Add some track_caller info to precondition panics #129658

saethlin · 2024-08-27T15:06:59Z

Currently, when you encounter a precondition check, you'll always get the caller location of the implementation of the precondition checks. But with this PR, you'll be told the location of the invalid call. Which is useful.

I thought of this while looking at #129642 (comment).

The changes to tests/ui/const* happen because the const-eval interpreter skips #[track_caller] frames in its backtraces.

The perf implications of this are:

Increased debug binary sizes. The caller_location implementation requires that the additional data we want to display here be stored in const allocations, which are deduplicated but not across crates. There is no impact on optimized build sizes. The panic path and the caller location data get optimized out.
The compile time hit to opt-incr-patched bitmaps happens because the patch changes the line number of some function calls with precondition checks, causing us to go from 0 dirty CGUs to 1 dirty CGU.
The other compile time hits are marginal but real, and due to doing a handful of new queries. Adding more useful data isn't completely free.

saethlin · 2024-08-27T16:30:13Z

@bors try @rust-timer queue

bors · 2024-08-27T16:31:24Z

⌛ Trying commit 18e2a95 with merge 7798f9b...

Add some track_caller info to precondition panics r? `@ghost` Thought of this while looking at rust-lang#129642 (comment)

library/core/src/panicking.rs

bors · 2024-08-27T18:24:54Z

☀️ Try build successful - checks-actions
Build commit: 7798f9b (7798f9b35d0cd727f26631c015620e3dfe62e1f6)

rust-timer · 2024-08-27T19:39:49Z

Finished benchmarking commit (7798f9b): comparison URL.

Overall result: no relevant changes - no 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.

@bors rollup=never
@rustbot label: -S-waiting-on-perf -perf-regression

Instruction count

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

Max RSS (memory usage)

Results (secondary -2.7%)

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%	[-3.1%, -2.3%]	2
All ❌✅ (primary)	-	-	0

Cycles

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

Binary size

Results (primary 0.1%, secondary 0.1%)

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)	0.1%	[0.0%, 0.6%]	47
Regressions ❌ (secondary)	0.1%	[0.0%, 0.4%]	35
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	0.1%	[0.0%, 0.6%]	47

Bootstrap: 749.925s -> 752.783s (0.38%)
Artifact size: 338.74 MiB -> 338.79 MiB (0.02%)

saethlin · 2024-08-27T19:55:42Z

That looks possibly acceptable. Let's just see how bad this becomes?

@bors try @rust-timer queue

bors · 2024-08-27T19:56:53Z

⌛ Trying commit 197c7b3 with merge 0e77a71...

Add some track_caller info to precondition panics r? `@ghost` Thought of this while looking at rust-lang#129642 (comment)

bors · 2024-08-27T21:50:53Z

☀️ Try build successful - checks-actions
Build commit: 0e77a71 (0e77a71199b6b2f7fac064cdf0b55e84d7ccca61)

rust-timer · 2024-08-27T23:24:55Z

Finished benchmarking commit (0e77a71): 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)	0.2%	[0.2%, 0.2%]	1
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-0.3%	[-0.3%, -0.3%]	6
All ❌✅ (primary)	0.2%	[0.2%, 0.2%]	1

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

Results (primary 0.3%, secondary 0.1%)

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)	0.3%	[0.0%, 1.4%]	65
Regressions ❌ (secondary)	0.1%	[0.0%, 0.5%]	35
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	0.3%	[0.0%, 1.4%]	65

Bootstrap: 749.925s -> 751.312s (0.18%)
Artifact size: 338.74 MiB -> 338.81 MiB (0.02%)

Add some track_caller info to precondition panics Currently, when you encounter a precondition check, you'll always get the caller location of the implementation of the precondition checks. But with this PR, you'll be told the location of the invalid call. Which is useful. I thought of this while looking at #129642 (comment). The changes to `tests/ui/const*` happen because the const-eval interpreter skips `#[track_caller]` frames in its backtraces. The perf implications of this are: * Increased debug binary sizes. The caller_location implementation requires that the additional data we want to display here be stored in const allocations, which are deduplicated but not across crates. There is no impact on optimized build sizes. The panic path and the caller location data get optimized out. * The compile time hit to opt-incr-patched bitmaps happens because the patch changes the line number of some function calls with precondition checks, causing us to go from 0 dirty CGUs to 1 dirty CGU. * The other compile time hits are marginal but real, and due to doing a handful of new queries. Adding more useful data isn't completely free.

bors · 2025-05-25T00:13:08Z

⌛ Testing commit e36dc78 with merge 6f12eca...

bors · 2025-05-25T01:06:06Z

💔 Test failed - checks-actions

saethlin · 2025-05-27T04:36:39Z

@bors try

bors · 2025-05-27T04:37:51Z

⌛ Trying commit 6d0e04d with merge 1cd73d5...

