GitLab

At the end of the run we were losing the coverage and XML output info.
We now stop filtering on a row of hyphens.

The terminalsize module is taken from elsewhere, so we don't want to
test it directly ourselves.

Break up the text in the description of what it does into particular
features. This should make it easier to pick out what the useful parts
of the plugin are.

Not displaying in GitLab, so let's see if a .md helps.

It's high time I tried to explain what this actually does, and shows
how much nicer the output is. So, this is the first shot at some
description.

Pylint complained about a lot of unused variables, all of which should
now be fixed. The print statement is now using a function form, so as
to work with Python 3. Raising exceptions a second time is now performed
through six, rather than faffing about ourselves.

When an error occurs within the setup or teardown code for a class,
there is no corresponding test for the message to be associated with.
As such, it is very useful to be able to report the test failure in the
output so that we can see where the problem occurred.

At present this is only reporting on the failure of the fixture in the
output. In a future change we may ensure that this is recorded in the
final results of the tests as well.

We introduce some new exceptions - FixtureError based exceptions - which
indicate that the problem was with the test fixture, not the test itself
and not the SUT (not necessarily). This allows us to report on problems
which lie with the environment (and possibly the SUT) as distinct from
the test actually being performed.

Where the functions which failed are in known, we can report them
with the failure message. This means that we report 'assertEqual' rather
than 'AssertionError', which is a lot more descriptive. The functions
that provide this ability are extensions to the plugin interface. They
can be provided by other plugins to declare their abilities as well.
This means that it should be possible for the other plugins to be able
to extend the mechanism used to report failures, without having to play
with the innards of the TreeOutput plugin.

Many different file types that we don't care about are ignored.

By default the test name is overridden by the DocString for the test
function. That's ugly and not very helpful when trying to associate the
results and the tests themselves. So we now have checks that the tests
themselves do not get their function names overridden.

The 'longtraceback' module was used to give more information about the
failures. It isn't necessary to have that module present for the
functioning of longtraceback, so we can make its failure non-fatal.

The TreeOutput plugin appears to be functional for the simple operations
that in the tests I'm running. There is a simple manual test available,
but this relies on examining the output of a given test run to be sure
of the results.