CliEnv('py37, py36') != CliEnv('py37,py36')

[0cjs]

Issue

Looks like there might be more undocumented behaviour going on in CliEnv. In one of my checkouts I am regularly getting the following failure:

_____________________________________________ test_ini_exhaustive_parallel_values ______________________________________________
tests/config/cli/test_cli_ini.py:127: in test_ini_exhaustive_parallel_values
    assert vars(options.parsed) == {
E   AssertionError: assert {'colored': '...tualenv', ...} == {'colored': '...tualenv', ...}
E     Omitting 30 identical items, use -vv to show
E     Differing items:
E     {'env': CliEnv('py37, py36')} != {'env': CliEnv('py37,py36')}
E     Use -v to get more diff
        core_handlers = {'c': <function show_config at 0x7f141ee70220>, 'config': <function show_config at 0x7f141ee70220>, 'd': <function devenv at 0x7f141ee7b420>, 'de': <function depends at 0x7f141ee7b100>, ...}
        options    = Options(parsed=Parsed(colored='yes', verbose=5, quiet=1, exit_and_dump_after=0, config_file=None, work_dir=None, root_...e': <function legacy at 0x7f141ee700e0>}, log_handler=<ToxHandler <_io.FileIO name=6 mode='rb+' closefd=True> (DEBUG)>)
----------------------------------------------------- Captured stdout call -----------------------------------------------------
ROOT: 1125 D setup logging to DEBUG on pid 294190 [tox/report.py:221]

I can't seem to reproduce it in other checkouts on the same machine using the same version of Python. I do not know why.

Nor do I understand how we ended up with an environment name that apparently starts with a space: py36.

[0cjs]

It turned out to be a change in my tree to CliEnv that triggered this; I got distracted by a completely unrelated problem in that checkout and had forgotten about it. Apparently even with the additional new tests in session/test_env_select.py, that still doesn't give full coverage for some behaviour that's being used by test_ini_exhaustive_parallel_values. There's something going on in the 200+ lines of code used by StrConvert().to(value, of_type=List[str], factory=None) that's not clearly defined. Sigh.

Maybe someone who knows what it is can add a test for it in the proper place.

[0cjs]

Ok, I think that #3208 and #3209 essentially fix this issue in the narrower sense. (What's going into that function that causes the extra space is clear once you look at that fixture; it's just that it was way far away from where it was being used, so it wasn't obvious unless you knew to track it down.)

In the wider sense, there are probably other things like this, but I don't think it's worth keeping an issue open for them; they should just be fixed as we come across them.