[python-native-code] breaks Pantsd cache invalidation

[Eric-Arellano]

We should not be using os.environ. It needs to use the engine to request Environment instead.

pants/src/python/pants/backend/python/subsystems/python_native_code.py

Lines 19 to 20 in 8f62a1c

	"--cpp-flags",
	default=safe_shlex_split(os.environ.get("CPPFLAGS", "")),

We might need to add a new magic string like <CPPFLAGS>, similar to how we handle <PATH>.

pants/src/python/pants/backend/go/subsystems/golang.py

Lines 65 to 75 in 8f62a1c

	def go_search_paths(self, env: Environment) -> tuple[str, ...]:
	def iter_path_entries():
	for entry in self._go_search_paths:
	if entry == "<PATH>":
	path = env.get("PATH")
	if path:
	yield from path.split(os.pathsep)
	else:
	yield entry
	return tuple(OrderedSet(iter_path_entries()))

While fixing, if we don't use a magic string like <CPPFLAGS>, we should also use default_str so that our docs don't put whatever the value of the docs release is.

--

Even better would be removing [python-native-code] and consolidating it into something like [subprocess-environment] or [python]. The subsystem is confusing and a holdover from v1. I tried consolidating, but gave up because I couldn't avoid breaking the deprecation policy.

consolidating it into something like [subprocess-environment] or [python]

It seems like we've gone in the direction of env vars per language, so [subprocess-environment] is probably a misnomer.

[chrisjrn]

For deprecation purposes, I'm proposing adding a deprecation message for the envvar CPPFLAGS being set if the environment target is a local environment, and an error if not. The magic string approach seems OK for a new method that should explicitly only work on local environments.

Consolidating the various Python subsystems looks like a good idea, probably worth undertaking as a separate ticket though.