benchmark.yml

name: Benchmark

on:
  workflow_dispatch:
    inputs:
      lines:
        description: Source lines in the deterministic synthetic corpus
        required: false
        default: '2000'
        type: string
      runs:
        description: Number of cold/warm measurement pairs
        required: false
        default: '1'
        type: string

env:
  CARGO_TERM_COLOR: always
  RUSTFLAGS: "-D warnings"

jobs:
  scan:
    name: small scan benchmark
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v5
      - uses: dtolnay/rust-toolchain@stable
      # This produces evidence only. Performance targets are deliberately not
      # CI gates, so a changed runner does not block unrelated correctness work.
      - name: Measure a deterministic synthetic corpus
        shell: bash
        run: |
          bench_root="$(mktemp -d)"
          cargo build --release -p codehelion
          cargo run --release -p codehelion-eval --bin codehelion-bench -- \
            corpus --out "$bench_root/corpus" --lines "${{ inputs.lines }}" --seed 42 --clone-percent 5
          for mode in fast structural; do
            cargo run --release -p codehelion-eval --bin codehelion-bench -- \
              scan --corpus "$bench_root/corpus" --binary target/release/codehelion --mode "$mode" --runs "${{ inputs.runs }}"
          done