voyage/tests/lib/retterunde.test.mjs

// tests/lib/retterunde.test.mjs
// The fix round before 6.3.0 (order 10), the points that are behaviour, not
// prose. Each test was red before its fix.
//   1  A step whose `Files:` are all tests can never be proven red: the
//      behaviour it tests is already at its start point (voyage-lekerepo
//      trial 3, 2026-10-09, finding 1). (a) the validator warns
//      PLAN_STEP_TESTS_ONLY; (c) red-proof says `tests_only` on its
//      `not-red-at-base`, so the executor sends the step up to the plan instead
//      of into fix rounds that cannot succeed. One reading: stepTestsOnly.
//   3  The example plan's `On failure:` lines open with a K11 word.
//   5  The criteria runner's commands never see NODE_TEST_CONTEXT: a nested
//      `node --test` would report to the parent and exit 0 on a red test.

import { test } from 'node:test';
import { strict as assert } from 'node:assert';
import { spawnSync } from 'node:child_process';
import { mkdirSync, mkdtempSync, readFileSync, rmSync, writeFileSync } from 'node:fs';
import { tmpdir } from 'node:os';
import { dirname, join } from 'node:path';
import { fileURLToPath } from 'node:url';
import { validatePlan, validatePlanContent } from '../../lib/validators/plan-validator.mjs';
import { parseStepFields, sliceSteps, stepTestsOnly } from '../../lib/parsers/plan-schema.mjs';
import { proveRed } from '../../lib/verification/red-proof.mjs';
import { runCriteria } from '../../lib/verification/criteria-runner.mjs';
import { buildPlan } from '../helpers/plan-v18-builder.mjs';

const HERE = dirname(fileURLToPath(import.meta.url));
const ROOT = join(HERE, '..', '..');
const codes = (list) => list.map((e) => e.code);

// --- 1 (a): the validator -----------------------------------------------------

// Step 2 changes only tests for behaviour step 1 delivered, as in the trial.
const TESTS_ONLY_PLAN = buildPlan({
  steps: [{}, { file: 'tests/e2e.sh', test: 'tests/e2e-guard.test.mjs', endToEnd: 'bash tests/e2e.sh' }],
});

test('1a: a step whose Files are all tests gives PLAN_STEP_TESTS_ONLY, a warning on that step', () => {
  const r = validatePlanContent(TESTS_ONLY_PLAN);
  const w = r.warnings.filter((x) => x.code === 'PLAN_STEP_TESTS_ONLY');
  assert.equal(w.length, 1, codes(r.warnings).join(','));
  assert.match(w[0].message, /Step 2/);
  assert.ok(!codes(r.errors).includes('PLAN_STEP_TESTS_ONLY'), 'a warning, never an error');
  assert.equal(r.valid, true, codes(r.errors).join(','));
});

test('1a: a step with code and its test gives no PLAN_STEP_TESTS_ONLY (as before)', () => {
  const r = validatePlanContent(buildPlan());
  assert.ok(!codes(r.warnings).includes('PLAN_STEP_TESTS_ONLY'), codes(r.warnings).join(','));
});

test('1a: the reading - Files all tests (by path or as the Test first file) or not; no Files = no warning', () => {
  const body = (files, tf = 'tests/x.test.mjs') => [
    '### Step 1: x', '',
    ...(files === null ? [] : [`- **Files:** ${files}`]),
    '- **Test first:**', `  - File: \`${tf}\` (new)`, '  - Verifies: x',
  ].join('\n');
  assert.equal(stepTestsOnly(body('`tests/e2e.sh`, `tests/e2e-guard.sh`')).testsOnly, true);
  assert.equal(stepTestsOnly(body('`src/a.test.ts` (new file), `lib/b.spec.js`, `pkg/c_test.go`')).testsOnly, true);
  assert.equal(stepTestsOnly(body('`__tests__/a.js`, `test/b.js`')).testsOnly, true);
  // The Test first file counts as a test wherever it lives.
  assert.equal(stepTestsOnly(body('`scripts/check.sh`', 'scripts/check.sh')).testsOnly, true);
  assert.equal(stepTestsOnly(body('`src/cli.mjs`, `tests/cli.test.mjs`')).testsOnly, false);
  assert.equal(stepTestsOnly(body('`src/cli.mjs`')).testsOnly, false);
  assert.equal(stepTestsOnly(body(null)).testsOnly, false);
  assert.deepEqual(stepTestsOnly(body('`tests/e2e.sh`, `tests/e2e-guard.sh`')).files, ['tests/e2e.sh', 'tests/e2e-guard.sh']);
});

