ci: keep completed performance reports visible as PRs advance

[mrousavy]

Completed performance results are currently skipped when the PR's head or base advances before publishing. Keep those results visible until a newer completed report replaces them, with a trusted label linking the exact measured commits and workflow attempt. Closed PRs still skip comments; advanced or superseded reports still skip Bencher history.

Order updates by the source workflow's run number, then attempt. Resolve the existing comment's source through GitHub and require a successful completed attempt from the same PR source repository and branch, including when migrating older comments. An old run finishing late—or being rerun after a newer run—cannot replace newer results. Per-source concurrency with GitHub's documented queue: max serializes the read/update sequence without dropping the previous pending publication.

Builds on merged #1623, which already shows the report in the run summary. This PR changes the trusted publishing policy; it takes effect after reaching main. It does not execute artifact code or change the rendered report's contents.

Validation: 133 performance-tool tests pass, including completion-order permutations, reruns, repeated publication, advanced/closed PRs, legacy comments, forged ordering metadata and fork/branch provenance. TypeScript, ESLint and diff checks pass. Actionlint 1.7.12 passes with only its unsupported concurrency.queue diagnostic excluded; a workflow test checks the exact queue configuration against GitHub's documented syntax.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
nitro-docs	Ready	Preview	Sep 7, 2026 7:02pm UTC

[nitro-modules-bot]

Performance Report

⚠️ Advisory: Results do not fail this PR.

iOS

Benchmark	Before	After	Difference
C++ large array (1,024)	47.60 µs	50.00 µs	🔴 +5.03% slower
C++ synchronousCallback()	214.2 ns	190.1 ns	🟢 -11.24% faster
C++ nested Car struct	2.55 µs	2.39 µs	🟢 -6.26% faster
Swift bounce(1 MiB)	255.5 ns	238.7 ns	🟢 -6.59% faster
Swift copy(1 MiB)	69.28 µs	65.39 µs	🟢 -5.61% faster
Swift synchronousCallback()	217.8 ns	206.9 ns	🟢 -5.01% faster
Swift create()	27.78 µs	30.04 µs	🔴 +8.16% slower
Swift optional trailing string	154.6 ns	171.4 ns	🔴 +10.88% slower
Swift deferred worker Promise (includes trigger call)	23.35 µs	21.27 µs	🟢 -8.89% faster
TurboModule addNumbers()	1.02 µs	922.2 ns	🟢 -9.97% faster

All Benchmarks

Benchmark	Before	After	Difference
JavaScript addNumbers()	24.3 ns	23.4 ns	🟢 -3.8% faster
C++ bounce(1 MiB)	241.0 ns	245.3 ns	🔴 +1.79% slower
C++ bounce(4 KiB)	238.8 ns	241.7 ns	🔴 +1.23% slower
C++ bounce native-owned buffer (1 MiB)	3.70 µs	3.78 µs	🔴 +2.11% slower
C++ bounce native-owned buffer (4 KiB)	391.4 ns	395.9 ns	🔴 +1.14% slower
C++ copy(1 MiB)	75.45 µs	76.01 µs	🔴 +0.75% slower
C++ copy(4 KiB)	1.01 µs	1.01 µs	🟢 -0.2% faster
C++ small array (16)	910.6 ns	933.5 ns	🔴 +2.51% slower
C++ create()	429.0 ns	431.7 ns	🔴 +0.61% slower
C++ returnExisting()	202.9 ns	202.9 ns	🟢 -0% faster
C++ typed map (8 entries)	1.76 µs	1.79 µs	🔴 +1.37% slower
C++ optional trailing string	151.0 ns	151.1 ns	🔴 +0.07% slower
C++ addNumbers()	90.5 ns	86.5 ns	🟢 -4.44% faster
C++ simpleFunc()	71.6 ns	68.3 ns	🟢 -4.61% faster
C++ deferred worker Promise (includes trigger call)	23.35 µs	22.81 µs	🟢 -2.33% faster
C++ immediatePromise()	844.4 ns	840.9 ns	🟢 -0.41% faster
C++ number property get/set	131.6 ns	132.7 ns	🔴 +0.8% slower
C++ short ASCII string	130.7 ns	135.5 ns	🔴 +3.7% slower
C++ Unicode string	287.3 ns	277.9 ns	🟢 -3.25% faster
C++ number | string variant	129.8 ns	135.4 ns	🔴 +4.27% slower
Swift bounce(4 KiB)	237.4 ns	235.4 ns	🟢 -0.85% faster
Swift bounce native-owned buffer (1 MiB)	3.47 µs	3.49 µs	🔴 +0.56% slower
Swift bounce native-owned buffer (4 KiB)	376.6 ns	375.9 ns	🟢 -0.19% faster
Swift copy(4 KiB)	982.7 ns	1.02 µs	🔴 +3.62% slower
Swift large array (1,024)	50.57 µs	49.62 µs	🟢 -1.87% faster
Swift small array (16)	1.05 µs	1.02 µs	🟢 -2.38% faster
Swift returnExisting()	267.1 ns	264.9 ns	🟢 -0.81% faster
Swift typed map (8 entries)	2.61 µs	2.49 µs	🟢 -4.68% faster
Swift addNumbers()	77.7 ns	75.8 ns	🟢 -2.46% faster
Swift simpleFunc()	60.3 ns	63.0 ns	🔴 +4.43% slower
Swift immediatePromise()	1.24 µs	1.22 µs	🟢 -1.87% faster
Swift number property get/set	141.6 ns	138.6 ns	🟢 -2.13% faster
Swift short ASCII string	158.7 ns	162.8 ns	🔴 +2.57% slower
Swift Unicode string	352.2 ns	348.7 ns	🟢 -1% faster
Swift nested Car struct	2.42 µs	2.46 µs	🔴 +1.62% slower
Swift number | string variant	212.3 ns	209.9 ns	🟢 -1.16% faster

