According to the docs,

pex_binary

will use the

interpreter_constraints

, which on my build I have:

[python]
interpreter_constraints = ["==3.12.*"]
enable_resolves = true

However, the shebang on my

pex_binary

is

#!/usr/bin/env python3.9

. I can perhaps override it, but I'm worried that something is going wrong. I see a few other odd things that might be caused by inappropriately using 3.9 instead of 3.12, such as parse errors using black that break on lines that should be fine with 3.12. Any help on how to debug into this would be appreciated.

I think I saw something like this happen to me a few days ago, but I thought I was imagining it (I was also doing some pex craziness). As a sanity, would you mind re-running (if it works) with

interpreter_constraints = ["==3.11.*"]

Sure, let me try.

Okay, this is what I was seeing - something to this effect, anyways. I had an issue packaging a pex using python 3.12, but converting to 3.11 and it worked (I also did have some

complete_platforms

shenanigans to deal with too.- but unsure if that was a blocker, or an anecdote)

Thank you, this gives me an interesting new datapoint. Any idea what might be going wrong, or is it just likely some pants bug? I could open an issue.

I'm not sure off the top of my head, because as soon as I worked around it, I promptly forgot it ever happened (literally days ago). I just assumed it was how I was messing around building SCIEs. I don't know which pex version is currently bundled with Pants, and I know there have been some shebang bugs/mods in the past few weeks - so there are a few things to look into.

Changing it back to 3.12 and removing

complete_platforms

also gave a correct shebang. Weird, since my JSON file I'm using seems pretty clear about 3.12 being the python version.

That's an even better datapoint 🙂

Interesting, thanks - possibly! I'd have to see if I can repro this just using pex, which I haven't looked at yet. But if so, then this isn't indicative of any wider issues, even though I still think pants is using 3.9 sometimes when it shouldn't.

Well, if Pex is doing it, then Pants definitely is 🙂 But if pex isn't doing it (on 2.3.0), then it's a Pants problem. However, on top of that, it is possible that somewhere we're using the downloaded interpreter version, in lieu of the desired interpreter_constraints. I haven't run into that, other than this past week

Interestingly, I just ran with -

ldebug

and it gives me a json parsing error for my json file in

complete_platforms

, and even more interestingly, I see from the stacktrace that it's running 3.9. That certainly seems wrong, but I know that goals default to some other python unless you tell it otherwise. On the other hand, the docs do say it pays attention to the python interpret constraints, but that isn't generally how other goals work.

So, pants itself will run 3.9, for sure - is that what you're seeing?

Yes

This isn't exactly answering your question, but using

sh_boot

might be a work around: https://github.com/pantsbuild/pants/pull/19925

FWIW, it isn't causing me an issue at the moment, but good to know, thank you. It would be nice to have some control over what Python is running, though.