test-symmetry.yml

name: Test Cross-Language Symmetry

on:
  push:
    branches: [main]
  # No branch filter: see ci.yml.
  pull_request:
  workflow_dispatch:

concurrency:
  group: ${{ github.workflow }}-${{ github.ref }}
  cancel-in-progress: true

jobs:
  symmetry:
    name: MATLAB ↔ Python symmetry tests
    runs-on: ubuntu-latest

    steps:
      # ── Checkout both repos ────────────────────────────────────────────
      - name: Check out NDI-python
        uses: actions/checkout@v4
        with:
          path: NDI-python

      - name: Check out NDI-matlab (matching branch, fall back to main)
        uses: actions/checkout@v4
        with:
          repository: VH-Lab/NDI-matlab
          # Use the same branch name as NDI-python so paired feature
          # branches stay in sync; fall back to main when the matlab
          # branch does not exist.
          ref: ${{ github.head_ref || github.ref_name }}
          path: NDI-matlab
        continue-on-error: true
        id: matlab_checkout_branch

      - name: Check out NDI-matlab (main fallback)
        if: steps.matlab_checkout_branch.outcome == 'failure'
        uses: actions/checkout@v4
        with:
          repository: VH-Lab/NDI-matlab
          ref: main
          path: NDI-matlab

      # ── Runtime setup ──────────────────────────────────────────────────
      - name: Start virtual display server
        run: |
          sudo apt-get install -y xvfb
          Xvfb :99 &
          echo "DISPLAY=:99" >> $GITHUB_ENV

      - name: Set up MATLAB
        uses: matlab-actions/setup-matlab@v2
        with:
          release: latest
          cache: true
          products: |
            Communications_Toolbox
            Statistics_and_Machine_Learning_Toolbox
            Signal_Processing_Toolbox

      - name: Install MatBox
        uses: ehennestad/matbox-actions/install-matbox@v1

      - name: Set up Python
        uses: actions/setup-python@v5
        with:
          python-version: "3.12"

      - name: Install NDI-python
        working-directory: NDI-python
        run: |
          python -m pip install --upgrade pip
          python ndi_install.py --dev --no-validate --verbose

      # ── Download test datasets ────────────────────────────────────────
      - name: Download test datasets
        run: |
          curl -L -o /tmp/69a8705aa9ab25373cdc6563.tgz \
            https://github.com/Waltham-Data-Science/file-passing/raw/refs/heads/main/69a8705aa9ab25373cdc6563.tgz

      # ── Stage 1: MATLAB makeArtifacts ──────────────────────────────────
      # Scope by directory: matbox.installRequirements points at the
      # +ndi/+symmetry folder (which has its own narrow requirements.txt),
      # and TestSuite.fromFolder discovers tests in that directory tree.
      - name: "Stage 1: MATLAB makeArtifacts"
        uses: matlab-actions/run-command@v2
        with:
          command: |
            cd("NDI-matlab");
            addpath(genpath("src"));
            addpath(genpath("tests"));
            addpath(genpath("tools"));

            symDir = fullfile(pwd, "tests", "+ndi", "+symmetry");
            matbox.installRequirements(symDir);

            % Persist the fully-resolved search path (NDI dirs + the matbox
            % dependencies just installed) so Stage 3 can reuse this one
            % installation instead of running installRequirements a second
            % time. A second install re-validates the already-installed repos
            % and would fail any dependency that lacks a LICENSE/README.
            savedPathFile = fullfile(getenv("GITHUB_WORKSPACE"), "ndi_symmetry_matlabpath.txt");
            fid = fopen(savedPathFile, "w");
            assert(fid > 0, "Could not write saved MATLAB path file.");
            fprintf(fid, "%s", path());
            fclose(fid);

            import matlab.unittest.TestSuite
            import matlab.unittest.TestRunner

            suite = TestSuite.fromFolder( ...
                fullfile(symDir, "+makeArtifacts"), "IncludingSubfolders", true);
            fprintf("\n=== MATLAB makeArtifacts: %d test(s) ===\n", numel(suite));
            assert(~isempty(suite), "No MATLAB makeArtifacts tests found.")

            runner = TestRunner.withTextOutput("Verbosity", "Detailed");
            results = runner.run(suite);

            fprintf("\n=== MATLAB makeArtifacts: %d passed, %d failed, %d incomplete out of %d ===\n", ...
                nnz([results.Passed]), nnz([results.Failed]), ...
                nnz([results.Incomplete]), numel(results));
            assert(all(~[results.Failed]), "MATLAB makeArtifacts failed")

