ci: preserve Xcode compiler discovery with ccache launchers

[mrousavy]

Xcode could not discover libclang.dylib when the iOS builds replaced its compiler with bare clang names. Use Xcode's compiler defaults across the example, Release, Harness, and benchmark builds. The example and Harness workflows use C_COMPILER_LAUNCHER for ccache and enable its Clang explicit-module opt-in, following Apple's compiler launcher guidance.

Remove ccache compiler symlinks and React Native's compiler/linker wrapper injection together, preserving the real-compiler discovery required by ASan in #1417. Use a generic arm64 simulator destination for the build-only example job, remove the matching local example compiler overrides, and remove redundant benchmark libc++ and Swift toolchain search entries.

Validation for 7ed9557c5fd8a366ec9c9412bbf035c3396435bf, rebased onto main at 576a7bc8c70490d580ef458dcd05f5fda88c325b:

• Actionlint, ShellCheck, shell syntax, Xcode project syntax, package JSON parsing, and diff checks pass.
• Hosted Xcode 26.5 Debug builds with and without static frameworks and the Release build pass. Main includes perf: cover deferred worker promises and native-owned buffers #1607's public-header fix for the previous missing ThreadPool.hpp failure. The fresh Debug workflow passed on its second attempt after an initial Bun dependency-tarball extraction failure before compilation.
• iOS Harness Default, ASan and TSan each pass all 552 tests, including the final TSan report check. Android Default and ASan, TypeScript, Nitrogen on Ubuntu/Windows, and init checks also pass.
• Performance builds and measurements and trusted publishing pass for this exact revision.
• The benchmark build logs show 28 dependency scans and 74 module precompilations across the two apps, without compiler/libclang fallback messages. The Harness builds restore cached Pods and DerivedData and successfully use the compiler launcher without discovery or launcher fallback warnings.

These checks verify compiler discovery and compatibility. They do not establish a controlled build-time or runtime speedup: the fresh launcher builds recorded 158 cacheable calls each with zero hits, and runtime measurements remain variable. The performance workflow also uses the head build script for both revisions, so its before/after table cannot isolate the compiler-script change.

CocoaPods may still generate an obsolete Swift toolchain search path; that warning is outside the checked-in project settings.

[vercel]

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

Project	Deployment	Actions	Updated
nitro-docs	Ready	Preview	Sep 7, 2026 6:22pm UTC

[nitro-modules-bot]

Performance Report

⚠️ Advisory: Results do not fail this PR.

iOS

Benchmark	Before	After	Difference
JavaScript addNumbers()	21.1 ns	23.8 ns	🔴 +13.02% slower
C++ bounce native-owned buffer (1 MiB)	3.78 µs	3.46 µs	🟢 -8.44% faster
C++ copy(1 MiB)	65.96 µs	59.15 µs	🟢 -10.32% faster
C++ copy(4 KiB)	993.6 ns	896.5 ns	🟢 -9.77% faster
C++ optional trailing string	160.6 ns	152.4 ns	🟢 -5.1% faster
C++ addNumbers()	94.4 ns	83.1 ns	🟢 -12.02% faster
C++ simpleFunc()	68.2 ns	63.9 ns	🟢 -6.4% faster
C++ deferred worker Promise (includes trigger call)	29.14 µs	24.05 µs	🟢 -17.47% faster
C++ number property get/set	129.2 ns	138.0 ns	🔴 +6.81% slower
C++ nested Car struct	2.38 µs	2.62 µs	🔴 +10.21% slower
Swift copy(1 MiB)	67.62 µs	64.09 µs	🟢 -5.22% faster
Swift copy(4 KiB)	1.05 µs	1.19 µs	🔴 +12.52% slower
Swift synchronousCallback()	234.0 ns	216.7 ns	🟢 -7.42% faster
Swift create()	26.24 µs	27.69 µs	🔴 +5.51% slower
TurboModule addNumbers()	934.6 ns	1.03 µs	🔴 +10.2% slower

