Macro definitions should be regular HIR items

[cjgillot]

Macro definitions are currently handled separately from other item kinds in HIR, either as MacroDef nodes inside hir::Crate::exported_macros or as using the hir::Crate::non_exported_macro_attrs.

They should be transformed into additional variants in the hir::ItemKind enum, and iterated over like any other item.

Instructions:

• introduce two additional variants hir::ItemKind::ExportedMacro { ast: ast::MacroDef } and hir::ItemKind::NonExportedMacro;
• move the macro case in rustc_ast_lowering::item::LoweringContext::lower_item into lower_item_kind;
• remove hir::OwnerNode::{MacroDef, NonExportedMacro} variants.

I recomment waiting on #83723 and #87234 to be merged.

Original idea by @petrochenkov in #83723 (comment)
Should help with #73754

[jyn514]

See also #77862, #87257, #79396 (comment).

[inquisitivecrystal]

ExportedMacro (or indeed, exported_macros) is a fairly misleading name, since this includes non exported MBE 2.0 macros. I brought this up in #86968, but I haven't opened a follow-up discussion (partly because of a lack of bandwidth, and partly because it seemed like the relevant code needed to be rewritten at some point).

Incidentally, how difficult do you think this task will be?

[jyn514]

@inquisitivecrystal my understanding without looking into it much is that basically everything that touches exported_macros today will have to be rewritten. So at least E-medium, possibly E-hard. But @petrochenkov would have a better idea than me.

[inquisitivecrystal]

Alright. I'm currently looking at whether this might be within my ability to implement. I don't like saying I'll do something until I I've checked that I have some chance of figuring it out, but the E-mentor helps a bit with that.

[inquisitivecrystal]

I'm going to give this a go! There's no way of knowing for sure if I can do it until I actually try it out, and I can't do that properly until the PRs mentioned in the initial post are merged.

[cjgillot]

@inquisitivecrystal : #87234 contains a perf regression which may very well be caused by this bug. I think you can go ahead with the implementation, I will rebase it on top of your work.
923ae28c0d60720a76782803ff478e49b099746c introduces the hir::OwnerNode::NonExportedMacro variant I mention in the top post. This commit will point the right place to modify, but you should not need to cherry-pick it.
Don't hesitate to contact me on zulip if you need more precisions.

[inquisitivecrystal]

I'm on it!