msd-core/tests/test-gate-watch-mode.test.cjs

// allow-test-rule: source-text-is-the-product #1857
// Workflow .md files — their text IS the deployed contract the orchestrator runs.

/**
 * #1857: a MSD test gate must not hang forever on a watch-mode runner.
 *
 * Three gates were UNBOUNDED or silently-continued and are the core fix. They must:
 *   - route the resolved command through the shared `normalize-test-command`
 *     helper (defeats vitest/jest watch mode), so the paths cannot drift,
 *   - bound execution with `msd_run run-with-timeout` using the
 *     `workflow.test_gate_timeout` budget, and
 *   - surface a timeout (exit 124) with a watch-mode hint — the regression gate
 *     ABORTS, the others surface clearly (never silently ignored).
 *
 * verify-phase's gate (a fourth, already-bounded surface) was deleted with its orphan
 * workflow in #1892; the gates above are the complete live set.
 */

'use strict';

const { test, describe } = require('node:test');
const assert = require('node:assert/strict');
const fs = require('node:fs');
const path = require('node:path');

const ROOT = path.join(__dirname, '..');
// #2932: the regression-gate step was split — steps/regression-gate.md now only
// discovers prior-phase test files and delegates (via "Read and execute") to
// steps/regression-gate-run.md, which carries the actual command resolution,
// normalize-test-command call, timeout bound, and watch-mode abort mechanics.
const REGRESSION_GATE = path.join(ROOT, 'msd-core', 'workflows', 'execute-phase', 'steps', 'regression-gate.md');
const REGRESSION_GATE_RUN = path.join(ROOT, 'msd-core', 'workflows', 'execute-phase', 'steps', 'regression-gate-run.md');
const POST_MERGE_GATE = path.join(ROOT, 'msd-core', 'workflows', 'execute-phase', 'steps', 'post-merge-gate.md');

// #4784: bounds the one-shot sed pipeline the behavioral extraction test runs
// against a fixture line (a pure text filter on ~50 lines — never the 30s-plus
// subprocess classes in timeouts.cjs; named so the two copies in this file
// cannot drift).
const SIMCTL_EXTRACT_TIMEOUT_MS = 10_000;
const AUDIT_FIX = path.join(ROOT, 'msd-core', 'workflows', 'audit-fix.md');
const EXECUTE_PHASE = path.join(ROOT, 'msd-core', 'workflows', 'execute-phase.md');

function read(p) { return fs.readFileSync(p, 'utf-8'); }

// The three gates that were unbounded/silently-continued: full normalize + configured timeout.
// Regression gate's mechanics now live in the delegated regression-gate-run.md (#2932).
const FULL_GATES = [
  ['regression gate', REGRESSION_GATE_RUN],
  ['post-merge gate', POST_MERGE_GATE],
  ['audit-fix gate', AUDIT_FIX],
];

describe('#1857: test gates normalize to one-shot and bound with a timeout', () => {
  for (const [label, file] of FULL_GATES) {
    describe(label, () => {
      test('routes the resolved command through the shared normalize-test-command helper', () => {
        const c = read(file);
        assert.match(c, /normalize-test-command/, `${label} must call the shared normalize-test-command helper`);
      });
      test('bounds execution with a timeout using the workflow.test_gate_timeout budget', () => {
        const c = read(file);
        assert.match(c, /workflow\.test_gate_timeout/, `${label} must read workflow.test_gate_timeout`);
        assert.match(c, /run-with-timeout "\$TEST_GATE_TIMEOUT"/, `${label} must wrap the test command in run-with-timeout with the configured budget`);
      });
      test('surfaces a timeout (exit 124) with a watch-mode hint — never a silent hang', () => {
        const c = read(file);
        assert.match(c, /-eq 124/, `${label} must handle the timeout exit code (124)`);
        assert.match(c, /watch\/dev mode/, `${label} must name watch/dev mode as the likely cause on timeout`);
      });
    });
  }

