[internal] Better error message for ./pants repl when using multiple resolves

[Eric-Arellano]

First part of #14295. Finishes Python - JVM still needs it.

[ci skip-rust]
[ci skip-build-wheels]

[Eric-Arellano]

Blocked by figuring out what the command should be, specifically whether to recommend using JQ vs if we need first-class support for filter --field. Imo, JQ is still better than the status quo!

[stuhood]

The JVM should already have a good error message for this: locating a resolve for the roots is a common operation across all of the JVM goal implementations, and the only difference for repl is that it does zero partitioning beforehand.

pants/src/python/pants/jvm/resolve/coursier_fetch.py

Lines 76 to 90 in 6f7fcfc

	class NoCompatibleResolve(Exception):
	"""No compatible resolve could be found for a set of targets."""
	def __init__(self, jvm: JvmSubsystem, msg_prefix: str, incompatible_targets: Iterable[Target]):
	targets_and_resolves_str = bullet_list(
	f"{t.address.spec}\t{jvm.resolves_for_target(t)}" for t in incompatible_targets
	)
	super().__init__(
	f"{msg_prefix}:\n"
	f"{targets_and_resolves_str}\n"
	"Targets which will be merged onto the same classpath must have at least one compatible "
	f"resolve (from the [resolve]({doc_url('reference-deploy_jar#coderesolvecode')}) or "
	f"[compatible_resolves]({doc_url('reference-java_sources#codecompatible_resolvescode')}) "
	"fields) in common."
	)

[Eric-Arellano]

Python already has that same error message :)

pants/src/python/pants/backend/python/util_rules/pex_from_targets.py

Lines 204 to 234 in 6f7fcfc

	# Note: Inspired by `coursier_fetch.py`.
	class NoCompatibleResolveException(Exception):
	"""No compatible resolve could be found for a set of targets."""
	def __init__(
	self, python_setup: PythonSetup, msg_prefix: str, relevant_targets: Iterable[Target]
	) -> None:
	resolves_to_addresses = defaultdict(list)
	for tgt in relevant_targets:
	if tgt.has_field(PythonResolveField):
	resolve = tgt[PythonResolveField].normalized_value(python_setup)
	resolves_to_addresses[resolve].append(tgt.address.spec)
	elif tgt.has_field(PythonRequirementCompatibleResolvesField):
	resolves = tgt[PythonRequirementCompatibleResolvesField].normalized_value(
	python_setup
	)
	for resolve in resolves:
	resolves_to_addresses[resolve].append(tgt.address.spec)
	formatted_resolve_lists = "\n\n".join(
	f"{resolve}:\n{bullet_list(sorted(addresses))}"
	for resolve, addresses in sorted(resolves_to_addresses.items())
	)
	super().__init__(
	f"{msg_prefix}:\n\n"
	f"{formatted_resolve_lists}\n\n"
	"Targets which will be used together must all have the same resolve (from the "
	f"[resolve]({doc_url('reference-python_test#codeexperimental_resolvecode')}) or "
	f"[compatible_resolves]({doc_url('reference-python_requirement#codeexperimental_compatible_resolvescode')}) "
	"fields) in common."
	)

And it will trigger with repl already becauase we use pex_from_targets.py. The only difference with this PR is that we will more eagerly throw that exception, specifically with a new msg_suffix that gives you the JQ workaround. Do we believe giving you the clunky JQ + peek workaround to run is better than saying no workaround at all? I think so. Wdyt?

[stuhood]

Do we believe giving you the clunky JQ + peek workaround to run is better than saying no workaround at all? I think so. Wdyt?

I think that it is worthwhile, yea. But maybe not a high priority. I think that "filter by field value" is worth doing sooner rather than later in support of #13882, and we can recommend it more widely / link to a page about it.

[Eric-Arellano]

But maybe not a high priority.

It's easy to finish - all that's left is figuring out the JQ snippet, then writing one simple unit test. I'm not very fluent with JQ but I think we have enough examples in the docs I could scrape something together.

[stuhood]

I'm not very fluent with JQ but I think we have enough examples in the docs I could scrape something together.

I added an example here: https://docs.google.com/document/d/1mzZWnXiE6OkgMH_Dm7WeA3ck9veusQmr_c6GWPb_tsQ/edit#heading=h.hzxpmsxjb6jw

[Eric-Arellano]

Bump