test.yml

# The test suite, factored out so two callers can share one definition: `ci.yml`
# on every push and PR, and `release.yml` as the gate a tag has to pass before
# any binary is built. A release that shipped without these having run would be
# a release nobody checked.
name: Test

on:
  workflow_call:

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

jobs:
  test:
    name: test (${{ matrix.os }})
    runs-on: ${{ matrix.os }}
    strategy:
      fail-fast: false
      matrix:
        os: [ubuntu-latest, windows-latest, macos-latest]
    steps:
      - uses: actions/checkout@v5

      - uses: dtolnay/rust-toolchain@stable

      - uses: Swatinem/rust-cache@v2
        with:
          key: ${{ matrix.os }}

      - name: Build
        run: cargo build --workspace --all-targets

      - name: Test
        run: cargo test --workspace

      # The corpus is generated, not hand-committed: if the generator and the
      # committed binaries ever disagree, the goldens are testing fiction.
      - name: Corpus is up to date
        run: python3 corpus/generate.py --check

  # CI gate 1 of Phase 15 (`docs/technical-analysis-presentations.md` §5.5): the
  # `.pptx` parts this project writes, validated against the ECMA-376 XSDs
  # vendored in `schemas/`. Ubuntu only, and once: the gate measures the bytes
  # the writer produces, which are the same on all three platforms, and running
  # it three times would only buy three copies of the same answer.
  schema-gate:
    name: pptx schema validation
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v5

      - uses: dtolnay/rust-toolchain@stable

      - uses: Swatinem/rust-cache@v2

      # libxml2's validator. Not a workspace dependency and deliberately not one:
      # it never reaches a shipped binary (`AGENTS.md` §7 rule 4).
      - name: Install xmllint
        run: sudo apt-get update && sudo apt-get install -y libxml2-utils

      - name: Build the CLI
        run: cargo build -p docsai-cli

      - name: Schema gate
        run: python3 corpus/schema_gate.py

  # CI gate 2 of Phase 15: what a corpus deck *looks like*, before and after the
  # round trip. Headless LibreOffice renders each package to PDF, `pdftoppm`
  # rasterises every page, and a low-frequency perceptual diff scores the pairs.
  # Ubuntu only and once, for the same reason as the schema gate: the bytes the
  # writer produces are the same on all three platforms.
  #
  # `--require-renderer` is the point of this job. A missing LibreOffice is a
  # *skip* on a developer's machine — a clean message and exit 3, never a green
  # tick — but here it is a **failure**: this runner installs LibreOffice
  # deliberately, so «no renderer» means the install step broke, not that the
  # gate is inapplicable. A CI job that silently measured nothing is exactly the
  # failure this gate was written against.
  render-gate:
    name: pptx render diff
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v5

      - uses: dtolnay/rust-toolchain@stable

      - uses: Swatinem/rust-cache@v2

      # Neither is a workspace dependency, and neither reaches a shipped binary
      # (`AGENTS.md` §7 rule 4). `--no-install-recommends` keeps the LibreOffice
      # install to the Impress filters and the fonts they need.
      - name: Install LibreOffice and poppler
        run: |
          sudo apt-get update
          sudo apt-get install -y --no-install-recommends \
            libreoffice-impress poppler-utils fonts-liberation

      - name: Build the CLI
        run: cargo build -p docsai-cli

      # The comparator itself, on synthetic pairs whose answer is known. It
      # needs no LibreOffice, so it measures something even on a machine where
      # the renderer is unavailable, and it is what pins the threshold.
      - name: Comparator self-test
        run: python3 corpus/render_gate.py --self-test

      - name: Render gate
        run: python3 corpus/render_gate.py --require-renderer

  lint:
    name: fmt + clippy
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v5

      - uses: dtolnay/rust-toolchain@stable
        with:
          components: rustfmt, clippy

      - uses: Swatinem/rust-cache@v2

      - name: Format
        run: cargo fmt --all -- --check

      - name: Clippy
        run: cargo clippy --workspace --all-targets -- -D warnings

      # Named separately from `cargo test` so a blown token budget reads as a
      # blown token budget in the checks list, not as "some test failed".
      - name: Token budget
        run: cargo test -p docsai-convert --test token_budget

      - name: Docs
        run: cargo doc --workspace --no-deps
        env:
          RUSTDOCFLAGS: "-D warnings"