Execute Cram via pytest

[vanguard737]

Description of proposed changes

@huddlej As discussed in PR #543, here's the PR to add just the pytest wrapper for the Cram functional tests, including coverage report.

One noteworthy change: When I submitted this in #543, I got different coverage numbers on my local machine vs. Travis. In this PR, that's no longer the case - I'm now seeing 41% overall coverage both locally and on Travis. Not sure why or how that fixed itself - but, seems to be working now.

This PR does the following:

• Adds a Pytest wrapper, tests/test_cram.py, to execute all Cram CLI tests located under tests/.
• Updates .travis.yml to execute this test independently from existing Augur unit tests, and to generate a coverage report showing code coverage just for the Cram tests.
• Updates pytest.python3.ini to exclude the pytest Cram wrapper when executing unit tests.
• Updates .gitignore to ignore the coverage datafile .coverage_pytest_cram.

Couple of additional notes:

• The .travis.yml and .gitignore changes are worth calling out. pytest-cov writes a coverage-data file to the default location, .coverage; Travis then uploads this to codecov.io. I don't want the Cram coverage data to overwrite the unit-test coverage data. So, to keep these separate, I've updated .travis.yml to set COVERAGE_FILE=.coverage_pytest_cram when I call pytest.

On my Travis and Codecov accounts, this seems to work as intended. Codecov correctly shows ~19% unit-test coverage, not the ~40% coverage you'd see if Codecov were picking up the Cram-test coverage data. But, the Codecov bash uploader is a little vague about how it detects coverage files. So, you might want to keep an eye on this. I've added a note for this in the 'Testing' section below.

• In .travis.yml, I've left in place the call to execute Cram tests in the 'normal' way (cram --shell=/bin/bash tests/functional/*.t tests/builds/*.t). This is redundant to the pytest-Cram call, of course - I kept the original just as a control, to prove that pytest-Cram and normal-Cram are both running and succeeding. Up to you guys, but my instinct would be to keep both calls side-by-side for a while, then delete the normal-cram call when everybody's comfortable with pytest-Cram.

Related issue(s)

Related to #542 #543

Testing

Locally:
COVERAGE_FILE=.coverage_pytest_cram pytest -vv -s tests/test_cram.py --cov-report=term --cov augur

On Travis:
Run a build. Observe that tests/test_cram.py executes tests/builds/zika.t and tests/functional/mask.t, and both tests pass (example). Observe that this writes a coverage report to terminal, currently showing 41% coverage (example).

Check codecov.io to verify that it's still showing coverage for just the augur unit tests. Should be ~19% or so - if it's ~40%, then Codecov may have unexpectedly pulled pytest-Cram's coverage data into the mix.

[codecov]

Codecov Report

Merging #549 into master will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##           master     #549   +/-   ##
=======================================
  Coverage   19.16%   19.16%           
=======================================
  Files          31       31           
  Lines        5072     5072           
  Branches     1289     1289           
=======================================
  Hits          972      972           
  Misses       4077     4077           
  Partials       23       23           

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 704f369...ab30fbe. Read the comment docs.