fix: cfg_select supports non token-tree tokens

[A4-Tacks]

Fixes #21702

• Fix single non-curly groups is ignored true => ((),), old: (),, new: ((),)

Example

const _: i32 = cfg_select! { true => 2 + 3, _ => 3 + 4 };

Before this PR

const _: i32 = cfg_select! { true => 2 /* expected a token tree after `=>` */+ 3, _ => 3 + 4 };

After this PR

const _: i32 = 2 + 3;

[ChayimFriedman2]

That's not how rustc does this and this is not the right way. This will break on e.g. paths with commas like foo::<T, U>. You need to parse an expression with the parser.

[A4-Tacks]

But it still needs to support pat, so I roughly delimited it with commas

[ChayimFriedman2]

What do you mean by "it still need to support pat"? You need to use the parser to parse an expr like with macro_rules matcher.

[A4-Tacks]

Like let cfg_select! { _ => () } = ();,
but now I realize that it's actually parsing according to the expression, it just happens to be a valid pat

[ChayimFriedman2]

That is definitely not enough. The only way to parse an expression reliably is to invoke the parser.

[A4-Tacks]

That is definitely not enough.

Is there anything else besides the closure returning the location?

The only way to parse an expression reliably is to invoke the parser.

Do we have an expression parser for TtIter?

[ChayimFriedman2]

Is there anything else besides the closure returning the location?

What do you mean?

Do we have an expression parser for TtIter?

No, but you can look at the macro_rules matcher. Look at the function named expect_fragment().

[rustbot]

This PR was rebased onto a different master commit. Here's a range-diff highlighting what actually changed.

Rebasing is a normal part of keeping PRs up to date, so no action is needed—this note is just to help reviewers.