All Benchmarks

Benchmark	Before	After	Difference
C++ bounce(1 MiB)	237.1 ns	241.1 ns	🔴 +1.71% slower
C++ bounce(4 KiB)	247.8 ns	237.2 ns	🟢 -4.3% faster
C++ bounce native-owned buffer (4 KiB)	372.0 ns	383.4 ns	🔴 +3.08% slower
C++ large array (1,024)	46.36 µs	46.14 µs	🟢 -0.47% faster
C++ small array (16)	895.0 ns	925.7 ns	🔴 +3.43% slower
C++ synchronousCallback()	188.2 ns	184.2 ns	🟢 -2.15% faster
C++ create()	440.7 ns	433.2 ns	🟢 -1.71% faster
C++ returnExisting()	208.6 ns	213.3 ns	🔴 +2.28% slower
C++ typed map (8 entries)	1.86 µs	1.89 µs	🔴 +1.66% slower
C++ immediatePromise()	918.7 ns	944.3 ns	🔴 +2.79% slower
C++ short ASCII string	141.9 ns	145.3 ns	🔴 +2.41% slower
C++ Unicode string	271.9 ns	272.7 ns	🔴 +0.27% slower
C++ number | string variant	132.9 ns	133.5 ns	🔴 +0.4% slower
Swift bounce(1 MiB)	228.1 ns	222.7 ns	🟢 -2.4% faster
Swift bounce(4 KiB)	236.8 ns	226.1 ns	🟢 -4.53% faster
Swift bounce native-owned buffer (1 MiB)	3.41 µs	3.32 µs	🟢 -2.7% faster
Swift bounce native-owned buffer (4 KiB)	359.1 ns	353.1 ns	🟢 -1.67% faster
Swift large array (1,024)	50.73 µs	50.53 µs	🟢 -0.39% faster
Swift small array (16)	997.9 ns	1.02 µs	🔴 +2.44% slower
Swift returnExisting()	256.9 ns	263.4 ns	🔴 +2.5% slower
Swift typed map (8 entries)	2.47 µs	2.40 µs	🟢 -2.79% faster
Swift optional trailing string	164.2 ns	167.9 ns	🔴 +2.26% slower
Swift addNumbers()	82.4 ns	83.0 ns	🔴 +0.63% slower
Swift simpleFunc()	66.5 ns	66.2 ns	🟢 -0.4% faster
Swift deferred worker Promise (includes trigger call)	25.57 µs	25.89 µs	🔴 +1.23% slower
Swift immediatePromise()	1.19 µs	1.21 µs	🔴 +2.28% slower
Swift number property get/set	144.7 ns	143.7 ns	🟢 -0.69% faster
Swift short ASCII string	169.2 ns	165.4 ns	🟢 -2.26% faster
Swift Unicode string	342.6 ns	345.9 ns	🔴 +0.96% slower
Swift nested Car struct	2.58 µs	2.50 µs	🟢 -3.45% faster
Swift number | string variant	216.9 ns	218.9 ns	🔴 +0.89% slower

Android

Benchmark	Before	After	Difference
JavaScript addNumbers()	21.3 ns	19.6 ns	🟢 -7.82% faster
C++ large array (1,024)	62.96 µs	35.93 µs	🟢 -42.93% faster
C++ deferred worker Promise (includes trigger call)	20.35 µs	22.05 µs	🔴 +8.37% slower
C++ short ASCII string	148.8 ns	134.2 ns	🟢 -9.79% faster
C++ Unicode string	273.5 ns	233.7 ns	🟢 -14.54% faster
C++ number | string variant	130.9 ns	123.2 ns	🟢 -5.88% faster
Kotlin bounce(4 KiB)	995.9 ns	1.05 µs	🔴 +5.64% slower
Kotlin copy(1 MiB)	223.07 µs	243.00 µs	🔴 +8.93% slower
Kotlin typed map (8 entries)	5.81 µs	6.12 µs	🔴 +5.37% slower
Kotlin immediatePromise()	2.26 µs	2.39 µs	🔴 +5.72% slower
Kotlin number property get/set	208.9 ns	220.7 ns	🔴 +5.68% slower

