Is there any way to activate the virtual environment a test is run in for that test execution? I'm dealing with a bunch of horrible things that Flink does that seem to only work if the python interpreter corresponds to one built with a virtual environment. It's running commands from a jvm and bash script like

python -m <some things it expects to be on path>

. When the test is executing,

which python

just points to the system install at

/usr/bin/python

.

It looks like pytest is already run in a venv: https://github.com/pantsbuild/pants/blob/3e62816f6e2bbe180962d855438ad6fcddb669fa/src/python/pants/backend/python/goals/pytest_runner.py#L281-L290

Hmm, when I do

subprocess.check_output(['which', 'python'])

it points back to system

Well, there are a few things getting mixed up here. If you want

PATH

adjustment, that's one thing. Your example claimed

sys.path

was the issue (-m ...)

Here is the specific part that is unhappy. And yeah, I'm trying to play some

PYTHONPATH

games to get things happy currently.

To get

PATH

adjustment, this would need to be `--venv prepend`: https://github.com/pantsbuild/pants/blob/3e62816f6e2bbe180962d855438ad6fcddb669fa/src/python/pants/backend/python/util_rules/pex.py#L888

How do you mean export? By environment variable or something else?

Yes, env var. Read the ~1st line of the script you pointed to.

Yeah, mostly good things and then a few things like this that are so broken out of the box. Unfortunately that env var setting didn't affect things and I'm pretty sure it was high up enough

Ok, and, at the end of the day, don't forget the

--venv prepend

thing. I think that could be a safe, global, change to all Pants VenvPex setups; so might be worth bugging & fixing.

I think I found a suitable workaround for now. Setting

python

envvar didn't work, but setting

PYTHONPATH

to where my source code and deps are did. So basically copying what they do in https://github.com/apache/flink/blob/af681c63821c027bc7a233560f7765b686d0d244/flink-python/pyflink/bin/pyflink-udf-runner.sh#L28

Just checking that you're aware of

--keep-sandboxes=always

which lets you play in the sandbox the test ran in