I've found an interesting issue. Haven't quite figured out what is happening; but if I take my pants-backends repo and add

pants.backend.shell

as a backend, it crashes with rule graph errors on all unreleased versions I can test. Without the shell backend it works fine. Without the OCI backend it works fine. The combo itself is toxic. This repros on 2.16.0rc5, 2.17.0rc0. Wondering if this is related to

RunInSandboxBehavior

, just unexpected that it explodes like that. 2.16.0.dev2 is the last one that works.

Likely. We're ooching towards a world where for unions you're expected to register the rule using

*(subclass).rules()

. (which contains maybe some "inner rules"). From your

RunInSandboxBehavior

hint, it might be that it's missing the relevant rule for sandbox running

Yeah; maybe. I'll have to debug later; this'd be a blocker for upgrading to 2.16 so hopefully not too hard to resolve. Seems a bit iffy in general if two backends can explode each other like that. I only found this by accident because I wanted to test a quick thing with go-code in my prod repo... Thus using PANTS_VERSION=2.16.

If the current thinking is true, you can hide the import/attribute behind

if PANTS_VERSION

flags

Yeah; I do that in so many places already 😄

🪦

Oh yeah for sure. I'm happy things are changing... But rule graph errors are one of my least favourite things to debug.

You are in very good company

Yeah, having two backends explode together is very ungood

"ungood" ™️

So; this is related to the

RunFieldSet.rules()

change, not the one I thought. I'm going to guess the shell backend consumes some Union that I wasn't properly setting since I had the old

UnionRule

setup.