fix handling of --all-groups and --no-default-groups flags

[Gankra]

This is a rewrite of the groups subsystem to have more clear semantics, and some adjustments to the CLI flag constraints. In doing so, the following bugs are fixed:

• --no-default-groups --no-group foo is no longer needlessly rejected
• --all-groups --no-default-groups now correctly evaluates to --all-groups where previously it was erroneously being interpretted as just --no-default-groups
• --all-groups --only-dev is now illegal, where previously it was accepted and mishandled, as if it was a mythical --only-all-groups flag

Fixes #10890
Closes #10891

[zanieb]

--only-dev and --dev do not conflict, --no-dev?

Suggested change

	#[arg(long, conflicts_with_all = ["group", "dev", "all_groups"])]
	pub only_dev: bool,
	#[arg(long, conflicts_with_all = ["group", "no_dev", "all_groups"])]
	pub only_dev: bool,

[Gankra]

The idea I'm trying to express here is essentially that --dev --only-dev is "what are you even doing".

This is more clear with --group A --only-group B, which is clearly "what no", but rejecting that and not rejecting --group A --only-group A is a pain for very little benefit. Since --dev --only-dev essentially desugars to that, I figured it's best to reject it for uniformity (and the followup code essentially assumes only one of these flags is ever true, although in a trivial way where we could fix it to "saturate" to --only-dev).

[zanieb]

That makes sense. I'd prefer to just allow redundant options, but understand it can get complicated and if it's not worth it it's not a big deal. Is it a breaking change though?

--only-dev --no-dev should still error though, no? Or does it not because we say that "exclusions override" and that's fine?

[Gankra]

Yeah I wasn't sure how exclusions and overrides interact. It should probably be tightened up but I didn't want to break something (I'll test it out / read the docs properly to figure out how this should be handled).

In terms of breaking... hmm... seems like the old impl tried harder to rationalize both inputs being provided... but in a bit of an adhoc way...

uv/crates/uv-configuration/src/dev.rs

Lines 234 to 256 in ac10042

	if matches!(dev, Some(DevMode::Only)) {
	unreachable!("cannot specify both `--only-dev` and `--group`")
	};
	// Ensure that `--no-group` and `--group` are mutually exclusive.
	group.retain(|group| !no_group.contains(group));
	Some(GroupsSpecification::Include {
	include: IncludeGroups::Some(group),
	exclude: no_group,
	})
	} else if !only_group.is_empty() {
	if matches!(dev, Some(DevMode::Include)) {
	unreachable!("cannot specify both `--dev` and `--only-group`")
	};
	// Ensure that `--no-group` and `--only-group` are mutually exclusive.
	only_group.retain(|group| !no_group.contains(group));
	Some(GroupsSpecification::Only {
	include: only_group,
	exclude: no_group,
	})

[zanieb]

In general, I'd say a breaking change here would not be ideal. I'd suggest PR'ing some tests for edge-cases to main first so we can see if they change here? If it seems hard to avoid a regression, then I guess we should wait for 0.6?

[Gankra]

Ok cool so this is the Final Potential Breakage to hash out (collapsed a few others that I regard as Equivalent).

Looking closer, the current code:

• Hard panics(!) on --group --only-dev and --dev --only-group
• CLI conflicts --group --only-group
• CLI conflicts --dev --only-dev
• Saturates --only-dev --dev to --only-dev
• CLI conflicts --only-dev --no-dev
• Positionally Overrides --dev --no-dev --dev ...

So overall the current code hard rejects trying to mix --only-* with --group/--dev, except in the exact case of --only-dev --dev, which it turns into --only-dev. Which is fair enough, if a bit inconsistent. It's interesting that there's no overriding between --only-dev and --no-dev but I don't want to poke that bear in this PR.

I've adjusted things to preserve this behaviour, although the hard panics are now CLI conflicts.

[zanieb]

This looks a little weird?

Related https://github.com/astral-sh/uv/pull/11224/files#r1942059001

[Gankra]

I agree it's a bit weird, but on the other hand I do kind of like that it's really hammering in "there were two different things I was theoretically supposed to be inputting and I'm doing Neither of them".

[zanieb]

Weird to allow this but not // --group and --only-group should error if they name same things but... eh.

[Gankra]

Preserving previous behaviour :)