Switch from nose to pytest

[scasagrande]

It has come to my attention that the test framework nose (including nose2) is not being as actively developed as py.test. It has a lot more features and nicer output. It also claims to support nose-tests straight out of the box. I believe the only thing we will need to do is replace any nose.eq_ calls with just straight assert statements, as well as update travis & docs.

https://github.com/pytest-dev/pytest/
http://pytest.org/latest/

Trello card: https://trello.com/c/VHzlJk2U/11-switch-test-framework-from-nose-to-py-test

Acceptance Criteria

• Tests run and pass using the pytest framework
• Travis updated
• Dev docs updated
• dev-requirements.txt updated

[bilderbuchi]

OK, so I took a first stab at this (on Python 3.6).

• I fixed the mock import occurences to import mock from instruments.tests, where I have placed an import guard to choose the stdlib mock that's available in Python>=3.3
• I realised that the dependency on pyserial is actually pyserial>=3.3, since you are using list_ports_common. (which is a little problem since Anaconda/conda-forge only has pyserial 2.7)
• With those changes, and nose installed, pytest runs the same 675 tests, and succeeds, as nosetests. So, as long as nose is installed, people can already use pytest
• Nose functionality used is @raises, eq_ and nottest (once).
  • eq_ I think is easy to replace with a nice regular expression search/replace: basically, eq_(a,b) to assert a == b.
  • nottest occurs only once, so probably not much effort to replace.
  • @raises decorator is used 161 times, and is more tedious to replace, since pytest uses this with a context manager.

[scasagrande]

Neat, thanks for taking a look.

I forgot about the pyserial minimum version requirement. There is some more stuff in PR that uses newer features of pyserial, so I'll have to think about what to do there.

[bilderbuchi]

Optimally, we should find out how much effort it is to create a conda-forge recipe for pyserial, then the min version won't be a problem. I think the pymeasure folks have tried this before, correct @cjermain?

[bilderbuchi]

update: Recently, a current version of pyserial has become available on conda-forge, so that can be installed via conda now.

[bilderbuchi]

So, this was a lot of manual work, but I have created a PR (#183).