test('1a: a plain path beside a backticked one is still read - a step that also changes code is not tests-only', () => {
  const body = (files) => ['### Step 1: x', '', `- **Files:** ${files}`].join('\n');
  const mixed = stepTestsOnly(body('src/cli.mjs (modify), `tests/x.test.mjs` (new)'));
  assert.deepEqual(mixed.files, ['src/cli.mjs', 'tests/x.test.mjs']);
  assert.equal(mixed.testsOnly, false);
  const list = stepTestsOnly(['### Step 1: x', '', '- **Files:**', '  - `tests/x.test.mjs` (new)', '  - src/cli.mjs'].join('\n'));
  assert.equal(list.testsOnly, false, JSON.stringify(list.files));
});

test('1a: a document under spec/ or specs/ is not a test for this reading', () => {
  const body = (files) => ['### Step 1: x', '', `- **Files:** ${files}`].join('\n');
  assert.equal(stepTestsOnly(body('`docs/specs/api.md`, `specs/001-auth/plan.md`')).testsOnly, false);
  assert.equal(stepTestsOnly(body('`spec/cli_spec.rb`')).testsOnly, true);
});

// --- 1 (c): red-proof says tests_only on not-red-at-base ----------------------

function sh(cwd, cmd, args) {
  const r = spawnSync(cmd, args, { cwd, encoding: 'utf8' });
  assert.equal(r.status, 0, `${cmd} ${args.join(' ')}: ${r.stderr}`);
  return r.stdout.trim();
}

const SUM_TEST = [
  "import { test } from 'node:test';",
  "import { strict as assert } from 'node:assert';",
  "import { sum } from '../lib/sum.mjs';",
  "test('sum adds', () => { assert.equal(sum(1, 2), 3); });",
  '',
].join('\n');

/** The behaviour is already at the start point; the step adds only its test. */
function greenAtBase() {
  const dir = mkdtempSync(join(tmpdir(), 'retterunde-red-'));
  sh(dir, 'git', ['init', '-q']);
  sh(dir, 'git', ['config', 'user.email', 't@example.invalid']);
  sh(dir, 'git', ['config', 'user.name', 'test']);
  mkdirSync(join(dir, 'lib'));
  writeFileSync(join(dir, 'lib', 'sum.mjs'), 'export function sum(a, b) { return a + b; }\n');
  sh(dir, 'git', ['add', 'lib/sum.mjs']);
  sh(dir, 'git', ['commit', '-q', '-m', 'base']);
  mkdirSync(join(dir, 'tests'));
  writeFileSync(join(dir, 'tests', 'sum.test.mjs'), SUM_TEST);
  return { dir, base: sh(dir, 'git', ['rev-parse', 'HEAD']) };
}

function planWithStep2Files(dir, files) {
  const p = join(dir, 'plan.local.md');
  writeFileSync(p, buildPlan({ steps: [{}, { file: files, test: 'tests/sum.test.mjs', endToEnd: 'bash tests/e2e.sh' }] }));
  return p;
}

const CMD = 'node --test tests/sum.test.mjs';

test('1c: not-red-at-base on a step whose Files are all tests says tests_only: true (red-only and full)', () => {
  const { dir, base } = greenAtBase();
  try {
    const plan = planWithStep2Files(dir, 'tests/sum.test.mjs');
    for (const redOnly of [true, false]) {
      const r = proveRed({ cwd: dir, base, cmd: CMD, tests: ['tests/sum.test.mjs'], redOnly, plan, step: 2 });
      assert.equal(r.reason, 'not-red-at-base', JSON.stringify(r));
      assert.equal(r.tests_only, true, JSON.stringify(r));
    }
    const cli = spawnSync(process.execPath, [join(ROOT, 'lib', 'verification', 'red-proof.mjs'),
      '--cwd', dir, '--base', base, '--cmd', CMD, '--tests', 'tests/sum.test.mjs', '--plan', plan, '--step', '2', '--json'],
    { encoding: 'utf8' });
    assert.equal(cli.status, 1, cli.stdout + cli.stderr);
    assert.equal(JSON.parse(cli.stdout).tests_only, true);
    // Text mode names the field too (the step helper reads it there).
    const text = spawnSync(process.execPath, [join(ROOT, 'lib', 'verification', 'red-proof.mjs'),
      '--red-only', '--cwd', dir, '--base', base, '--cmd', CMD, '--tests', 'tests/sum.test.mjs', '--plan', plan, '--step', '2'],
    { encoding: 'utf8' });
    assert.match(text.stdout, /tests_only: true/);
    // --step without --plan is a usage error, never silently dropped.
    const lone = spawnSync(process.execPath, [join(ROOT, 'lib', 'verification', 'red-proof.mjs'),
      '--cwd', dir, '--base', base, '--cmd', CMD, '--tests', 'tests/sum.test.mjs', '--step', '2'], { encoding: 'utf8' });
    assert.equal(lone.status, 2, lone.stderr);
  } finally { rmSync(dir, { recursive: true, force: true }); }
});

