voyage/tests/commands/trekexecute-parallel-portability.test.mjs
Kjell Tore Guttormsen ed6699da58
refactor(simplify): one source per rule, shared test helpers, a faster isolation test [skip-docs]
/simplify over a97efb6..HEAD (the plan_version 1.9 / trial-4 delivery).

- coordinator-contract: engine names come from engine-findings ENGINES;
  one markedMajorCodeReview predicate under both lowering checks;
  isEngineFinding everywhere; the unreachable branch (c) of
  illegalBriefLowering folded into (e).
- armB workflow: RULE_KEYS = Object.keys(RULE_TIERS); the engine map is a
  literal pinned equal to ENGINES by the workflow test.
- plan-schema: usesFilesGrammar() is the one 1.9 check (validator, sweep,
  testFirstFiles, stepTestsOnlyFromFields); opensEntry built on opensChunk.
- plan-validator: one issue builder for the Files and Test first grammar.
- plan-scope-sweep: the 1.9 and the 6.2.0 reader are one if/else.
- artifact-commit: TRAILER_RE/trailerProblem no longer exported (no
  importer); the order comment says the CLI finds the repo root first.
- tests: the lock drivers in session-id-isolation (d) run side by side per
  mode (~100 s -> ~30 s); a shared artifact-repo helper; severityRank and
  extractFindingsBlock reused; childEnv(env) without a spread; one copy of
  the PLAN_VERSION_CURRENT check.

Plan readers: validator errors/warnings, plannedPaths, fieldPaths,
testFirstFiles and stepTestsOnlyFromFields compared byte for byte against
HEAD over 510 plans / 1692 steps plus 210 synthetic 1.7-1.9 cases: same.
Suite 2384 (2382/0/2).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-12 21:06:47 +02:00

350 lines
16 KiB
JavaScript

// tests/commands/trekexecute-parallel-portability.test.mjs
//
// Executable tests for the two shell blocks /trekexecute Phase 2.6 Step 2a' and
// Phase 2.55 Check 2 tell the agent to run. Both blocks wrecked a real voyage on
// macOS (order 20260831T214411Z-941965142-from-.claude):
//
// Defect 1: `realpath --relative-to` is GNU coreutils. On BSD realpath the
// command substitution fails, PROJECT_REL becomes EMPTY, and `mkdir -p
// "$wt/"` + `cp ... "$wt//"` both SUCCEED — brief.md/plan.md land at the
// worktree root instead of the project relpath. Exit status stays 0; only
// file location tells the truth. Every assertion here checks placement.
// Defect 2: Check 2 ran `git add {plan-path}` unconditionally. When the
// project directory is gitignored (normal — .claude/projects/ is tool-
// managed and local-only) the add fails and the plan never reaches HEAD.
//
// The blocks are EXTRACTED from commands/trekexecute.md and executed, so the
// test binds to what an agent actually copies, not to prose about it. Anchors
// are grep-able strings, never line numbers.
import { test } from 'node:test';
import { strict as assert } from 'node:assert';
import { readFileSync, writeFileSync, mkdirSync, mkdtempSync, rmSync, chmodSync } from 'node:fs';
import { execFileSync, spawnSync } from 'node:child_process';
import { tmpdir } from 'node:os';
import { realpathSync } from 'node:fs';
import { dirname, join } from 'node:path';
import { fileURLToPath } from 'node:url';
import { childEnv } from '../helpers/child-env.mjs';
const HERE = dirname(fileURLToPath(import.meta.url));
const ROOT = join(HERE, '..', '..');
const COMMAND_FILE = join(ROOT, 'commands', 'trekexecute.md');
const COPY_BLOCK_ANCHOR = "**2a'. Copy gitignored project artifacts";
const CHECK2_ANCHOR = '### Check 2 —';
// The pre-fix form, kept ONLY as the negative control for Defect 1.
