triage for 2024-05-07.

Author: pnkfelix

triage/2024-05-06.md

@@ -1,8 +1,10 @@
 # 2024-05-07 Triage Log
 
-TODO: Summary
+Largely uneventful week; the most notable shifts were considered false-alarms
+that arose from changes related to cfg-checking (either cargo enabling it, or
+adding cfg's like `rustfmt` to the "well-known cfgs list").
 
-Triage done by **@???**.
+Triage done by **@pnkfelix**.
 Revision range: [c65b2dc9..69f53f5e](https://perf.rust-lang.org/?start=c65b2dc935c27c0c8c3997c6e8d8894718a2cb1a&end=69f53f5e5583381267298ac182eb02c7f1b5c1cd&absolute=false&stat=instructions%3Au)
 
 **Summary**:
@@ -31,6 +33,9 @@ Rollup of 7 pull requests [#124675](https://github.com/rust-lang/rust/pull/12467
 | Improvements ✅ <br /> (secondary) | -    | -            | 0     |
 | All ❌✅ (primary)                 | 0.5% | [0.2%, 1.2%] | 11    |
 
+* all primary regressions are to doc-full scenarios, and the 1.2% is to helloworld.
+* not worth teasing apart a rollup PR.
+* marking as triaged.
 
 Update cargo [#124684](https://github.com/rust-lang/rust/pull/124684) [(Comparison Link)](https://perf.rust-lang.org/compare.html?start=09cd00fea4aecaa6707f122d7e143196b8a12ee2&end=2c4bf249bd47f232de3c1e78ffe69b40c29bfcca&stat=instructions:u)
 
@@ -43,6 +48,12 @@ Update cargo [#124684](https://github.com/rust-lang/rust/pull/124684) [(Comparis
 | All ❌✅ (primary)                 | 2.4% | [0.2%, 19.1%] | 83    |
 
 
+* syn (mostly check builds, but also a debug incr-unchanged and opt incr-unchanged) had regressions ranging from 7.24% all the way up to 19.11%.
+* The most plausible hypothesis is that this is due to an explosion in the number of warnings emitted for this benchmark. (The number of warnings went from ~200 up to 1800, according to Urgau's analysis).
+* This means the code ends up becoming, at least in part, a benchmark of the lint machinery, regardless of whether that is our intent or not.
+* see also [rustc-perf#1819](https://github.com/rust-lang/rustc-perf/issues/1819) "Consider passing -Awarnings (or similar) to avoid false alarms from lint *reporting*"
+* marking as triaged.
+
 Rollup of 3 pull requests [#124784](https://github.com/rust-lang/rust/pull/124784) [(Comparison Link)](https://perf.rust-lang.org/compare.html?start=96f1da82687f499dd3f57006ae71548714532382&end=d287f3e4eeaf680e8fe875f1ec75cca68f357d30&stat=instructions:u)
 
 | (instructions:u)                   | mean | range        | count |
@@ -53,6 +64,10 @@ Rollup of 3 pull requests [#124784](https://github.com/rust-lang/rust/pull/12478
 | Improvements ✅ <br /> (secondary) | -    | -            | 0     |
 | All ❌✅ (primary)                 | 0.3% | [0.2%, 0.4%] | 5     |
 
+* all regressions were to syn, to various incr-unchanged and incr-patched:println scenarios.
+* current hypothesis is that this is due to PR #124742, which adds `rustfmt` to the well-known cfgs list.
+* that hypothesis implies that this is a (mostly-)false alarm, much like #124684.
+* marking as triaged
 
 #### Improvements
 
@@ -66,6 +81,9 @@ Rollup of 10 pull requests [#124646](https://github.com/rust-lang/rust/pull/1246
 | Improvements ✅ <br /> (secondary) | -0.9% | [-1.6%, -0.3%] | 9     |
 | All ❌✅ (primary)                 | -1.0% | [-2.8%, -0.2%] | 24    |
 
+* the bulk of the improvements are to variations of html5ever and serde_derive.
+* skimming over the rollup list, I cannot identify an immediate root cause for improvement
+* but for now will treat it like a happy accident
 
