Hi, I have a Python test that uses Spark, which depends on having a JVM. Is there a way to set a dependency on a JVM with a specific version? (for instance,

openjdk 11

)

Not sure which version has it, but most probably you can use

adhoc_tool

https://github.com/pantsbuild/pants/blob/a805788031b94477bf855b8e5a11ea7b03f894c1/docs/markdown/Ad-Hoc%20Tools/adhoc-tool.md#L4

Thanks, will give it a go.

I haven’t used it myself. I mostly saw this being recommended here for similar scenarios. This is still in beta.

Yeah if everything gets structured correctly, the output of your tool with be cached. And we're cranking performance of that caching in 2.17.x

👍 Thanks for confirming.

Don't take out word for it though, try it out! We have spark code in our codebase and right now I just turn a blind eye, so I'd love to hear how this works for you

Ack. Will probably take a few days before I have a result though.

This has been silently broken in our repo for months. Take your time 😉

Someone with more JVM knowledge than me should chime in (@witty-crayon-22786, @ancient-vegetable-10556) but I don't think adhoc_tool is pertinent. I'm pretty sure the JVM backends already support this?

Let me take a loooksie

Oh, NM, I missed that this was a python test, my bad

The best thing you could do is write an

experimental_test_shell_command

that executes the

pytest

runner and the relevant JVM dependencies as a

runnable_dependency

, but it’s not a workflow I have an example of off the top of my head, and it’s likely to be exceeeeeeeeeedingly awkward.

I see. Well, I'll think of it for a bit. I imagine maybe making a new

jvm

target that the

python_tests

and/or

python_sources

could depend on and thus to have a JVM available in the relevant contexts (testing, maybe somewhere else?). Any leads on that? (or maybe save me some time by telling me it's not such a great path to go through? 😛)

The infrastructure is all there in the form of

runnable_dependencies

, in

adhoc_tool

and

shell_command

, but it’s limited to working with those targets at the moment. We need to work on a more systematic way to add

runnable_dependencies

to rely on

One odd thing here too is that in order for this to work,

pyspark

needs to find Java. Meaning the target that represents "Java" should carry not just the files on disk, but an env var (

JAVA_HOME

) associated. I have other (non-Java) use-cases for having the output something also carry env vars for similar reasons. So something we need to figure out is how to make a target not just a digest, but a digest and env vars 😅

That’s

runnable_dependencies

🙂

I don't follow (also sorry @high-magician-46188 we might be derailing a bit)

@bitter-ability-32190 moving your tangent to #C0D7TNJHL