benchmarking-with-hyperfinelisted

# Benchmarking with Hyperfine ## Default workflow 1. Clarify the benchmark question before running anything: - Comparing alternatives? Use the same input, cache story, and environment for all commands. - Checking a regression? Lock the baseline with `--reference` unless parameter sweeps require listing the baseline as a normal command. - Measuring I/O? Decide whether the user needs warm-cache, cold-cache, or both. - Measuring a build? Decide whether clean-build time or incremental-build time matters. 2. Verify the tool: run `hyperfine --version`. If it is older than the flags needed, ask to install or pin a newer release. 3. Start from the closest recipe, then adapt only what the task needs: - Compare commands: `hyperfine --warmup 3 'cmd-a' 'cmd-b'` - Very fast command: `hyperfine -N --warmup 5 --min-runs 200 './bin'` - Clean build: `hyperfine --prepare 'make clean' 'make -j8'` - Numeric sweep: `hyperfine -P threads 1 16 'make -j {threads}'` - List sweep: `hyperfine -L compiler gcc,clang '{compiler} -O2 main.c'` - Named commands: `hyperfine -n old './old' -n new './new'` - Locked baseline: `hyperfine --reference './main' './candidate'` - Export: `hyperfine --export-json r.json --export-markdown r.md 'cmd'` 4. Control noise before trusting results: - Close heavy background processes. - Use `--warmup 3` for normal warm-cache timings. - Use `--prepare` to reset state before every run when measuring cold-cache or clean-state behavior