const LEGACY_GNU_LINE = 'PROJECT_REL="$(realpath --relative-to="$REPO_ROOT" "$PROJECT_SOURCE")"';
// --- helpers -------------------------------------------------------------
/** Extract the first ```bash fence that follows `anchor` in commands/trekexecute.md. */
function extractBashBlock(anchor) {
const text = readFileSync(COMMAND_FILE, 'utf8');
const at = text.indexOf(anchor);
assert.ok(at >= 0, `anchor not found in trekexecute.md: ${anchor}`);
const fenceOpen = text.indexOf('```bash', at);
assert.ok(fenceOpen >= 0, `no bash fence after anchor: ${anchor}`);
const bodyStart = text.indexOf('\n', fenceOpen) + 1;
const fenceClose = text.indexOf('```', bodyStart);
assert.ok(fenceClose > bodyStart, `unterminated bash fence after anchor: ${anchor}`);
return text.slice(bodyStart, fenceClose);
}
/**
* A PATH directory whose `realpath` behaves like BSD realpath: it rejects every
* GNU long option and resolves bare paths correctly. Stubbed, never assumed —
* the machine running the suite may or may not have GNU coreutils.
*/
function bsdRealpathStubDir() {
const dir = mkdtempSync(join(tmpdir(), 'trek-bsdstub-'));
const stub = join(dir, 'realpath');
writeFileSync(stub, [
'#!/bin/sh',
'# BSD realpath stand-in: no GNU long options.',
'for a in "$@"; do',
' case "$a" in',
' --*) echo "realpath: illegal option -- -" >&2; exit 1 ;;',
' esac',
'done',
"exec python3 -c 'import os,sys",
'for p in sys.argv[1:]: print(os.path.realpath(p))',
"' \"$@\"",
'',
].join('\n'));
chmodSync(stub, 0o755);
return dir;
}
/** A PATH directory whose `git check-ignore` dies fatally (128), everything else real. */
function fatalCheckIgnoreGitStubDir() {
const realGit = execFileSync('/usr/bin/env', ['sh', '-c', 'command -v git'], { encoding: 'utf8' }).trim();
assert.ok(realGit, 'git not on PATH — cannot build the check-ignore stub');
const dir = mkdtempSync(join(tmpdir(), 'trek-gitstub-'));
const stub = join(dir, 'git');
writeFileSync(stub, [
'#!/bin/sh',
'# `git check-ignore` fatal (128): NOT an answer about ignore status.',
'for a in "$@"; do',
' if [ "$a" = "check-ignore" ]; then',
' echo "fatal: simulated check-ignore failure" >&2',
' exit 128',
' fi',
'done',
`exec ${realGit} "$@"`,
'',
].join('\n'));
chmodSync(stub, 0o755);
return dir;
}
function git(cwd, ...args) {
return execFileSync('git', args, { cwd, encoding: 'utf8', stdio: ['ignore', 'pipe', 'pipe'] });
}
/**
* A temp repo with a committed file plus a project directory holding brief/plan
* (+ research unless opts.research === false). `gitignoreProject: true` adds the
* .gitignore line that makes the project directory invisible to `git add` —
* and, note, invisible to `git status --porcelain` too, which is exactly why
* this topology sails through Check 1 and only trips on Check 2.