  // verify-phase's normalize-only block was removed with the orphan workflow (#1892).

  test('the regression gate ABORTS (halts) on a watch-mode timeout', () => {
    const c = read(REGRESSION_GATE);
    assert.match(c, /REGRESSION GATE ABORTED/, 'regression gate must abort (not continue) on timeout');
  });

  test('execute-phase.md delegates the regression gate to the extracted step (size-frozen file stays lean)', () => {
    const c = read(EXECUTE_PHASE);
    assert.match(c, /steps\/regression-gate\.md/, 'execute-phase.md must reference the extracted regression-gate step');
  });

  test('the gates share ONE normalizer — the helper is a single source of truth', () => {
    // The behaviour lives in src/normalize-test-command.cts; every gate invokes it
    // by the same verb name, so a change to watch-defeat logic touches one place.
    for (const file of [REGRESSION_GATE_RUN, POST_MERGE_GATE, AUDIT_FIX]) {
      assert.match(read(file), /msd_run query normalize-test-command/);
    }
  });
});

// #2350: `config-get KEY --default ""` WITHOUT `--raw` prints the JSON-encoded
// empty string (the 2-byte literal `""`) for an unset key, so a `[ -z "$CMD" ]`
// guard sees a non-empty string, SKIPS the whole Makefile/Cargo/go/npm/…
// auto-detection cascade, and runs the literal `""` as a command → exit 127,
// reported as a false build/test failure on any repo with no override and no
// detectable tooling (docs-only / planning-only repos, or any repo before its
// first build file). Every gate that resolves a build/test command this way MUST
// pass `--raw` so an unset key is a genuinely empty bash string the `-z` guard
// catches. This is a defect CLASS — post-merge-gate.md was the reported instance,
// but regression-gate.md and audit-fix.md shared it, so the guard sweeps them (a
// single-file check gave false confidence; the fourth original member,
// verify-phase.md, was deleted as an orphan in #1892). config-get's
// own `--raw` behaviour is covered in config-get-default.test.cjs.
describe('#2350: every gate resolves build/test commands with --raw', () => {
  // Each gate file that reads workflow.build_command / workflow.test_command to
  // build a shell command it then runs. Add new gates here as they appear.
  const GATE_FILES = [
    ['post-merge gate', POST_MERGE_GATE],
    ['regression gate', REGRESSION_GATE_RUN],
    ['audit-fix gate', AUDIT_FIX],
  ];

  for (const [label, file] of GATE_FILES) {
    test(`${label}: no build/test_command config-get line is left without --raw`, () => {
      const lines = read(file)
        .split('\n')
        .filter((l) => /config-get\s+workflow\.(build|test)_command\s+--default\s+""/.test(l));
      // The gate must actually resolve a command this way (guards against the file
      // being renamed/refactored out from under this test without notice).
      assert.ok(lines.length > 0, `${label} (${path.basename(file)}) should resolve a build/test command via config-get`);
      const offending = lines.filter((l) => !/--raw/.test(l));
      assert.deepStrictEqual(
        offending,
        [],
        `${label}: every build/test_command config-get must pass --raw so an unset key is empty, not the literal ""; offending: ${offending.join(' | ')}`,
      );
    });
  }
});

// ─── #4784: the Xcode arms carry -project and a machine-independent destination ──

describe('#4784: Xcode gate construction', () => {
  const gate = read(POST_MERGE_GATE);

  test('#4784: the Xcode gates pass -project to every constructed xcodebuild command', () => {
    // The gate resolves XCODEPROJ (find . -maxdepth 2) and even uses it for
    // `xcodebuild -list -json -project` — but the build/test commands used to
    // drop it, so any project one directory down (App/App.xcodeproj) failed
    // with exit 66: xcodebuild does not search subdirectories.
    const constructed = gate.match(/xcodebuild (?:build|test)[^\n]*'/g) || [];
    assert.ok(constructed.length >= 4, `expected the four Xcode command constructions, got: ${JSON.stringify(constructed)}`);
    for (const cmd of constructed) {
      assert.ok(
        cmd.includes("-project '$XCODEPROJ'"),
        `every constructed xcodebuild command must carry -project '$XCODEPROJ', got: ${cmd}`,
      );
    }
  });

