Performance

This page records benchmark results from the v0.12.x performance cycle. Numbers are updated each release cycle. All measurements are from the Go benchmark suite in benchmarks/ — see the v0.12.x performance post for the full story behind each number.

Machine: Apple M4, macOS, Go 1.26.

Package installation

go test -bench=BenchmarkInstall -benchmem -benchtime=3s -count=5 ./benchmarks/

The parallel pool uses min(len(pins), GOMAXPROCS*2) workers. The gain is larger on a slow disk or network where the sequential path serialised all I/O latency.

Package resolution

go test -bench=BenchmarkPMLock -benchmem -benchtime=3s -count=5 ./benchmarks/

The fixture benchmark uses an in-process index (no network). It measures pure resolver overhead. The cold-cache number uses a real PyPI request and reflects network latency; your numbers will differ by connection speed.

Test runner

go test -bench=BenchmarkTestRunner -benchmem -benchtime=3s -count=5 ./benchmarks/

Throughput is the same on a machine with sufficient RAM. The improvement from B-4 is peak goroutine count and GC pressure: the unbounded implementation launched one goroutine per file; the bounded pool holds at GOMAXPROCS*2. On a 200-file suite on a 4-core machine the difference is 200 live interpreter allocations versus 8.

Build cache

go test -bench=BenchmarkBuild -benchmem -benchtime=3s -count=3 ./benchmarks/

On a real project where file collection, minification, and zip writing dominate, the cache hit path reduces second-build time to ~55 µs of hash checks regardless of project size. The remaining ~8 ms in the CLI benchmark is process startup.

Startup

go test -bench=BenchmarkStartup -benchmem -benchtime=3s -count=3 ./benchmarks/

bunpy -c "pass" at ~7.2 ms is inside the 10 ms target and below CPython 3.14’s 14 ms cold start on M-series hardware. The lazy module loading in v0.12.8 skips all 40+ bunpy.* factory calls for scripts that never import bunpy. The remaining startup cost is Go runtime init and goipy.New().

Running the benchmarks yourself

# Generate fixtures once
go run ./benchmarks/fixtures/build_fixtures.go

# Run all benchmarks
go test -bench=. -benchmem -benchtime=3s -count=3 ./benchmarks/

# Run a specific benchmark
go test -bench=BenchmarkStartup -benchmem -benchtime=5s -count=5 ./benchmarks/

# Cross-tool comparison (bunpy vs uv vs CPython)
go test -bench=. -benchmem -benchtime=3s -count=3 ./benchmarks/compare/

The scripts/bench.sh script runs all benchmarks and writes a snapshot to benchmarks/baseline.txt.

Environment variables that affect performance