*/
function makeRepo(opts = {}) {
const root = realpathSync(mkdtempSync(join(tmpdir(), 'trek-exec-')));
git(root, 'init', '-q', '-b', 'main');
git(root, 'config', 'user.email', 'test@example.invalid');
git(root, 'config', 'user.name', 'Test');
writeFileSync(join(root, 'README.md'), '# fixture\n');
if (opts.gitignoreProject) writeFileSync(join(root, '.gitignore'), '.claude/projects/\n');
git(root, 'add', 'README.md', ...(opts.gitignoreProject ? ['.gitignore'] : []));
git(root, 'commit', '-qm', 'init');
const projectRel = join('.claude', 'projects', 'demo');
const projectDir = join(root, projectRel);
mkdirSync(projectDir, { recursive: true });
writeFileSync(join(projectDir, 'brief.md'), '# brief\n');
writeFileSync(join(projectDir, 'plan.md'), '# plan\n');
if (opts.research !== false) {
mkdirSync(join(projectDir, 'research'), { recursive: true });
writeFileSync(join(projectDir, 'research', '01-x.md'), '# r\n');
}
const worktreeDir = join(root, '.claude', 'trekplan-sessions', 'demo', 'worktrees');
mkdirSync(join(worktreeDir, 'session-1'), { recursive: true });
mkdirSync(join(worktreeDir, 'session-2'), { recursive: true });
return { root, projectRel, projectDir, worktreeDir };
}
function runBlock(script, { cwd, env }) {
return spawnSync('bash', ['-c', script], { cwd, env: childEnv(env), encoding: 'utf8' });
}
function copyBlockEnv(repo, pathPrefixDir) {
return {
REPO_ROOT: repo.root,
PROJECT_DIR: repo.projectDir,
WORKTREE_DIR: repo.worktreeDir,
PATH: `${pathPrefixDir}:${process.env.PATH}`,
};
}
function exists(p) {
try { readFileSync(p); return true; } catch { return false; }
}
// --- Defect 1: portable relpath derivation -------------------------------
test("2a' — the BSD realpath stub is known-positive: bare path resolves, --relative-to is rejected", () => {
const stubDir = bsdRealpathStubDir();
try {
const bare = runBlock('realpath "$HOME"', { cwd: ROOT, env: { PATH: `${stubDir}:${process.env.PATH}` } });
assert.equal(bare.status, 0, 'stub must resolve a bare path (proves it can succeed)');
assert.equal(bare.stdout.trim(), realpathSync(process.env.HOME));
const gnu = runBlock('realpath --relative-to=/ "$HOME"', { cwd: ROOT, env: { PATH: `${stubDir}:${process.env.PATH}` } });
assert.notEqual(gnu.status, 0, 'stub must reject the GNU long option');
assert.match(gnu.stderr, /illegal option/, 'stub must fail the way BSD realpath fails');
assert.equal(gnu.stdout.trim(), '', 'no stdout — this is what leaves PROJECT_REL empty');
} finally {
rmSync(stubDir, { recursive: true, force: true });
}
});
test("2a' — NEGATIVE CONTROL: the GNU form drops plan.md at the worktree ROOT under BSD realpath", () => {
const repo = makeRepo();
const stubDir = bsdRealpathStubDir();
try {
const legacy = [
'PROJECT_SOURCE="$(realpath "${PROJECT_DIR}")"',
LEGACY_GNU_LINE,
'for wt in "$WORKTREE_DIR"/session-*; do',
' [ -d "$wt" ] || continue',
' mkdir -p "$wt/$PROJECT_REL"',
' cp "$PROJECT_SOURCE"/brief.md "$wt/$PROJECT_REL/"',
' cp "$PROJECT_SOURCE"/plan.md "$wt/$PROJECT_REL/"',
'done',
].join('\n');
runBlock(legacy, { cwd: repo.root, env: copyBlockEnv(repo, stubDir) });
const wt = join(repo.worktreeDir, 'session-1');
assert.equal(exists(join(wt, repo.projectRel, 'plan.md')), false,
'the broken form must NOT put plan.md at the project relpath');
assert.equal(exists(join(wt, 'plan.md')), true,
'the broken form silently drops plan.md at the worktree root — the measured havari');
} finally {
rmSync(repo.root, { recursive: true, force: true });
rmSync(stubDir, { recursive: true, force: true });
}
});
test("2a' — the shipped block copies brief/plan/research to $wt/$PROJECT_REL without GNU realpath", () => {
const repo = makeRepo();
const stubDir = bsdRealpathStubDir();
try {
