Shorten rustc_middle::ty::mod

[Nicholas-Baron]

Related to #60302.

This PR moves all Adt*, Assoc*, Generic*, and UpVar* types to separate files.
This, alongside some use reordering, puts mod.rs at ~2,200 lines, thus removing the // ignore-tidy-filelength.

The particular groups were chosen as they had 4 or more "substantive" members.

[rust-highfive]

r? @petrochenkov

(rust-highfive has picked a reviewer for you, use r? to override)

[petrochenkov]

Not sure who maintains this code now.
r? @jackh726 @nikomatsakis

[cjgillot]

What is the status of the work to move the type definitions to rustc_type_ir?

[jackh726]

What is the status of the work to move the type definitions to rustc_type_ir?

Somewhat blocked on figuring out the Encodable/Decodable situation. @detrumi and @zaharidichev have been doing here

[jackh726]

I big 👍 on me for the PR in general. For upvar.rs, I would like that to either be reverted or to add some other closure capture related things and changed to closure_capture or something like that :)

[jackh726]

This looks better. I think anything related directly to closure captures can go in closure.rs

[jackh726]

Looks good to me! r=me when CI is green

[Nicholas-Baron]

Looks good to me! r=me when CI is green

So when the CI returns all good, I write r=jackh726 to bors?

[jackh726]

@bors delegate+

Yes :)

[bors]

✌️ @Nicholas-Baron can now approve this pull request

[jackh726]

Let's also @bors rollup=iffy to this

[Nicholas-Baron]

@bors: r=jackh726

[bors]

📌 Commit d022142 has been approved by jackh726

[Nicholas-Baron]

@bors: rollup=iffy

[Nicholas-Baron]

I hope I did not mess up anything.

[jackh726]

@Nicholas-Baron should be fine :) CI should finish before bors can get to this. I'll keep in eye on it.

[bors]

⌛ Testing commit d022142 with merge b3ac526...

[bors]

☀️ Test successful - checks-actions
Approved by: jackh726
Pushing b3ac526 to master...

[Mark-Simulacrum]

This seems to have been a pretty big (~30%) regression on diesel-doc builds, presumably due to inlining changes - self-profile data in particular points at the evaluate_obligation query. It's possible we can just leave this, but might be good to add a couple inline attributes which should resolve the regression.

[Nicholas-Baron]

This seems to have been a pretty big (~30%) regression on diesel-doc builds, presumably due to inlining changes - self-profile data in particular points at the evaluate_obligation query. It's possible we can just leave this, but might be good to add a couple inline attributes which should resolve the regression.

Are there any particular functions which would benefit from inlining?

[Mark-Simulacrum]

It's hard to say without more investigation; probably a cachegrind or similar tooling might give a hint. I unfortunately don't have time to write up some directions, but a start would be to:

• checkout the rust compiler tree at the before & after commits around this PR's merge and build locally to get debuginfo
• run cachegrind on each with the diesel doc benchmark (either via the perf collector, some docs here, or more directly)
• look to see if any small functions stopped getting inlined

[Nicholas-Baron]

@Mark-Simulacrum I tried to use the rustc-perf/collector locally. However, linking LLVM with release-debuginfo = true leads to OOM for my 16 GB dev machine. I never got to building even stage0.

[Mark-Simulacrum]

Oh, you almost certainly only need debug info for the compiler, not llvm

[Nicholas-Baron]

@Mark-Simulacrum I have run into an issue which I have filed here. It looks like I am roadblocked for the current moment.

[Nicholas-Baron]

@Mark-Simulacrum I guess I am leaving this alone. There are probably larger performance gains to be had elsewhere in the compiler.