hopefully this will help with figuring out why our HTTP tests sometimes
fail on travis.

just to be sure, we're not breaking something simple here.

added test coverage

Adjusted pattern for matching media links to properly handle a single '}' in the link text. Fixes #1587.

This fixes a problem with those tests in the most recent version of
phpunit. I thought it was caused by the use of the deprecated var
keyword, but some other weirdness was going on. This way it's more
proper anyway.

This should fix some tests on PHP 7.1. The work is far from done, we
still have a lot of variables without proper accessibility defined.

loadAction() and checkAction() are now public and could be used within
actionOK(). However some weird circular references prevent that. In
addition, actionOK is also used to check for things that aren't Actions
(yet) like 'rss' and 'top'.

also fixes the tests

Lots of FIXMEs and the routing isn't integrated, yet

This is neither complete nor working, yet

Those tests either just check that an exception is NOT thrown or are
there to check dependencies for other tests. Newer phpunits mark all
tests as risky that have no assertion.

actually we could right now. but I don't like to introduce that change in
a feature branch.

We still need 5 to test on old php versions, but travis uses 6 for newer
PHP versions (and 5 may break on those).

Luckily both versions aren't that far apart, yet. This adds a thin
adapter class when running on 6.

See for reference #1837 and sebastianbergmann/phpunit#2503

However those classes are already supported in PHP 5.7 so we can switch
to them without breaking backward-compatibility.

This makes it more flexible to use on the expense of needing one echo
more.

now comments and line breaks between tags are removed

Without the change in issue #1801, or commit
1b26d16d respectively, this adjusted
test would throw an error since it would attempt to use another array as
an array-key.

The fact that the test passes without error means that the bug is fixed.

What happened was that in the place of the new `waitForTick`-call an
actual delay happened and therefore the file modification time and the
deletion entry were two different points in time, resulting in random
test failures. The test enforces this now, the condition to test has
been adjusted.