Fix inline const pattern unsafety checking in THIR

[matthewjasper]

Fix THIR unsafety checking of inline constants.

• Steal THIR in THIR unsafety checking (if enabled) instead of MIR lowering.
• Represent inline constants in THIR patterns.

[rustbot]

r? @b-naber

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

[rustbot]

Some changes might have occurred in exhaustiveness checking

cc @Nadrieril

[bors]

☔ The latest upstream changes (presumably #116623) made this pull request unmergeable. Please resolve the merge conflicts.

[Nadrieril]

:D I like this now, thanks for your patience with my comments!

[bors]

☔ The latest upstream changes (presumably #116737) made this pull request unmergeable. Please resolve the merge conflicts.

[b-naber]

I don't think it's that intuitive to hide the range pattern behind this nested PatKind::InlineConstant construction. Having an explicit range pattern makes this clearer imo. To me this seems to add complexity, but I don't know off the top of my head how these range patterns with inline constants are currently handled in later stages. Does this make things easier in later stages in a way that justifies introducing the complexity here?

[Nadrieril]

This makes things simpler for the consuming side: unsafety checking only cares that there are some inline constants around somewhere, pattern checking only cares that there's a range pattern whose extremities can be evaluated to bits. Carrying the unevaluated constant around felt invisible and brittle, my gut said danger.

But I am biased. I have an open PR (#116692) that reworks PatRange a lot, and I'm hoping I can eventually evaluate the consts eagerly and store bits in PatRange.

[b-naber]

Carrying the unevaluated constant around felt invisible and brittle, my gut said danger.

I don't understand this. Looking at const_to_pat, we always error if we find an unevaluatable constant. When we lower inline consts we currently always go through const_to_pat in the non-error case. Why would we carry unevaluated constants around?

Can't we introduce a new enum, that is used in Patkind::Constant and includes variants for InlineConst(mir::Const, LocalDefId) and mir::Const instead of modelling inline constants at the pattern level? The visitor in thir unsafety checking could still detect inline constants, pattern checking should be able to work with this afaict and we could continue to encode pattern ranges with inline constants as PatKind::Range directly, which seems less hacky than the nested PatKind::InlineConst imo.

[matthewjasper]

As I said above

Inline constants, like other constants, can be lowered to actual patterns in THIR. We could keep them as constants here and lower to patterns in both MIR building and exhaustiveness checking, but I don't think that's any better.

[Nadrieril]

Why would we carry unevaluated constants around?

Unsafety checking needs to look at the body of the inline constant, so if we want to use Const to carry it around, we must not evaluate it. Pretty sure an evaluated const doesn't know about the body it came from. This is what was done in ranges before I asked for this special InlineConstant behavior.

Can't we introduce a new enum, that is used in Patkind::Constant

We can't if we have an inline constant that evaluates to something that const_to_pat will deconstruct, e.g.:

fn main() {
    match Some(0) {
        const { Some(0) } => println!("got 0!"),
        Some(0) => unreachable!(),
        _ => println!("something else"),
    }
}

Here the constant needs to be deconstructed into a Pat that represents Some(0) so that we can do proper exhaustiveness/unreachable checking with it. The resulting Pat might not have any PatKind::Constant left at the end. Hence why we attach the inline constant in the middle of the pattern.

[b-naber]

Thanks for the clarifications.

@bors r+

[bors]

📌 Commit 8aea0e9 has been approved by b-naber

It is now in the queue for this repository.

[bors]

⌛ Testing commit 8aea0e9 with merge 848a387...

[bors]

☀️ Test successful - checks-actions
Approved by: b-naber
Pushing 848a387 to master...

[rust-timer]

Finished benchmarking commit (848a387): comparison URL.

Overall result: ❌✅ regressions and improvements - 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)	3.7%	[3.7%, 3.7%]	1
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-0.3%	[-0.3%, -0.2%]	3
All ❌✅ (primary)	3.7%	[3.7%, 3.7%]	1

Max RSS (memory usage)

Results

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-2.7%	[-2.7%, -2.7%]	1
All ❌✅ (primary)	-	-	0

Cycles

Results

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	5.1%	[2.6%, 7.4%]	4
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-	-	0

Binary size

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

Bootstrap: missing data
Artifact size: 304.40 MiB -> 304.40 MiB (-0.00%)