All Benchmarks

Benchmark	Before	After	Difference
C++ bounce(1 MiB)	393.9 ns	393.0 ns	🟢 -0.22% faster
C++ bounce(4 KiB)	397.4 ns	409.7 ns	🔴 +3.08% slower
C++ bounce native-owned buffer (1 MiB)	3.02 µs	3.06 µs	🔴 +1.16% slower
C++ bounce native-owned buffer (4 KiB)	417.2 ns	417.0 ns	🟢 -0.03% faster
C++ copy(1 MiB)	24.44 µs	25.31 µs	🔴 +3.55% slower
C++ copy(4 KiB)	757.2 ns	783.5 ns	🔴 +3.48% slower
C++ small array (16)	728.6 ns	754.1 ns	🔴 +3.5% slower
C++ synchronousCallback()	301.8 ns	295.2 ns	🟢 -2.21% faster
C++ create()	520.8 ns	515.3 ns	🟢 -1.06% faster
C++ returnExisting()	210.3 ns	209.0 ns	🟢 -0.61% faster
C++ typed map (8 entries)	1.68 µs	1.62 µs	🟢 -3.59% faster
C++ optional trailing string	142.0 ns	142.5 ns	🔴 +0.34% slower
C++ addNumbers()	83.2 ns	87.1 ns	🔴 +4.72% slower
C++ simpleFunc()	71.6 ns	70.6 ns	🟢 -1.43% faster
C++ immediatePromise()	757.1 ns	791.0 ns	🔴 +4.47% slower
C++ number property get/set	149.5 ns	153.3 ns	🔴 +2.53% slower
C++ nested Car struct	2.04 µs	2.04 µs	🟢 -0.19% faster
Kotlin bounce(1 MiB)	1.01 µs	974.7 ns	🟢 -3.28% faster
Kotlin bounce native-owned buffer (1 MiB)	3.80 µs	3.68 µs	🟢 -3.01% faster
Kotlin bounce native-owned buffer (4 KiB)	1.09 µs	1.09 µs	🟢 -0.17% faster
Kotlin copy(4 KiB)	4.25 µs	4.20 µs	🟢 -1.14% faster
Kotlin large array (1,024)	39.51 µs	40.12 µs	🔴 +1.56% slower
Kotlin small array (16)	958.4 ns	983.8 ns	🔴 +2.65% slower
Kotlin synchronousCallback()	1.01 µs	1.06 µs	🔴 +4.66% slower
Kotlin create()	1.48 µs	1.52 µs	🔴 +2.38% slower
Kotlin returnExisting()	556.9 ns	559.7 ns	🔴 +0.51% slower
Kotlin optional trailing string	317.8 ns	312.0 ns	🟢 -1.82% faster
Kotlin addNumbers()	114.9 ns	114.7 ns	🟢 -0.21% faster
Kotlin simpleFunc()	90.2 ns	93.7 ns	🔴 +3.93% slower
Kotlin deferred worker Promise (includes trigger call)	32.64 µs	33.25 µs	🔴 +1.87% slower
Kotlin short ASCII string	420.9 ns	400.0 ns	🟢 -4.96% faster
Kotlin Unicode string	675.4 ns	673.6 ns	🟢 -0.27% faster
Kotlin nested Car struct	4.31 µs	4.42 µs	🔴 +2.69% slower
Kotlin number | string variant	371.5 ns	386.9 ns	🔴 +4.14% slower
TurboModule addNumbers()	211.0 ns	213.4 ns	🔴 +1.1% slower

Benchmarking Code Diff 576a7bc8...7ed9557c (view raw output)

Raw measurements: performance-report-1 (JSON artifact). Run 34151299577, attempt 1. Download requires GitHub access.

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