Revert #656 on master branch or improve skipping of tests

[corranwebster]

Having a look at the awkwardness of avoiding problems by skipping tests with Traits 6.0, I would strongly recommend either we:

• revert Add CI job targeting Traits 6.0 #656 on master
• use the unittest test loader mechanisms to detect Traits 6.0 and skip over the tests in data_view modules (eg. inject something here https://github.com/enthought/pyface/blob/master/pyface/__init__.py#L33)

I'm generally in favour of the first as long as we are good about maintenance releases, as I think it just adds complexity and makes implicit promises that I'm not sure are worth the extra effort.

[kitchoi]

It does not seem very difficult to me to be wrapping imports and test case with conditions actually. It has the benefit of skipping the tests even if you are running the test module individually without using unittest discover.

But I think the second option of skipping the tests using load_tests is fine too. I would avoid using environment variable like the EXCLUDE_TESTS because people running the test suite not via etstool.py won't have the environment variable set.

I'd still advocate we keep compatibility with Traits 6.0.

[corranwebster]

Actually, having just had a look at this in some detail, I would strongly recommend using the EXCLUDE_TESTS environment variable to do this:

• it is an existing mechanism
• it is designed to solve the same problem (avoiding importing things that you don't want to)
• if you run unittest discover across the entire codebase outside of etstool without the environment variable set, then we already get test failures because of the Qt/Wx import issue; this will make no difference
• if you run unittest or unittest discover on specific sub-packages, it will either work or not in a completely predictable way (eg. python -m unittest discover pyface.tests works just fine without any environment variables set).

[kitchoi]

if you run unittest discover across the entire codebase outside of etstool without the environment variable set, then we already get test failures because of the Qt/Wx import issue; this will make no difference

Indeed that is the awkwardness I run into when I try to release and test the release like this:

$ python3.8 -m venv test
$ source test/bin/activate
$ pip install traits pyface[pyside2] traitsui
$ python -munittest discover pyface

I always get a test error due to failure to import wx (likewise for qt if the environment is using wx), and I have to excuse that knowing it is because of the toolkit. Likewise, if we use an environment variable with load_tests for handling traits 6.0, then one will get test errors if they install pyface from pypi along with traits 6.0 (with whatever favourite deployment manager they have). That's not very nice for the majority of users who don't use etstool.py. You want all the tests to pass if they are not skipped.

IOW I actually disagree with the existing mechanism, but for toolkit import, the error is more likely to be taken as "expected". For importing traits, it won't be as obvious.