Android

Benchmark	Before	After	Difference
JavaScript addNumbers()	22.3 ns	21.1 ns	🟢 -5.59% faster
C++ copy(1 MiB)	24.64 µs	29.66 µs	🔴 +20.37% slower
C++ returnExisting()	220.2 ns	204.6 ns	🟢 -7.09% faster
C++ addNumbers()	82.7 ns	97.6 ns	🔴 +18.08% slower
C++ short ASCII string	139.7 ns	131.4 ns	🟢 -5.96% faster
C++ number | string variant	126.8 ns	143.5 ns	🔴 +13.18% slower
Kotlin copy(1 MiB)	292.67 µs	275.91 µs	🟢 -5.72% faster
Kotlin copy(4 KiB)	4.10 µs	5.01 µs	🔴 +22.22% slower
Kotlin synchronousCallback()	934.0 ns	1.14 µs	🔴 +22.02% slower
Kotlin create()	1.53 µs	1.39 µs	🟢 -8.95% faster
Kotlin simpleFunc()	97.6 ns	90.7 ns	🟢 -7.13% faster
Kotlin number property get/set	210.4 ns	222.1 ns	🔴 +5.56% slower
Kotlin short ASCII string	438.4 ns	415.5 ns	🟢 -5.23% faster
Kotlin Unicode string	663.2 ns	696.7 ns	🔴 +5.05% slower

All Benchmarks

Benchmark	Before	After	Difference
C++ bounce(1 MiB)	426.5 ns	429.5 ns	🔴 +0.71% slower
C++ bounce(4 KiB)	410.3 ns	401.6 ns	🟢 -2.11% faster
C++ bounce native-owned buffer (1 MiB)	3.36 µs	3.26 µs	🟢 -2.99% faster
C++ bounce native-owned buffer (4 KiB)	440.4 ns	447.0 ns	🔴 +1.5% slower
C++ copy(4 KiB)	763.1 ns	736.6 ns	🟢 -3.49% faster
C++ large array (1,024)	37.39 µs	37.74 µs	🔴 +0.94% slower
C++ small array (16)	749.2 ns	782.6 ns	🔴 +4.46% slower
C++ synchronousCallback()	292.1 ns	300.8 ns	🔴 +2.96% slower
C++ create()	569.6 ns	548.8 ns	🟢 -3.66% faster
C++ typed map (8 entries)	1.66 µs	1.58 µs	🟢 -4.6% faster
C++ optional trailing string	138.4 ns	145.2 ns	🔴 +4.92% slower
C++ simpleFunc()	67.9 ns	67.8 ns	🟢 -0.23% faster
C++ deferred worker Promise (includes trigger call)	21.50 µs	22.21 µs	🔴 +3.3% slower
C++ immediatePromise()	810.5 ns	777.6 ns	🟢 -4.06% faster
C++ number property get/set	144.9 ns	145.9 ns	🔴 +0.67% slower
C++ Unicode string	233.5 ns	236.7 ns	🔴 +1.4% slower
C++ nested Car struct	2.08 µs	2.09 µs	🔴 +0.55% slower
Kotlin bounce(1 MiB)	1.03 µs	1.02 µs	🟢 -0.88% faster
Kotlin bounce(4 KiB)	985.4 ns	974.5 ns	🟢 -1.11% faster
Kotlin bounce native-owned buffer (1 MiB)	3.99 µs	4.01 µs	🔴 +0.55% slower
Kotlin bounce native-owned buffer (4 KiB)	1.12 µs	1.12 µs	🔴 +0.06% slower
Kotlin large array (1,024)	39.24 µs	39.64 µs	🔴 +1.04% slower
Kotlin small array (16)	995.2 ns	1.02 µs	🔴 +2.01% slower
Kotlin returnExisting()	527.8 ns	526.2 ns	🟢 -0.29% faster
Kotlin typed map (8 entries)	6.21 µs	5.92 µs	🟢 -4.64% faster
Kotlin optional trailing string	318.6 ns	323.3 ns	🔴 +1.47% slower
Kotlin addNumbers()	109.5 ns	110.3 ns	🔴 +0.7% slower
Kotlin deferred worker Promise (includes trigger call)	30.94 µs	31.16 µs	🔴 +0.72% slower
Kotlin immediatePromise()	2.42 µs	2.33 µs	🟢 -3.92% faster
Kotlin nested Car struct	4.24 µs	4.22 µs	🟢 -0.55% faster
Kotlin number | string variant	385.3 ns	390.7 ns	🔴 +1.41% slower
TurboModule addNumbers()	214.8 ns	224.6 ns	🔴 +4.57% slower

Benchmarking Code Diff 2e815f49...0fd15aa4 (view raw output)

Raw measurements: performance-report-1 (JSON artifact). Run 34153977762, attempt 1. Download requires GitHub access.

Android: measurements, attempt 1, apps, attempt 1.
iOS: measurements, attempt 1, apps, attempt 1.