 Some hir cleanups [#124401](https://github.com/rust-lang/rust/pull/124401) [(Comparison Link)](https://perf.rust-lang.org/compare.html?start=d2d24e395a1e4fcee62ca17bf4cbddb1f903af97&end=09cd00fea4aecaa6707f122d7e143196b8a12ee2&stat=instructions:u)
 
@@ -77,6 +95,9 @@ Some hir cleanups [#124401](https://github.com/rust-lang/rust/pull/124401) [(Com
 | Improvements ✅ <br /> (secondary) | -1.1% | [-2.0%, -0.2%] | 2     |
 | All ❌✅ (primary)                 | -0.1% | [-0.2%, -0.1%] | 3     |
 
+* all improvements are to variations of typenum
+* the hir cleanups in question are largely to store `AnonConst` (e.g. for array lengths) in the HIR arena, and then move the ConstArg span over to AnonConst span instead.
+* inspection of typenum didn't show any particular cases that seemed like the would stress `AnonConst`; maybe the benefit comes more from the places where we now pass a span by value instead of passing a pointer to it.
 
 #### Mixed
 
@@ -90,6 +111,10 @@ Account for immutably borrowed locals in MIR copy-prop and GVN [#123602](https:/
 | Improvements ✅ <br /> (secondary) | -0.5% | [-1.0%, -0.3%] | 8     |
 | All ❌✅ (primary)                 | 0.0%  | [-1.1%, 0.9%]  | 16    |
 
+* html5ever opt-full regressed by 0.92%; libc in various incremental scenarios regressed by 0.30% to 0.39%.
+* the libc changes were anticipated in the perf build prior to merge; html5ever opt-full was not predicted there.
+* pnkfelix hypothesizes that this just reflects some extra-work from the compiler attempting to do the copy-propagation and global-value-numbering mir-optimizations on a larger set of immutably-borrowed locals, and is acceptable given the expected benefits.
+* marking as triaged
 
 Rollup of 8 pull requests [#124703](https://github.com/rust-lang/rust/pull/124703) [(Comparison Link)](https://perf.rust-lang.org/compare.html?start=7dd170fccb3be6b1737af5df14dd736b366236c1&end=d7ea27808deb5e10a0f7384e339e4e6165e33398&stat=instructions:u)
 
@@ -101,6 +126,8 @@ Rollup of 8 pull requests [#124703](https://github.com/rust-lang/rust/pull/12470
 | Improvements ✅ <br /> (secondary) | -1.0% | [-1.5%, -0.5%] | 4     |
 | All ❌✅ (primary)                 | 0.5%  | [0.2%, 0.6%]   | 4     |
 
+* image opt-full regressed by 0.63%; html5ever debug-{incr-full,full} by ~0.5%, html5ever opt-incr-unchaged by 0.21%
+* already triaged by Kobzol, who hypothesizes that PR #124700 modified some inlining decisions.
 
 Rollup of 4 pull requests [#124716](https://github.com/rust-lang/rust/pull/124716) [(Comparison Link)](https://perf.rust-lang.org/compare.html?start=1a851da73cdeb02e2c62d301aa6bd98e515a50da&end=d568423a7a4ddb4b49323d96078a22f94df55fbd&stat=instructions:u)
 
@@ -113,6 +140,11 @@ Rollup of 4 pull requests [#124716](https://github.com/rust-lang/rust/pull/12471
 | All ❌✅ (primary)                 | -0.8% | [-0.8%, -0.8%] | 1     |
 
 
+* all regressions are secondary (specifically on unused-warnings benchmark)
+* regression identified by Kobzol as caused by [PR #124584](https://github.com/rust-lang/rust/pull/124584) "Various improvements to entrypoint code"
+* seems like noise to pnkfelix
+* marked as triaged
+
 #### Untriaged Pull Requests
 
 - [#124784 Rollup of 3 pull requests](https://github.com/rust-lang/rust/pull/124784)
@@ -146,7 +178,3 @@ Rollup of 4 pull requests [#124716](https://github.com/rust-lang/rust/pull/12471
 - [#119002 Rollup of 5 pull requests](https://github.com/rust-lang/rust/pull/119002)
 - [#118661 Restore `const PartialEq`](https://github.com/rust-lang/rust/pull/118661)
 
-#### Nags requiring follow up
-
-TODO: Nags
-