            % An Incomplete test verified nothing. In THIS workflow every
            % artifact is produced by an earlier stage, so a filtered
            % assumption means a generator broke -- it does not mean the
            % artifact legitimately does not exist. So gate on it. NDI-matlab's
            % own symmetry job deliberately does NOT: Python artifacts really
            % are absent there, and its incomplete count is honest reporting.
            % See Waltham-Data-Science/NDI-python#77.
            incompleteNames = string({results([results.Incomplete]).Name});
            % "%s" as the format, message as the argument: a test name
            % containing a percent sign must not be read as a format spec.
            assert(isempty(incompleteNames), "%s", ...
                "MATLAB makeArtifacts: " + numel(incompleteNames) + ...
                " test(s) verified nothing because an assumption was" + ...
                " filtered, which in this workflow means a missing" + ...
                " artifact:" + newline + "  " + ...
                strjoin(incompleteNames, newline + "  "))

      # ── Stage 2: Python makeArtifacts + readArtifacts ──────────────────
      - name: "Stage 2: Python symmetry tests (make + read)"
        working-directory: NDI-python
        run: |
          echo "=== Running Python makeArtifacts ==="
          pytest tests/symmetry/make_artifacts/ -v --tb=short

          echo ""
          echo "=== Running Python readArtifacts (reads both MATLAB and Python artifacts) ==="
          # Strict mode: a missing artifact is a failure here, not a skip.
          # Stage 1 produced the matlabArtifacts and the step above produced
          # the pythonArtifacts, so nothing is legitimately absent. Locally
          # the same tests still skip, since most contributors have no MATLAB.
          NDI_SYMMETRY_STRICT=1 pytest tests/symmetry/read_artifacts/ -v --tb=short

      # ── Stage 3: MATLAB readArtifacts ──────────────────────────────────
      # Reuse the installation from Stage 1: restore the saved search path
      # instead of re-running installRequirements. The dependency files
      # persist on the runner between steps; only the path is lost when the
      # fresh MATLAB session starts. This avoids a redundant install and the
      # re-validation that install would perform on the existing repos.
      - name: "Stage 3: MATLAB readArtifacts (reads Python artifacts)"
        uses: matlab-actions/run-command@v2
        with:
          command: |
            cd("NDI-matlab");

            savedPathFile = fullfile(getenv("GITHUB_WORKSPACE"), "ndi_symmetry_matlabpath.txt");
            assert(isfile(savedPathFile), ...
                "Saved MATLAB path from Stage 1 is missing; cannot reuse install.");
            path(fileread(savedPathFile));

            symDir = fullfile(pwd, "tests", "+ndi", "+symmetry");

            import matlab.unittest.TestSuite
            import matlab.unittest.TestRunner

            suite = TestSuite.fromFolder( ...
                fullfile(symDir, "+readArtifacts"), "IncludingSubfolders", true);
            fprintf("\n=== MATLAB readArtifacts: %d test(s) ===\n", numel(suite));
            assert(~isempty(suite), "No MATLAB readArtifacts tests found.")

            runner = TestRunner.withTextOutput("Verbosity", "Detailed");
            results = runner.run(suite);

            fprintf("\n=== MATLAB readArtifacts: %d passed, %d failed, %d incomplete out of %d ===\n", ...
                nnz([results.Passed]), nnz([results.Failed]), ...
                nnz([results.Incomplete]), numel(results));
            assert(all(~[results.Failed]), "MATLAB readArtifacts failed")

            % An Incomplete test verified nothing. In THIS workflow every
            % artifact is produced by an earlier stage, so a filtered
            % assumption means a generator broke -- it does not mean the
            % artifact legitimately does not exist. So gate on it. NDI-matlab's
            % own symmetry job deliberately does NOT: Python artifacts really
            % are absent there, and its incomplete count is honest reporting.
            % See Waltham-Data-Science/NDI-python#77.
            incompleteNames = string({results([results.Incomplete]).Name});
            % "%s" as the format, message as the argument: a test name
            % containing a percent sign must not be read as a format spec.
            assert(isempty(incompleteNames), "%s", ...
                "MATLAB readArtifacts: " + numel(incompleteNames) + ...
                " test(s) verified nothing because an assumption was" + ...
                " filtered, which in this workflow means a missing" + ...
                " artifact:" + newline + "  " + ...
                strjoin(incompleteNames, newline + "  "))