  test('#4784: the Xcode destination derives from the machine\'s available simulators', () => {
    // A concrete device NAME is machine state (the reporter's machine had zero
    // devices matching the hardcoded name). The gate must resolve the first
    // available simulator (id= destination form) and must not pin a name.
    assert.ok(
      gate.includes('simctl list devices available'),
      'the gate must resolve the destination from simctl list devices available',
    );
    assert.ok(gate.includes('id=$XCODE_SIM'), 'the destination must reference the resolved simulator id');
    assert.ok(
      !gate.includes('name=iPhone 16'),
      'the hardcoded simulator name must be gone',
    );
  });

  test('#4784: with no available simulator the gates skip loudly instead of guaranteed failure', () => {
    // Pin BOTH skip echoes: the build arm names workflow.build_command, the
    // test arm names workflow.test_command (a deleting-the-echo edit must go red).
    assert.ok(
      /No available iOS Simulator[^\n]*workflow\.build_command/.test(gate),
      'build-arm skip must name workflow.build_command',
    );
    assert.ok(
      /No available iOS Simulator[^\n]*workflow\.test_command/.test(gate),
      'test-arm skip must name workflow.test_command',
    );
    // set -u robustness: the skip paths must leave the command variables ASSIGNED.
    const skipArms = gate.match(/if \[ -z "\$XCODE_SIM" \]; then[\s\S]*?\n {4}fi/g) || [];
    assert.ok(skipArms.length >= 2, `expected both skip arms, got ${skipArms.length}`);
    for (const arm of skipArms) {
      assert.ok(
        /(?:BUILD|TEST)_CMD=""/.test(arm),
        'each skip arm must assign its command variable (a set -u agent shell must not abort on an unbound variable)',
      );
    }
  });

  test('#4784: the gate\'s simulator-extraction pipeline yields a UDID from a real-format simctl line', () => {
    // Adversarial-review Finding 1 (behavioral, would have been red): the first
    // cut anchored the sed at end-of-line, but real simctl lines end with a
    // (Shutdown)/(Booted) state suffix (often + trailing space) — the anchored
    // form matched NOTHING and the gates skipped on every machine. Extract the
    // gate's OWN pipeline and execute it against a real-format fixture line.
    const pipeMatch = gate.match(/simctl list devices available 2[^\n]*?(sed -n '[^']+')[^\n]*?head -1/);
    assert.ok(pipeMatch, 'gate must contain the simctl→sed→head extraction pipeline');
    const realFormatLine = '    iPhone 17 Pro (94B8198C-8CF1-4860-994A-5669D3388BE8) (Shutdown) ';
    const { execFileSync } = require('node:child_process');
    const udid = execFileSync('sh', ['-c', `printf '%s\\n' "$1" | ${pipeMatch[1]} | head -1`, 'sh', realFormatLine], {
      encoding: 'utf-8',
      timeout: SIMCTL_EXTRACT_TIMEOUT_MS,
    }).trim();
    assert.match(udid, /^[A-F0-9-]{8,}$/, `the pipeline must extract the UDID from a real-format line, got: ${JSON.stringify(udid)}`);
  });

  test('#4784: the test-gate timeout guidance names the Xcode diagnostics collector', () => {
    // A real iOS suite measured 854s wall clock, ~600s of it simctl diagnose
    // collecting a sysdiagnose AFTER the suite passed — the default timeout
    // returns 124 on a passing suite. The gate text must surface the flag.
    assert.ok(
      gate.includes('-collect-test-diagnostics never'),
      'the gate must mention -collect-test-diagnostics never for Xcode suites',
    );
  });
});