Need option for not skipping missing interpreters

[wsanchez]

I really like to have skip_missing_interpreters = True in my tox.ini, because I'm often working on machines that don't have every Python interpreter installed.

But it's a really bad idea in a CI scenario, because environments just pass without doing anything.

I'd love to see an option to set that value back to False, but there's only a CLI option for setting it to True… every flag really should have an opposite.

[obestwalter]

Hi @wsanchez, you could make the value dependent on an env var and get what you want this way.

[tox]
skip_missing_interpreters = {env:TOX_SKIP_MISSING_INTERPRETERS:False}

On your devbox have the env var permanently set by adding it to the initialization of the shell or directly on the command line:

$ TOX_SKIP_MISSING_INTERPRETERS=True tox  #posix

$ set TOX_SKIP_MISSING_INTERPRETERS=True & tox  #windows

CI can then be blissfully unaware of that configuration option and developers can adjust it to their needs.

[asottile]

I for one think this is a good idea, patch incoming :)

[obestwalter]

I am -1 on this. Feels like command line argument inflation for something that is easily solvable with what we have.

[obestwalter]

I would rather introduce accepting an argument for --skip-missing-interpreters that behaves exactly like the configuration option. That would be a breaking change, but much cleaner.

[asottile]

Devils advocate: why even have the command line argument if you could do it with environment variables 😆

[asottile]

more seriously: I think we could change --skip-missing-interpreters to take an argument and preserve the existing behaviour if no argument is passed -- if that's a better option :)

import argparse
import distutils.util

no_arg = object()


parser = argparse.ArgumentParser()
parser.add_argument(
    '--foo', default=no_arg, nargs='?', type=distutils.util.strtobool,
)
args = parser.parse_args()

if args.foo is no_arg:
    print('not present: false')
elif args.foo is None:
    print('old behaviour: true')
else:
    print(f'new behaviour {args.foo}')

$ python3.6 test.py
not present: false
$ python3.6 test.py --foo
old behaviour: true
$ python3.6 test.py --foo=True
new behaviour 1
$ python3.6 test.py --foo=False
new behaviour 0
$ python3.6 test.py --foo=garbage
usage: test.py [-h] [--foo [FOO]]
test.py: error: argument --foo: invalid strtobool value: 'garbage'

[obestwalter]

Now we're getting somewhere :)

[The-Compiler]

That will break when there are arguments following:

$ python3.6 t.py --foo test.py
usage: t.py [-h] [--foo [FOO]]
t.py: error: argument --foo: invalid strtobool value: 'test.py'

[obestwalter]

TBH my feeling is anyway to try to simplify tox configuration instead of piling the features on, so I will stay with my -1 on this.

[The-Compiler]

Maybe tox should just have an -o option to override any ini option, like pytest has? With that, you could do tox -o skip_missing_interpreters=false but also e.g. tox -o distdir=foo, and maybe even tox -o '[testenv:py35]basepython=python3.5'.

[obestwalter]

That sounds like a great direction to me.

[asottile]

Ooh that sounds like the best option.

If we wanted to make it generic perhaps something like this?

• -o tox.skip_missing_interpreters=True
• --option testenv:py35.basepython=python3.5

Or with brackets?

• -o [tox]skip_missing_interpreters=True
• -o [testenv:py35]basepython=True

EDIT: that argparse snippet above only fails on positional arguments afterwards -- it still does the right thing on options:

$ python3.6 test.py --foo --bar
usage: test.py [-h] [--foo [FOO]]
test.py: error: unrecognized arguments: --bar

[obestwalter]

I'd say we have good look at how pytest does this and ideally implement the exact same behaviour. I'm off to bed for now though. Wrestling with that horrible release workflow I concocted left me utterly drained.

[The-Compiler]

@asottile The syntax with brackets I proposed was inspired by the interpolation syntax in tox.ini where you'd use {[testenv:py35]basepython} as well, so I think it'd make sense to keep that syntax.

EDIT: that argparse snippet above only fails on positional arguments afterwards -- it still does the right thing on options:

Right. It still means invocations which worked before will give a confusing error afterwards.

I'd say we have good look at how pytest does this and ideally implement the exact same behaviour.

Note that pytest doesn't have to handle different config sections, it only has [pytest].

[obestwalter]

closing in favour of #634 (and because it is possible without this option already as dexcribed in #631 (comment)).