const r = runBlock(extractBashBlock(COPY_BLOCK_ANCHOR), { cwd: repo.root, env: copyBlockEnv(repo, stubDir) });
assert.equal(r.status, 0, `block must succeed without GNU realpath. stderr: ${r.stderr}`);
for (const s of ['session-1', 'session-2']) {
const dest = join(repo.worktreeDir, s, repo.projectRel);
assert.equal(exists(join(dest, 'plan.md')), true, `${s}: plan.md must reach $wt/$PROJECT_REL`);
assert.equal(exists(join(dest, 'brief.md')), true, `${s}: brief.md must reach $wt/$PROJECT_REL`);
assert.equal(exists(join(dest, 'research', '01-x.md')), true, `${s}: research/ must reach $wt/$PROJECT_REL`);
assert.equal(exists(join(repo.worktreeDir, s, 'plan.md')), false,
`${s}: nothing may land at the worktree root`);
}
} finally {
rmSync(repo.root, { recursive: true, force: true });
rmSync(stubDir, { recursive: true, force: true });
}
});
test("2a' — a project without research/ is not a failure (block still exits 0)", () => {
const repo = makeRepo({ research: false });
const stubDir = bsdRealpathStubDir();
try {
const r = runBlock(extractBashBlock(COPY_BLOCK_ANCHOR), { cwd: repo.root, env: copyBlockEnv(repo, stubDir) });
assert.equal(r.status, 0, `missing research/ must not fail the wave. stderr: ${r.stderr}`);
assert.equal(exists(join(repo.worktreeDir, 'session-1', repo.projectRel, 'plan.md')), true);
} finally {
rmSync(repo.root, { recursive: true, force: true });
rmSync(stubDir, { recursive: true, force: true });
}
});
test("2a' — an underivable relpath fails LOUDLY instead of dropping files at the worktree root", () => {
const repo = makeRepo();
const stubDir = bsdRealpathStubDir();
const outside = realpathSync(mkdtempSync(join(tmpdir(), 'trek-outside-')));
try {
mkdirSync(join(outside, 'p'), { recursive: true });
writeFileSync(join(outside, 'p', 'brief.md'), 'b');
writeFileSync(join(outside, 'p', 'plan.md'), 'p');
const env = { ...copyBlockEnv(repo, stubDir), PROJECT_DIR: join(outside, 'p') };
const r = runBlock(extractBashBlock(COPY_BLOCK_ANCHOR), { cwd: repo.root, env });
assert.notEqual(r.status, 0, 'a project outside REPO_ROOT must abort the wave');
assert.match(r.stderr, /relpath/i, 'the abort must name the cause');
assert.equal(exists(join(repo.worktreeDir, 'session-1', 'plan.md')), false,
'nothing may be dropped at the worktree root');
} finally {
rmSync(repo.root, { recursive: true, force: true });
rmSync(outside, { recursive: true, force: true });
rmSync(stubDir, { recursive: true, force: true });
}
});
test("2a' — the GNU-only form is gone from the block agents copy", () => {
const block = extractBashBlock(COPY_BLOCK_ANCHOR);
assert.equal(block.includes('--relative-to'), false,
'the copied block must not contain `realpath --relative-to` (GNU-only)');
assert.ok(block.includes('os.path.relpath'), 'the copied block must derive the relpath portably');
});
test("both copied blocks are ASCII-clean (bash 3.2 dies on a multibyte char under set -u)", () => {
const nonAscii = (s) => s.split('\n')
.map((line, i) => [i + 1, line])
.filter(([, line]) => /[^\x00-\x7F]/.test(line));
// Known-positive: the detector must actually fire on a multibyte char.
assert.equal(nonAscii('echo "a — b"').length, 1, 'detector must find an em-dash');
for (const anchor of [COPY_BLOCK_ANCHOR, CHECK2_ANCHOR]) {
assert.deepEqual(nonAscii(extractBashBlock(anchor)), [],
`non-ASCII inside the shell block after ${anchor} (prose outside the fence is fine)`);
}
});
// --- Defect 2: gitignored project directory ------------------------------
// The block carries the `{plan-path}` placeholder the way every other block in
// trekexecute.md does. Substitute it exactly as an agent would — never inject
// PLAN_PATH through the environment: that would supply what the doc must supply
// itself, and a block that never assigns the variable would still pass.