Add some track_caller info to precondition panics Currently, when you encounter a precondition check, you'll always get the caller location of the implementation of the precondition checks. But with this PR, you'll be told the location of the invalid call. Which is useful. I thought of this while looking at #129642 (comment). The changes to `tests/ui/const*` happen because the const-eval interpreter skips `#[track_caller]` frames in its backtraces. The perf implications of this are: * Increased debug binary sizes. The caller_location implementation requires that the additional data we want to display here be stored in const allocations, which are deduplicated but not across crates. There is no impact on optimized build sizes. The panic path and the caller location data get optimized out. * The compile time hit to opt-incr-patched bitmaps happens because the patch changes the line number of some function calls with precondition checks, causing us to go from 0 dirty CGUs to 1 dirty CGU. * The other compile time hits are marginal but real, and due to doing a handful of new queries. Adding more useful data isn't completely free. try-job: test-various

bors · 2025-05-27T05:59:31Z

☀️ Try build successful - checks-actions
Build commit: 1cd73d5 (1cd73d518b9babf013044d90e9d757a210a41214)

saethlin · 2025-05-27T13:42:19Z

@bors r=jhpratt

bors · 2025-05-27T13:42:22Z

📌 Commit 6d0e04d has been approved by jhpratt

It is now in the queue for this repository.

bors · 2025-05-27T22:11:56Z

⌛ Testing commit 6d0e04d with merge be42293...

bors · 2025-05-28T01:19:43Z

☀️ Test successful - checks-actions
Approved by: jhpratt
Pushing be42293 to master...

github-actions · 2025-05-28T01:22:19Z

What is this?

This is an experimental post-merge analysis report that shows differences in test outcomes between the merged PR and its parent PR.

Comparing 45f256d (parent) -> be42293 (this PR)

Test differences

Show 2448 test diffs

2448 doctest diffs were found. These are ignored, as they are noisy.

Test dashboard

Run

cargo run --manifest-path src/ci/citool/Cargo.toml -- \
    test-dashboard be422939446d7c5b27ba98debb6b4b8d6a261f1a --output-dir test-dashboard

And then open test-dashboard/index.html in your browser to see an overview of all executed tests.

Job duration changes

aarch64-gnu: 6386.1s -> 8098.2s (26.8%)
aarch64-gnu-debug: 4005.6s -> 4950.7s (23.6%)
dist-aarch64-apple: 6589.2s -> 5537.3s (-16.0%)
dist-apple-various: 8252.1s -> 7289.1s (-11.7%)
x86_64-mingw-2: 7027.3s -> 7622.8s (8.5%)
dist-i686-mingw: 8764.8s -> 8117.7s (-7.4%)
x86_64-msvc-ext1: 7718.6s -> 7204.1s (-6.7%)
x86_64-apple-2: 5063.5s -> 4730.0s (-6.6%)
dist-x86_64-mingw: 7534.3s -> 7967.5s (5.7%)
x86_64-gnu-llvm-20-2: 5942.4s -> 6279.6s (5.7%)

How to interpret the job duration changes?

Job durations can vary a lot, based on the actual runner instance
that executed the job, system noise, invalidated caches, etc. The table above is provided
mostly for t-infra members, for simpler debugging of potential CI slow-downs.

rust-timer · 2025-05-28T06:48:29Z

Finished benchmarking commit (be42293): comparison URL.

Overall result: ❌ regressions - please read the text below

Our benchmarks found a performance regression caused by this PR.
This might be an actual regression, but it can also be just noise.

Next Steps:

If the regression was expected or you think it can be justified,
please write a comment with sufficient written justification, and add
@rustbot label: +perf-regression-triaged to it, to mark the regression as triaged.
If you think that you know of a way to resolve the regression, try to create
a new PR with a fix for the regression.
If you do not understand the regression or you think that it is just noise,
you can ask the @rust-lang/wg-compiler-performance working group for help (members of this group
were already notified of this PR).

@rustbot label: +perf-regression
cc @rust-lang/wg-compiler-performance

Instruction count

This is the most reliable metric that we have; it was used to determine the overall result at the top of this comment. However, even this metric can sometimes exhibit noise.

	mean	range	count
Regressions ❌ (primary)	1.0%	[0.1%, 6.2%]	14
Regressions ❌ (secondary)	0.8%	[0.4%, 1.2%]	4
Improvements ✅ (primary)	-0.5%	[-0.5%, -0.5%]	1
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	0.9%	[-0.5%, 6.2%]	15

Max RSS (memory usage)

Results (primary 1.0%, secondary 3.6%)

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)	2.9%	[0.9%, 6.7%]	6
Regressions ❌ (secondary)	3.6%	[2.2%, 5.1%]	2
Improvements ✅ (primary)	-4.7%	[-5.7%, -3.7%]	2
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	1.0%	[-5.7%, 6.7%]	8

Cycles

Results (primary 5.7%)

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)	5.7%	[3.3%, 8.2%]	2
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	5.7%	[3.3%, 8.2%]	2

Binary size

Results (primary 0.6%, secondary 1.3%)

