Fix split index calculation in check_type_alias_where_clause_location

[stuuupidcat]

Fix the calculation of the split index in check_type_alias_where_clause_location by ignoring before-equals-sign-clauses with attributes.

Fix #138010.

cc @frank-king.

[rustbot]

r? @BoxyUwU

rustbot has assigned @BoxyUwU.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[rustbot]

Some changes occurred in src/tools/rustfmt

cc @rust-lang/rustfmt

[BoxyUwU]

Filtering things which have attributes applied to them isn't really correct because it's not necessarily the case that all cfgd where clauses will be removed during macro expansion, e.g.

#![feature(where_clause_attrs, cfg_boolean_literals, lazy_type_alias)]
#![expect(incomplete_features)]

struct Foo;
trait Trait {}

impl Trait for Foo {}

type MixedWhereBounds
where
    #[cfg(true)]
    Foo: Trait,
= Foo
where
    (): Sized;

Here we have a where clause before the = that has an attribute applied to it, yet the where clause will still be present after macro expansion.

If you run the compiler on this example with your changes, and then remove the cfg, you'll get different diagnostics output even though post-expansion it'll be the same set of where clauses. That's caused by this before_with_attr_count being 1 and then when we split the where clauses into before/after you wind up splitting at 0 so both clauses end up in the "after" bucket.

Can you add both version as a test (with/without cfg(true))?

[BoxyUwU]

I'm not really sure what the correct way to implement this is 🤔

[stuuupidcat]

Thank you for your detailed explanation, I will add a few tests first.