Move to py.test and take advantage of it

[gsnedders]

We want to move to py.test so that:

• we can select to only run parser tests with one tree builder (and tree walker tests with only one tree walker, etc.)
• we can make use of the plugin API to create a means of dumping test failures and build support for expected failures into the testsuites from html5lib-tests
• it is easier to create nicer error messages
• better support for skipped and expected failures so we can walk towards having a constant number of tests across all platforms, optionally skipped

https://github.com/gsnedders/html5lib-python/tree/expected_failures_pytest is my current WIP branch implementing this.

Status:

Needed to fix failures:

• Make sure tests from generators run in a deterministic order
• Use pytest-expect

The rest:

• Move tokenizer and serialiser tests over to generating test objects from the JSON files.
• Refactor support.py.
• Check the submodule is imported (see Warn on missing testdata submodule #122).

[gsnedders]

I've, uh, been totally over-engineering this. Something along the lines of https://github.com/dmerejkowsky/pytest-runfailed/blob/master/pytest_runfailed.py would work fine.

[gsnedders]

We've now landed the stuff most needed here. Yay!

[willkg]

I think there are three things left to do here:

1. Update html5lib/tests/README to talk about pytest and not notes.
2. Move tokenizer and serialiser tests over to generating test objects from the JSON files.
3. Refactor support.py.

I can do the 1. I need more details on what 2 and 3 entail. Part of me thinks we should split 2 and 3 into separate issues.

@gsnedders Is that ok with you?

[gsnedders]

@willkg lgtm

[willkg]

I split 2 and 3 into separate issues and did a PR to remove that one bit about nose in the tests README. After that PR lands, I'll close this out.

[willkg]

PR #366 landed and the other two things were spun off as separate issues, so everything outstanding is covered elsewhere now. Closing this out.