xcode-build-benchmark

Run the benchmark_builds.py script with your workspace or project, scheme, configuration, and destination. It performs a warm-up, then 3 clean builds and 3 zero-change or incremental builds, saving medians, min, max, and timing summaries to a JSON artifact in .build-benchmark.

Pass --touch-file with a representative Swift source file to benchmark_builds.py. The script touches that file before each of 3 builds, measuring a real edit-rebuild loop; without the flag it measures zero-change builds, which reveal fixed per-build overhead.

Yes. The script detects COMPILATION_CACHE_ENABLE_CACHING = YES via -showBuildSettings and automatically runs a cached-clean phase: it warms the compilation cache, deletes DerivedData before each run, and measures realistic clean builds with a warm cache. Pass --no-cached-clean to skip.

The first clean build after warm-up often runs 20-40% slower due to cold OS-level disk and linker caches. The script mitigates this with a warm-up clean and build cycle; if variance remains high, prefer the median or min and note it in the artifact.

SPM packages with exclude: paths referencing gitignored directories such as __Snapshots__ cause xcodebuild -resolvePackageDependencies to crash inside a worktree. Create the missing directories manually before running any benchmark builds.

Results are noisy under system load, so medians across 3 runs are reported rather than single fastest times. Timing-summary categories are cumulative parallel task times, not wall-clock time, so they indicate compiler workload but do not prove a category blocks the build.