Check for occupied niches

[saethlin]

Implementation of rust-lang/compiler-team#624

Crater run has 62 crates that hit the check, 43 of those are published to crates.io. I see a lot of null function pointers and use of mem::uninitialized where the 0x1-filling collides with an enum niche.

But that is with full niche checks; checking transmute, plus any place where that we Copy, Move, or Inspect. Such checking is definitely too thorough to be on by default because it is 2x compile time overhead.

---

During implementation, this ran into llvm/llvm-project#68381

r? @ghost

[rustbot]

Some changes occurred to MIR optimizations

cc @rust-lang/wg-mir-opt

[saethlin]

This is not done yet. I wish generalized niche checks were as simple as what I whipped up here, and also CI doesn't pass for... reasons.

[bors]

☔ The latest upstream changes (presumably #111082) made this pull request unmergeable. Please resolve the merge conflicts.

[rust-timer]

Finished benchmarking commit (a3ce34e): comparison URL.

Overall result: ❌ regressions - ACTION NEEDED

Benchmarking this pull request likely means that it is perf-sensitive, so we're automatically marking it as not fit for rolling up. While you can manually mark this PR as fit for rollup, we strongly recommend not doing so since this PR may lead to changes in compiler perf.

Next Steps: If you can justify the regressions found in this try perf run, please indicate this with @rustbot label: +perf-regression-triaged along with sufficient written justification. If you cannot justify the regressions please fix the regressions and do another perf run. If the next run shows neutral or positive results, the label will be automatically removed.

@bors rollup=never
@rustbot label: -S-waiting-on-perf +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)	2.5%	[0.2%, 11.0%]	17
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	2.5%	[0.2%, 11.0%]	17

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)	4.3%	[0.6%, 6.7%]	7
Regressions ❌ (secondary)	1.3%	[0.7%, 2.2%]	6
Improvements ✅ (primary)	-2.2%	[-2.7%, -1.8%]	2
Improvements ✅ (secondary)	-3.6%	[-3.6%, -3.6%]	1
All ❌✅ (primary)	2.9%	[-2.7%, 6.7%]	9

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)	4.6%	[1.1%, 11.1%]	8
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	4.6%	[1.1%, 11.1%]	8

Binary size

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)	0.8%	[0.1%, 3.4%]	54
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-0.0%	[-0.0%, -0.0%]	1
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	0.8%	[-0.0%, 3.4%]	55

Bootstrap: 635.311s -> 636.977s (0.26%)
Artifact size: 304.40 MiB -> 304.39 MiB (-0.01%)

[saethlin]

I've done a bit more micro-opt, by avoiding ptrtoint for pointer niches, and for partial niches that start at offset 0 I'm just doing a pointer load. It's less LLVM IR at least...
@bors try @rust-timer queue

[bors]

⌛ Trying commit 2ae0dbc with merge 57f535e...

[bors]

☀️ Try build successful - checks-actions
Build commit: 57f535e (57f535edd0b2e7c91f9de22525f6a9073f8745ec)

[rust-timer]

Finished benchmarking commit (57f535e): comparison URL.

Overall result: ❌ regressions - ACTION NEEDED

Benchmarking this pull request likely means that it is perf-sensitive, so we're automatically marking it as not fit for rolling up. While you can manually mark this PR as fit for rollup, we strongly recommend not doing so since this PR may lead to changes in compiler perf.

Next Steps: If you can justify the regressions found in this try perf run, please indicate this with @rustbot label: +perf-regression-triaged along with sufficient written justification. If you cannot justify the regressions please fix the regressions and do another perf run. If the next run shows neutral or positive results, the label will be automatically removed.

@bors rollup=never
@rustbot label: -S-waiting-on-perf +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)	1.3%	[0.1%, 4.0%]	17
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	1.3%	[0.1%, 4.0%]	17

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)	3.5%	[0.4%, 9.1%]	6
Regressions ❌ (secondary)	1.6%	[1.1%, 2.0%]	2
Improvements ✅ (primary)	-2.6%	[-2.6%, -2.6%]	1
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	2.6%	[-2.6%, 9.1%]	7

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)	2.7%	[1.2%, 4.6%]	6
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	2.7%	[1.2%, 4.6%]	6

Binary size

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)	0.5%	[0.0%, 1.6%]	50
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-0.0%	[-0.0%, -0.0%]	1
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	0.5%	[-0.0%, 1.6%]	51

Bootstrap: 635.492s -> 636.287s (0.13%)
Artifact size: 304.42 MiB -> 304.38 MiB (-0.01%)

[saethlin]

To reproduce the crash, compile this:

#[derive(thiserror::Error)]
#[error("...")]
pub enum ErrorEnum {
    Test {
        source: u8, 
    },  
}

With a toolchain built with

[rust]
debug = true
codegen-units = 1

[dianqk]

To reproduce the crash, compile this:

#[derive(thiserror::Error)]
#[error("...")]
pub enum ErrorEnum {
    Test {
        source: u8, 
    },  
}

With a toolchain built with

[rust]
debug = true
codegen-units = 1

I can confirm that this is the same issue as #116976. LLVM 17.0.4 should be released soon.

[bors]

☔ The latest upstream changes (presumably #117498) made this pull request unmergeable. Please resolve the merge conflicts.

[saethlin]

@bors try @rust-timer queue

[bors]

⌛ Trying commit 5c19cfd with merge eaa4669...

[bors]

☀️ Try build successful - checks-actions
Build commit: eaa4669 (eaa46697f8834416cadb541c3d8b28c9d7da8e68)

[rust-timer]

Finished benchmarking commit (eaa4669): comparison URL.

Overall result: ❌✅ regressions and improvements - ACTION NEEDED

Benchmarking this pull request likely means that it is perf-sensitive, so we're automatically marking it as not fit for rolling up. While you can manually mark this PR as fit for rollup, we strongly recommend not doing so since this PR may lead to changes in compiler perf.

Next Steps: If you can justify the regressions found in this try perf run, please indicate this with @rustbot label: +perf-regression-triaged along with sufficient written justification. If you cannot justify the regressions please fix the regressions and do another perf run. If the next run shows neutral or positive results, the label will be automatically removed.

@bors rollup=never
@rustbot label: -S-waiting-on-perf +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)	1.1%	[0.1%, 3.5%]	17
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-0.6%	[-1.2%, -0.2%]	3
All ❌✅ (primary)	1.1%	[0.1%, 3.5%]	17

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)	1.8%	[0.2%, 7.9%]	17
Regressions ❌ (secondary)	3.9%	[2.6%, 5.2%]	3
Improvements ✅ (primary)	-2.1%	[-4.6%, -0.7%]	5
Improvements ✅ (secondary)	-2.2%	[-2.2%, -2.2%]	1
All ❌✅ (primary)	0.9%	[-4.6%, 7.9%]	22

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)	1.6%	[0.4%, 3.3%]	8
Regressions ❌ (secondary)	2.0%	[2.0%, 2.0%]	1
Improvements ✅ (primary)	-0.4%	[-0.4%, -0.4%]	1
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	1.4%	[-0.4%, 3.3%]	9

Binary size

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)	0.5%	[0.0%, 1.1%]	45
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-0.1%	[-0.2%, -0.0%]	3
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	0.4%	[-0.2%, 1.1%]	48

Bootstrap: 674.395s -> 673.749s (-0.10%)
Artifact size: 311.13 MiB -> 311.07 MiB (-0.02%)

[Dylan-DPC]

@saethlin any updates on this?

[saethlin]

Yes; I'm working on an alternative implementation. It's almost ready for a draft PR.

[saethlin]

Replaced by #121174