function check2Script(planPath) {
const block = extractBashBlock(CHECK2_ANCHOR);
assert.ok(block.includes('{plan-path}'),
'Check 2 block must carry the {plan-path} placeholder for the agent to substitute');
return block.replace('{plan-path}', planPath);
}
function check2Env(repo, pathPrefixDir) {
const env = { REPO_ROOT: repo.root, PLAN_PATH: '' };
if (pathPrefixDir) env.PATH = `${pathPrefixDir}:${process.env.PATH}`;
return env;
}
test('Check 2 — gitignored project dir: no commit, no failure, and 2a\' still delivers the plan', () => {
const repo = makeRepo({ gitignoreProject: true });
const stubDir = bsdRealpathStubDir();
try {
// Known-positive on the premise: the plan file really is ignored here.
const ci = spawnSync('git', ['check-ignore', '-v', join(repo.projectRel, 'plan.md')],
{ cwd: repo.root, encoding: 'utf8' });
assert.equal(ci.status, 0, 'fixture premise: the plan file must actually be gitignored');
const head = git(repo.root, 'rev-parse', 'HEAD').trim();
const r = runBlock(check2Script(join(repo.projectRel, 'plan.md')),
{ cwd: repo.root, env: check2Env(repo) });
assert.equal(r.status, 0, `Check 2 must tolerate a gitignored plan file. stderr: ${r.stderr}`);
assert.equal(git(repo.root, 'rev-parse', 'HEAD').trim(), head,
'an ignored plan file must NOT be forced into history (origin is a public mirror)');
const copy = runBlock(extractBashBlock(COPY_BLOCK_ANCHOR), { cwd: repo.root, env: copyBlockEnv(repo, stubDir) });
assert.equal(copy.status, 0, `copy step must succeed. stderr: ${copy.stderr}`);
assert.equal(exists(join(repo.worktreeDir, 'session-1', repo.projectRel, 'plan.md')), true,
'the plan must reach the worktree even though git never tracked it');
} finally {
rmSync(repo.root, { recursive: true, force: true });
rmSync(stubDir, { recursive: true, force: true });
}
});
test('Check 2 — KNOWN-POSITIVE: an untracked, NOT-ignored plan file is still added and committed', () => {
const repo = makeRepo();
try {
const planPath = join(repo.projectRel, 'plan.md');
const head = git(repo.root, 'rev-parse', 'HEAD').trim();
const r = runBlock(check2Script(planPath), { cwd: repo.root, env: check2Env(repo) });
assert.equal(r.status, 0, `Check 2 must succeed on a normal untracked plan. stderr: ${r.stderr}`);
assert.notEqual(git(repo.root, 'rev-parse', 'HEAD').trim(), head,
'a trackable plan file must still be committed for worktree visibility');
const ls = spawnSync('git', ['ls-files', '--error-unmatch', planPath], { cwd: repo.root, encoding: 'utf8' });
assert.equal(ls.status, 0, 'the plan file must now be tracked');
} finally {
rmSync(repo.root, { recursive: true, force: true });
}
});
test('Check 2 — a FATAL git check-ignore (128) is not read as "not ignored"', () => {
const repo = makeRepo({ gitignoreProject: true });
const stubDir = fatalCheckIgnoreGitStubDir();
try {
const head = git(repo.root, 'rev-parse', 'HEAD').trim();
const r = runBlock(check2Script(join(repo.projectRel, 'plan.md')),
{ cwd: repo.root, env: check2Env(repo, stubDir) });
assert.notEqual(r.status, 0, 'a fatal check-ignore must stop, not fall through to git add');
assert.match(r.stderr, /check-ignore/, 'the stop must name the failing probe');
assert.equal(git(repo.root, 'rev-parse', 'HEAD').trim(), head, 'no commit may be made on a fatal probe');
} finally {
rmSync(repo.root, { recursive: true, force: true });
rmSync(stubDir, { recursive: true, force: true });
}
});