

helpers: don't truncate words when when truncating texts for urlification

This would especially cause confusion when URLs were truncated.

Added tag 0.3.2 for changeset a84d40e9481f
Added signature for changeset a84d40e9481f

auth: further sanitize requests to prevent GET CSRF (CVE-2016-3691)

Routes allows GET requests to override the HTTP method, which breaks
the Kallithea CSRF protection (which only applies to POST requests).

This commit blocks such GET request, preventing CSRF attacks.

auth: properly invoke PermFunctions (CVE-2016-3114)

This fixes a vulnerability that allowed logged-in users to edit or
delete open pull requests associated with any repository to which
they had read access, plus a related vulnerability allowing logged-in
users to delete any comment from any repository, provided they could
determine the comment ID and had read access to just one repository.

changelog: fix range selection from changelog in git repos (Issue #190)

The Mercurial backend can handle short revision ids so the problem does not
occur with hg repositories. Git needs the full hash.

vcs: fix repo size calculation

tip.walk already 'dives into' the dirs - don't do that twice.

tests: finally block also needs to be executed if api_call itself fails

The cleanup code (finally block) should be executed as soon as something what
changes the db fails.

I don't remember which one but one of these tests failed and caused all the
others to fail as well because the database wasn't cleaned up after the first
failure. That made finding the root cause more difficult then it should have
been.

pytest migration: model: convert all tests to TestControllerPytest

The model tests were based on BaseTestCase which does not exist currently
for pytest-style tests. Nevertheless, there seems to be no advantage of
directly subclassing BaseTestCase over subclassing TestControllerPytest.
Thus, keep things simple and use TestControllerPytest.

pytest migration: add temporary assertRaises wrapper

Similar to the other temporary unittest-assert wrappers, add the missing
assertRaises and forward to pytest.raises.

pytest migration: simplify hg+git test class hierarchies

No need to use multiple inheritance when single inheritance can do it just
as well. Let the git/hg test cases just derive from _BaseTestCase which in
turn derives from TestControllerPytest.

Also remove empty setup_class/teardown_class methods.

pytest migration: convert functional tests with setup/teardown methods

pytest also supports setup/teardown methods like unittest (even though
pytest fixtures are more powerful and should be the end goal).
Only difference is the naming and signature of setUp (setup_method) and
tearDown (teardown_method).

test_admin_notifications: fix index test dependency

The index test only worked because another test had prepared some database
state it relied on.

More specifically, the index test creates a notification
as a newly created user, but that user had not been committed to the
database yet. When running the index test standalone, this causes the error:

IntegrityError: (IntegrityError) NOT NULL constraint failed:
notifications.created_by u'INSERT INTO notifications (subject, body,
created_by, created_on, type) VALUES (?, ?, ?, ?, ?)'
(u'test_notification_1', u'notification_1', None, '2016-03-11
21:15:19.066402', u'message')

By committing the database after the user is created, the problem is solved.

i18n: updated translation for Chinese (China)

Currently translated at 39.8% (452 of 1135 strings)

i18n: updated translation for German

Currently translated at 58.6% (666 of 1135 strings)

i18n: updated translation for French

Currently translated at 100.0% (1135 of 1135 strings)

i18n: updated translation for French

Currently translated at 98.5% (1119 of 1135 strings)

tests: fix Git on Windows sometimes failing on ' or ` in file:/// URLs

Some Git versions will fail when cloning a repository like:

git clone file:///path/to/strange-file-'`foobar myclone

Since '` (and some other characters) are filtered anyway (in
lib.utils.repo_name_slug) I think this test doesn't need to test those
characters.

tests: fix test_nodes.py and test_files.py which fail on windows due to mimetype differences

The mimetype of files may differ on different platforms (e.g. the mimetype of
.py files on windows is not text/x-python but text/plain). Therefore, compare
the mimetype to the value the platform uses.

compare: close Dulwich internals for Git repositories to avoid leaking open files

Fix failing test_compare.py on Windows by explicitly closing internal
object_store objects.

api: avoid sending double Content-Length header (Issue #201)

WSGI will take of Content-Length - handling it explicitly was misguided.

docs: improve mod_wsgi documentation (Issue #203)

Based on feedback from Graham Dumpleton.

* Recommend setting global Apache config WSGIRestrictEmbedded On
* Don't set the default value for processes=1 - it has side effects
* Use python-home instead of python-path
* Fix confusing mentioning of root and clarify the use of specifying user and group
* Include WSGIProcessGroup in configuration excerpt

docs: add notes about IIS, Windows Authentication and Mercurial

(The original patch from Konstantin has been heavily copyedited and modified by
Mads Kiilerich but is still [based on] Konstantin's feedback and contribution.)

status: refactor internal dict for mapping status string to description

Slightly better performance ... but more importantly: do it "the right way".

docs: document Kallithea customization options

Create basic documentation for the various customization options in
Kallithea. Move the rcextensions section to this new page and reword.

settings: rescope 'Analytics HTML' option to generic HTML/JS/CSS customization

Update the setting name 'Analytics HTML block' to better cover the fact that
it can be used to make any HTML, JavaScript or CSS tweak.

vcs: fix processing of git commands that return output on stderr

The subprocessio module used for interfacing with Git is using threading. It
might not be fully deterministic.

The subprocessio module also had the "feature" that it stopped reading and
reported failure once it found output on stderr - even if the command completed
(or would complete) with success.

Sometimes (like https://drone.io/bitbucket.org/domruf/kallithea/24 ) tests on
linux could fail on totally normal informational output like:

Couldn't run git command (['git', '-c', 'core.quotepath=false', 'checkout', 'foobranch']).
Original error was: Subprocess exited due to an error:
Switched to branch 'foobranch'

On Windows it would fail even more often.

To fix that, ignore stderr while processing output. There is a risk that it in
some cases can make the process block on stderr and thus never finish ... but
there is no reports of that yet.

i18n: add translation for Spanish

Created new translation.