test('1c: not-red-at-base on a step that also changes code says tests_only: false (fix rounds, as before)', () => {
  const { dir, base } = greenAtBase();
  try {
    const plan = planWithStep2Files(dir, 'lib/sum.mjs');
    const r = proveRed({ cwd: dir, base, cmd: CMD, tests: ['tests/sum.test.mjs'], plan, step: 2 });
    assert.equal(r.reason, 'not-red-at-base');
    assert.equal(r.tests_only, false, JSON.stringify(r));
    // Without --plan nothing new is said: the field is absent.
    const bare = proveRed({ cwd: dir, base, cmd: CMD, tests: ['tests/sum.test.mjs'] });
    assert.equal(bare.reason, 'not-red-at-base');
    assert.equal('tests_only' in bare, false);
  } finally { rmSync(dir, { recursive: true, force: true }); }
});

// --- 3: the example plan reads with the K11 words -----------------------------

test('3: the example plan gives no PLAN_STEP_ON_FAILURE_UNKNOWN', () => {
  const r = validatePlan(join(ROOT, 'examples', '01-add-verbose-flag', 'plan.md'));
  assert.deepEqual(r.warnings.filter((w) => w.code === 'PLAN_STEP_ON_FAILURE_UNKNOWN').map((w) => w.message), []);
});

// --- 5: the criteria runner strips NODE_TEST_CONTEXT ---------------------------

test('5: a nested `node --test` on a failing test is red even when NODE_TEST_CONTEXT is set', () => {
  const dir = mkdtempSync(join(tmpdir(), 'retterunde-ctx-'));
  const saved = process.env.NODE_TEST_CONTEXT;
  try {
    mkdirSync(join(dir, 'tests'));
    writeFileSync(join(dir, 'tests', 'fail.test.mjs'), [
      "import { test } from 'node:test';",
      "import { strict as assert } from 'node:assert';",
      "test('fails', () => { assert.equal(1, 2); });",
      '',
    ].join('\n'));
    process.env.NODE_TEST_CONTEXT = 'child-v8';
    const [r] = runCriteria([{ label: 'C1', text: 'x', command: 'node --test tests/fail.test.mjs' }], { cwd: dir });
    assert.equal(r.status, 'failed', JSON.stringify(r));
    assert.notEqual(r.exitCode, 0);
  } finally {
    if (saved === undefined) delete process.env.NODE_TEST_CONTEXT;
    else process.env.NODE_TEST_CONTEXT = saved;
    rmSync(dir, { recursive: true, force: true });
  }
});

// Operator 2026-10-09: the example's verbose-content test moves into the step
// that wires the verbose output; steps 1-2 (golden capture and the stability
// harness: tests of today's behaviour, never red) keep the warning, and a step
// type for them is a wave 3 question.
test('1: the example plan warns only on its two golden/stability steps; the verbose test sits with the code', () => {
  const r = validatePlan(join(ROOT, 'examples', '01-add-verbose-flag', 'plan.md'));
  const w = r.warnings.filter((x) => x.code === 'PLAN_STEP_TESTS_ONLY').map((x) => x.message.match(/Step (\d+)/)[1]);
  assert.deepEqual(w, ['1', '2']);
  const text = readFileSync(join(ROOT, 'examples', '01-add-verbose-flag', 'plan.md'), 'utf8');
  const step = sliceSteps(text).find((s) => /tests\/cli-verbose-flag\.test\.mjs/.test(parseStepFields(s.body).get('Files')?.text ?? ''));
  assert.ok(step, 'a step lists tests/cli-verbose-flag.test.mjs in Files');
  assert.match(parseStepFields(step.body).get('Files').text, /src\/commands\/login\.mjs/);
  // The run record follows the plan: one entry per step.
  const progress = JSON.parse(readFileSync(join(ROOT, 'examples', '01-add-verbose-flag', 'progress.json'), 'utf8'));
  const n = r.parsed.steps.length;
  assert.equal(progress.total_steps, n);
  assert.equal(progress.current_step, n);
  assert.deepEqual(progress.steps.map((s) => s.n), Array.from({ length: n }, (_, i) => i + 1));
});

// Guard against a reading of the example that counts nothing: it has steps.
test('3: the example plan is read (six or more steps)', () => {
  const text = readFileSync(join(ROOT, 'examples', '01-add-verbose-flag', 'plan.md'), 'utf8');
  assert.ok(validatePlanContent(text).parsed.steps.length >= 6);
});