chore: Start infesting ide crates with 'db lifetime

[Veykril]

Got curious what this would look like, results in a bit of lifetime infestation :)

[ChayimFriedman2]

I would not merge this until we determine the effect on performance.

[Veykril]

analysis-stats for me varied by ~1% so very much within noise on my machine.

[ChayimFriedman2]

Did we even intern TraitEnvironment previously? And do we really need the fine-grained tracking here?

[Veykril]

Did we even intern TraitEnvironment previously?

No, I got partially curious if we would gain something from it being interned. It doesn't look like it though. (Technically hir::Type would become copy with this when we switch to the new trait solver)

And do we really need the fine-grained tracking here?

There is no fine grained tracking here really given its an interned struct, not tracked one.

[ChayimFriedman2]

There is no fine grained tracking here really given its an interned struct, not tracked one.

Are you sure? I think interned structs also have fine grained tracking.

Edit: It seems you're right, there's no fine grained tracking.

[ChayimFriedman2]

(Technically hir::Type would become copy with this when we switch to the new trait solver)

How so? We'll need to intern Ty for that too (and I'm completely not sure it's worth it)

[Veykril]

(Technically hir::Type would become copy with this when we switch to the new trait solver)

How so? We'll need to intern Ty for that too (and I'm completely not sure it's worth it)

From what I understood, we will be interning the type-ir Ty's within salsa such that we get copy-able handles which is a hard requirement by the new solver.

[ChayimFriedman2]

(Technically hir::Type would become copy with this when we switch to the new trait solver)

How so? We'll need to intern Ty for that too (and I'm completely not sure it's worth it)

From what I understood, we will be interning the type-ir Ty's within salsa such that we get copy-able handles which is a hard requirement by the new solver.

Didn't @jackh726 relax this requirement?

[Veykril]

Anyways this doesn't seem too useful, though I am inclined to include the 'db changes as this will have to occur at some point anyways the further we move into new salsa

[davidbarsky]

(Technically hir::Type would become copy with this when we switch to the new trait solver)

How so? We'll need to intern Ty for that too (and I'm completely not sure it's worth it)

From what I understood, we will be interning the type-ir Ty's within salsa such that we get copy-able handles which is a hard requirement by the new solver.

Didn't @jackh726 relax this requirement?

To some degree, but not really. I still think we should do copy-able handles if possible.

Does this have a memory usage impact, out of curiosity?

[Veykril]

Looks like this doesnt save memory (2MB) but it is apparently faster? 10 seconds speedup for type inference on CI which I wouldn't classify as within noise https://github.com/rust-lang/rust-analyzer/actions/runs/14282272667/job/40033194279?pr=19495 vs https://github.com/rust-lang/rust-analyzer/actions/runs/14282156180/job/40032940941

[davidbarsky]

Alright, nice! Maybe we should land this, since it could make Jack's life easier: https://rust-lang.zulipchat.com/#narrow/stream/185405-t-compiler.2Frust-analyzer/topic/Implmentating.20New.20Trait.20Solver/near/510313582

[Veykril]

Ah wait, forget that. We can't really take the numbers there seriously I think as those are debug builds (and they do seem to vary a lot). That is only memory usage there is sensible.