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)	0.6%	[0.0%, 2.8%]	66
Regressions ❌ (secondary)	1.3%	[0.1%, 2.9%]	25
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	0.6%	[0.0%, 2.8%]	66

Bootstrap: 779.523s -> 779.377s (-0.02%)
Artifact size: 366.37 MiB -> 366.47 MiB (0.03%)

Add some track_caller info to precondition panics Currently, when you encounter a precondition check, you'll always get the caller location of the implementation of the precondition checks. But with this PR, you'll be told the location of the invalid call. Which is useful. I thought of this while looking at rust-lang#129642 (comment). The changes to `tests/ui/const*` happen because the const-eval interpreter skips `#[track_caller]` frames in its backtraces. The perf implications of this are: * Increased debug binary sizes. The caller_location implementation requires that the additional data we want to display here be stored in const allocations, which are deduplicated but not across crates. There is no impact on optimized build sizes. The panic path and the caller location data get optimized out. * The compile time hit to opt-incr-patched bitmaps happens because the patch changes the line number of some function calls with precondition checks, causing us to go from 0 dirty CGUs to 1 dirty CGU. * The other compile time hits are marginal but real, and due to doing a handful of new queries. Adding more useful data isn't completely free.

eddyb · 2025-07-03T05:42:30Z

library/core/src/ub_checks.rs

-                    ::core::panicking::panic_nounwind(concat!("unsafe precondition(s) violated: ", $message,
+                    let msg = concat!("unsafe precondition(s) violated: ", $message,
                        "\n\nThis indicates a bug in the program. \
-                        This Undefined Behavior check is optional, and cannot be relied on for safety."));
+                        This Undefined Behavior check is optional, and cannot be relied on for safety.");
+                    ::core::panicking::panic_nounwind_fmt(::core::fmt::Arguments::new_const(&[msg]), false);


This might be a slight (unoptimized?) codegen regression, for things as simple as (godbolt):

#[unsafe(no_mangle)] pub fn r_next(mut r: std::ops::Range<usize>) -> Option<usize> { r.next() }

Before:

@alloc_3e1ebac14318b612ab4efabc52799932 = private unnamed_addr constant [186 x i8] c"unsafe precondition(s) violated: usize::unchecked_add cannot overflow\0A\0AThis indicates a bug in the program. This Undefined Behavior check is optional, and cannot be relied on for safety.", align 1 ; <usize>::unchecked_add::precondition_check define internal void @_RNvNvMs9_NtCs1p5UDGgVI4d_4core3numj13unchecked_add18precondition_checkCsb8r2Xq4BFYL_7example(i64 %lhs, i64 %rhs) unnamed_addr #0 { ; … (elided) … bb1: ; preds = %start ; call core::panicking::panic_nounwind call void @_ZN4core9panicking14panic_nounwind17h0c59dc9f7f043eadE(ptr align 1 @alloc_3e1ebac14318b612ab4efabc52799932, i64 186) #6

After:

@alloc_3e1ebac14318b612ab4efabc52799932 = private unnamed_addr constant [186 x i8] c"unsafe precondition(s) violated: usize::unchecked_add cannot overflow\0A\0AThis indicates a bug in the program. This Undefined Behavior check is optional, and cannot be relied on for safety.", align 1 ; … (elided) … ; <usize>::unchecked_add::precondition_check ; Function Attrs: inlinehint nounwind nonlazybind uwtable define internal void @_RNvNvMs9_NtCs25W7cjsTkyH_4core3numj13unchecked_add18precondition_checkCsgRHkggJuglA_7example(i64 %lhs, i64 %rhs, ptr align 8 %0) unnamed_addr #0 { ; … (elided) … bb1: ; preds = %start %2 = getelementptr inbounds nuw { ptr, i64 }, ptr %_6, i64 0 store ptr @alloc_3e1ebac14318b612ab4efabc52799932, ptr %2, align 8 ; … (elided) … ; call core::panicking::panic_nounwind_fmt call void @_ZN4core9panicking18panic_nounwind_fmt17h42ee446b0f35d307E(ptr align 8 %_4, i1 zeroext false, ptr align 8 %0) #6

Notably, even if we ignore the extra IR needed to implement fmt::Arguments::new_const, the fact that &[msg] is pointing to a stack array (as opposed to e.g. const { &[concat!(...)] }) results in the constant string being stored on the stack at runtime.

(and unlike even &[concat!(...)], the let msg blocks even incidental promotion)

That all aside, I am a bit surprised there isn't "just" a copy of panic_nounwind that has #[track_caller], to avoid invoking panic_nounwind_fmt and have this extra codegen (assuming it matters at all, ofc, I just noticed this in passing for unrelated reasons).

@saethlin , pinging so you can see this

(To be clear, this is not that important or any kind of a blocker, I just found it slightly notable that runtime code is being unnecessarily generated, and arguably by mistake, even)

rustbot added S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. T-compiler Relevant to the compiler team, which will review and decide on the PR/issue. T-libs Relevant to the library team, which will review and decide on the PR/issue. labels Aug 27, 2024