Add a lint against never type fallback affecting unsafe code

[WaffleLapkin]

I'm not very happy with the code quality... VecGraph not allowing you to get predecessors is very annoying. This should work though, so there is that. (ended up updating VecGraph to support getting predecessors)

First few commits are from #123934 #123980

[rustbot]

r? @wesleywiser

rustbot has assigned @wesleywiser.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[rustbot]

Some changes occurred to MIR optimizations

cc @rust-lang/wg-mir-opt

Some changes occurred in coverage instrumentation.

cc @Zalathar

[Noratrieb]

the lint description should mention what never type fallback actually is, I think most people are not aware of it at all

[WaffleLapkin]

That's one of the annoying problems of all the never type stuff, fallback can't be explained without going somewhat indepth :(

[WaffleLapkin]

I added an explanation, does it look good?

[WaffleLapkin]

Thinking about VecGraph more: we'll need to have predecessors for another lint anyway. One of the lints @nikomatsakis suggested checks for coercions like this:

?F -> ?1
       ^
       |
      ?N

Where ?F is a diverging variable which experiences fallback and ?N is non-diverging variable.

The algorithm to actually check this is

1. From the variable which is falling back do a DFS, recording all the nodes we see and sinks (variables which are not coerced to anything)
2. From sinks do a backwards DFS recording all nodes which we did not record in the 1-st step
3. The nodes recorded in the 2-nd step are the ones we need to lint against

But VecGraph does not have predecessors and so can't do a backwards DFS.

This PR (and a future one for the third lint) are struggling/will struggle from a similar problem: we need to be able to go backwards (because this lint does not care about coercion direction -- since rust infers both ways).

I will try to use just Graph, but I'm not entirely sure its interface will work here... TBQH I wish rustc would just use an off-the-shelf library for graphs, instead of rolling its own half baked and undocumented version.

[WaffleLapkin]

Okay. VecGraph actually exists for a reason -- having indices and edges upfront simplifies things quite a bit. Graph won't work nicely here, because we'd have to keep TyVid <-> NodeIndex map.

I'll try adding opt-in backreferences to VecGraph.

[bors]

☔ The latest upstream changes (presumably #124026) made this pull request unmergeable. Please resolve the merge conflicts.

[WaffleLapkin]

@compiler-errors I've added support for unsafe methods, alongside other unsafe things which may have a type infered from the environment (like pointer dereference, union access, etc). The change is annoyingly big, but yeah...

[compiler-errors]

Split this into 5 slugs and do:

#[derive(Diagnostic)]
enum NeverTypeFallbackFlowingIntoUnsafe {
  #[diag(hir_typeck_never_type_fallback_flowing_into_unsafe_call)]  
  Call,
  ...
}

And then

hir_typeck_never_type_fallback_flowing_into_unsafe_call = never type fallback affects this call to an `unsafe` function

---

https://github.com/projectfluent/fluent/wiki/Good-Practices-for-Developers#prefer-separate-messages-over-variants-for-ui-logic

[WaffleLapkin]

I've kept the reason enum and then convert to the diagnostic, cause I find using errors::* until the literal "emit" somewhat weird. But LMK if I should just return the diagnostic from the "find unsafe infer vars" function.

[compiler-errors]

r=me after splitting the fluent slugs

[bors]

☔ The latest upstream changes (presumably #124558) made this pull request unmergeable. Please resolve the merge conflicts.

[compiler-errors]

that's fine

@bors r+

[bors]

📌 Commit b562617 has been approved by compiler-errors

It is now in the queue for this repository.

[bors]

⌛ Testing commit b562617 with merge fcc06c8...

[bors]

☀️ Test successful - checks-actions
Approved by: compiler-errors
Pushing fcc06c8 to master...

[rust-timer]

Finished benchmarking commit (fcc06c8): comparison URL.

Overall result: ❌ regressions - no action needed

@rustbot label: -perf-regression

Instruction count

This is a highly reliable metric that was used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	1.3%	[1.1%, 1.5%]	2
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-	-	0

Max RSS (memory usage)

This benchmark run did not return any relevant results for this metric.

Cycles

This benchmark run did not return any relevant results for this metric.

Binary size

This benchmark run did not return any relevant results for this metric.

Bootstrap: 674.637s -> 671.059s (-0.53%)
Artifact size: 315.98 MiB -> 315.99 MiB (0.00%)