Providing custom cache directory through ini option

[RockBomber]

Providing custom cache directory name through ini option (issue #2543)

[coveralls]

Coverage increased (+0.005%) to 92.142% when pulling 931823e on RockBomber:features_cache_dir into b62aef3 on pytest-dev:features.

[nicoddemus]

Awesome, thanks!

[nicoddemus]

Thanks for following through with absolute paths. 👍

I would like to add support for environment variables as well, but I will do it myself some time later.

[RockBomber]

Would you like to add environment variable like PYTEST_CACHEDIR?
Which behaviour must be implemented? Environment variable must override default ini option, but custom ini option must override environment variable?

[nicoddemus]

Would you like to add environment variable like PYTEST_CACHEDIR?

I hadn't thought about this specifically, but this might be a good idea. 😁

Which behaviour must be implemented? Environment variable must override default ini option, but custom ini option must override environment variable?

I meant to be able to expand variables in the config option (#1089):

[pytest]
cache_dir=$XDG_CONFIG_DIR

This would also mean that we can announce that we plan to change the default from .cache into $XDG_CONFIG_DIR (if defined) in the future.

Are you interested to tackle this as well?

[RockBomber]

I can implement it like:

    parser.addini(
        "cache_dir", default=os.environ.get('PYTEST_CACHEDIR', '.cache'),
        help="cache directory path.")

In my opinion, it's look better than cache_dir=$XDG_CONFIG_DIR, because you don't need to create ini file or add advanced args.

[nicoddemus]

I see, but the main point that I would like to have is environment variable expansion support in the cache_dir variable.

Note that we can have both: PYTEST_CACHEDIR support and a cache_dir config var which knows how to expand environment variables.

Not sure which would take precedence... my initial gut feeling would be that PYTEST_CACHEDIR takes precedence over pytest.ini. Looking at how PYTEST_ADDOPTS is handled:

# args is sys.argv
args[:] = shlex.split(os.environ.get('PYTEST_ADDOPTS', '')) + args
args[:] = self.getini("addopts") + args

PYTEST_ADDOPTS is added after addopts (so it takes precedence), so I think my gut feeling makes sense.

[RockBomber]

In this case it may implemented like:

    @staticmethod
    def cache_dir_from_config(config):
        cache_dir = config.getini("cache_dir")
        if cache_dir.startswith('$'):
            env_cache_dir = os.environ.get(cache_dir)
            if not env_cache_dir:
                raise ValueError("environment variable %s is not setted" % cache_dir)
            cache_dir = env_cache_dir
        if _isabs(_expanduser(cache_dir)):
            return py.path.local(cache_dir, expanduser=True)
        else:
            return config.rootdir.join(cache_dir)

[nicoddemus]

We can let let os.path.expandvars take care of that:

@staticmethod
def cache_dir_from_config(config):
    cache_dir = config.getini("cache_dir")
    cache_dir = os.path.expanduser(cache_dir)
    cache_dir = os.path.expandvars(cache_dir)
    if os.path.isabs(cache_dir):
        return py.path.local(cache_dir)
    else:
        return config.rootdir.join(cache_dir)

[RockBomber]

Looks more clear :)
I can add some tests and update PR

[nicoddemus]

Looks more clear :)
I can add some tests and update PR

That would be great, thanks! 😁

[nicoddemus]

(Just noticed there is a "First time contributor" small badge on the PR header, didn't know about this GH feature. Nice! 👍)

[nicoddemus]

It is better to use a non-sibling of our tmpdir, please use the tmpdir_factory fixture to create a new temporary directory instead:

def test_custom_abs_cache_dir(self, testdir, tmpdir_factory):
    abs_cache_dir = str(tmpdir_factory.mktemp('custom_cache_dir'))

Better to use a non-sibling directory to avoid this feature breaking in the future by accident.

[coveralls]

Coverage decreased (-0.2%) to 91.893% when pulling 35469a7 on RockBomber:features_cache_dir into b62aef3 on pytest-dev:features.

[nicoddemus]

Oh just remembered that this also needs a new changelog file. 👍

[nicoddemus]

LGTM, just needs a changelog file and improve the test for absolute path a bit. 👍

[coveralls]

Coverage increased (+0.008%) to 91.893% when pulling 649a384 on RockBomber:features_cache_dir into ef62b86 on pytest-dev:features.

[RockBomber]

Something goes wrong. I will do new PR.
I was confused that in CONTRIBUTING.rst was wrote about adding a message to CHANGELOG.rst
Is it wrong? Must I add change log into changelog directory?