just opened https://github.com/pantsbuild/pants/issues/17354 to help describe a pitfall around where

EnvironmentNames

are consumed that will likely need to be present in 2.15.x at this point, but that we could maybe fix in 2.16.x.

@witty-crayon-22786 glad you could join me, the water’s wonderful 😉

drowning

@witty-crayon-22786 I think this may be where we need to start formalising the phases that Pants goes through and explain that all a bit better

yea. although it is a bit funky because rather than “phases”, per-se, it’s more like “chunks of the rule graph”. for example, it would be totally fine for codegen (running in a

remote

or

docker

environment) to call back to the BUILD file parsing rules (which would switch back to a local environment to run)

Stu: My only concern with that exact syntax is that

per_platform

looks like it can be used in other positions

This had me thinking if it would be possible/desirable to have platform specific BUILD files (or “sections” within i build file) so you define the whole target per platform (possibly with a common version and only the platform specific overrides in another…). maybe crazy idea..

so still not exactly like bazel’s phases, where the phase ends and you’re done

So, perhaps have different rule decorators for the various roles,

@local_rule

@runtime_rule

etc…

This had me thinking if it would be possible/desirable to have platform specific BUILD files (or “sections” within i build file) so you define the whole target per platform (possibly with a common version and only the platform specific overrides in another…). maybe crazy idea.

possibly… but as it stands, that would mean different build graphs / dependencies per platform, which pinning the graph to a single platform prevents currently… somewhat/mostly intentionally.

I had that idea too, but thought better of it so I can sleep at night 🙂

@curved-television-6568: it also gets into the domain of https://github.com/pantsbuild/pants/issues/11206 … because these “manually declared `dependencies`” which the

graph

@rules

compute are still manual, and are unlikely to be the fully accurate final dependencies of running a process

Glad I could muddy the water 🙂