Make SemanticsScope non-generic

[lnicola]

This slightly reduces the build times:

(compare to #1987 (comment))

[matklad]

Right, you are spot on @lnicola! I originally tried making Semantics non-generic, but didn't. I don't exactly remember why, it might be that either:

• there's some annoying dyn limitation
• I decided that sema.db being a concrete type would be useful

If it's the latter, we can have generic Semantics and concrete SemanticsImpl. I'll look into this if no one beats me to it.

[lnicola]

I'm not sure if Semantics can be non-generic, it's a bit of an annoying change.

[matklad]

I'm not sure if Semantics can be non-generic,

The following setup should work without breaking the API:

pub struct Semantics<'db, DB>{
  db: &'db DB
  imp: SemanticsImpl<'a>
}

struct SemanticsImpl<'a> {
  db: &'db dyn HirDatabase
  s2d_cache: RefCell<SourceToDefCache>,
  cache: RefCell<FxHashMap<SyntaxNode, HirFileId>>,
}

[lnicola]

What about the change in this PR? Do you think it's not a good idea?

[matklad]

Argh, I assumed this is an issue and not a PR. Gotta go make some pu erh I guess...

[matklad]

bors r+

This totally makes sense, but we need to do the rest as well!

[lnicola]

One issue is that some places seem to require SourceDatabase instead of HirDatabase, so we might need to add more Upcast impls?

[bors]

Build succeeded:

• TypeScript (ubuntu-latest)
• Rust (macos-latest)
• Rust (ubuntu-latest)
• TypeScript (windows-latest)
• Rust (windows-latest)
• TypeScript (macos-latest)

[matklad]

Adding upcasts should be fine!

[lnicola]

Ugh, after a while ra_hir_def is back to 176s for me, so maybe this PR didn't do anything 😕.