CStore switch FxHashMap to IndexVec

[Eh2406]

This is a first attempt to fix#46876.

[rust-highfive]

r? @eddyb

(rust_highfive has picked a reviewer for you, use r? to override)

[arielb1]

Your code does not compile:

[00:15:32] error[E0507]: cannot move out of borrowed content
[00:15:32] --> /checkout/src/librustc_metadata/cstore.rs:114:9
[00:15:32] |
[00:15:32] 114 | self.metas.borrow().get(cnum).unwrap().unwrap().clone()
[00:15:32] | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ cannot move out of borrowed content
[00:15:32] [00:15:32] error[E0596]: cannot borrow immutable local variable `met` as mutable
[00:15:32] --> /checkout/src/librustc_metadata/cstore.rs:121:13
[00:15:32] |
[00:15:32] 119 | let met = self.metas.borrow_mut();
[00:15:32] | --- consider changing this to `mut met`
[00:15:32] 120 | while met.len() <= cnum.index() {
[00:15:32] 121 | met.push(None);
[00:15:32] | ^^^ cannot borrow mutably
[00:15:32] [00:15:32] error[E0596]: cannot borrow immutable local variable `met` as mutable
[00:15:32] --> /checkout/src/librustc_metadata/cstore.rs:123:9
[00:15:32] |
[00:15:32] 119 | let met = self.metas.borrow_mut();
[00:15:32] | --- consider changing this to `mut met`
[00:15:32] ...
[00:15:32] 123 | met[cnum] = Some(data);
[00:15:32] | ^^^ cannot borrow mutably
[00:15:32] 

[arielb1]

Use self.metas.borrow()[cnum].clone().unwrap() to fix the borrow error

[arielb1]

This needs to be let mut met

[Eh2406]

Thanks for the heads up. I got distracted trying to test locally.

[Eh2406]

Ok, I am confused, what does that test fail have to do with my changes?

[pnkfelix]

@Eh2406 Assuming that you are referring to the change to the output for ui/print_type_sizes/niche-filling.rs ... here is my current understanding...

I think we are seeing some (currently undiagnosed) non-deterministic treatment of std::foo::bar vs core::foo::bar , and your change is causing a ripple effect there.

You can see some recent notes from me on this at #46905, in particular the commit message for 5f85764

Note in particular that that tests originally said core::cmp::Ordering; it was only changed to std::cmp::Orderingvery recently, in #46708 (which I am becoming increasingly concerned about and am considering reverting in favor of #46838 ...).

• Of course the ideal thing would be for someone to explain why it switched from core to std in the first place as part of Fix visible_parent_map to choose globally minimal paths #46708 ; I had thought it was obviously better at the time when I wrote Fix visible_parent_map to choose globally minimal paths #46708, but I have since come to understand the point of the comment noted in the log message for 5f85764 :

Entry::Occupied(mut entry) => {// If `child` is defined in crate `cnum`, ensure// that it is mapped to a parent in `cnum`.if child.krate == cnum && entry.get().krate != cnum {
entry.insert(parent);}}

• namely, I think the idea of that code is that its better to prefer the path that corresponds to an items definition, if that path is public, rather than a path of a mere re-export of the item.

---

In the meantime, I personally would be fine with you just updating the print-type-sizes.stdout to match the new output. (Which after all is the same as the old output...)

[Eh2406]

Ok so I switched from core::cmp::Ordering to std::cmp::Ordering in the test, we will see if that fixes things.

[Eh2406]

Any ideas what is up with these errors? Why would this change lead to differences in the error messages?

[eddyb]

r? @pnkfelix

[Eh2406]

Is there any Ideas on how to proceed? Should I try a rebase?

[Eh2406]

All the PRs referenced by @pnkfelix have been merged, so I tried a rebase (and I squashed while I was doing). We will see if it works.

[Eh2406]

Ping. The test got fixed by other prs. So this is now back to being a small simple change, with green CI.

[eddyb]

@bors r+

[bors]

📌 Commit 2c69473 has been approved by eddyb

[bors]

⌛ Testing commit 2c69473 with merge 10f980c9f3977872cf1fddb6a4e9986d945c2bbd...

[bors]

💔 Test failed - status-appveyor

[Eh2406]

Time out on appveyor. So did the last few commits. So we are on hold for that to get straightened out, yes?

[kennytm]

@bors retry #46903 (3 hour timeout)

[bors]

⌛ Testing commit 2c69473 with merge 8d1a302...

[bors]

☀️ Test successful - status-appveyor, status-travis
Approved by: eddyb
Pushing 8d1a302 to master...