mGCA: Track whether a const is a type const as part of `DefKind`

[BoxyUwU]

When replacing #[type_const] with proper type const syntax in #152324, a query was introduced to check whether a const item is a type const or not. This (may have) caused the perf regression (#152324 (comment)), we should experiment with replacing this query with a flag in DefKind::Const/AssocConst.

This would probably be a nice cleanup regardless of whether it fixes the perf regression

[JayanAXHF]

@rustbot claim

[eggyal]

I'd be happy to experiment with this, but won't be able to get onto it until Monday at the earliest.

[JayanAXHF]

Hey @BoxyUwU, i'd like to work on this. If you could point me to the relevant files/modules, i'll try to file a PR. Thanks

[BoxyUwU]

DefKind is defined here: https://doc.rust-lang.org/nightly/nightly-rustc/src/rustc_hir/def.rs.html#100-196. I'd start by changing the Const/AssocConst variants to have a bool on them determining whether its a type const or not.

From there it should be a matter of updating a lot of matches to be Const(_) or AssocConst(_). The interesting change will be updating the logic in rustc_resolve/src/def_collector.rs to make a DefKind::Const(true/false) corresponding to whether the const item is a type const or not.

Once that's done you can rewrite the is_type_const fn (https://doc.rust-lang.org/nightly/nightly-rustc/src/rustc_middle/ty/context.rs.html#1115-1124) to look at the DefKind instead of calling the is_rhs_type_const query. It should also be possible to remove the is_rhs_type_const query but we can get to that once you've gotten further along with this :)

[JayanAXHF]

Hey @BoxyUwU, i'm almost through with changing the instances of DefKind::(Const|AssocConst) throughout the codebase. After that, i'll remove the is_rhs_type_const query too, and then submit a PR. Thanks!