enhance(#2721): regenerating merge driver, regen:derived, and a name for the emitted-artifact family (#2730)

* test(#2721): failing-first suite for the gsd-regen driver and CONTEXT.md parity

Tests precede the implementation per the TDD gate. The driver module does not
exist yet, so tests/git-merge-regen-driver.test.cjs fails at require time; the
contributor-standards parity assertions fail against next as it stands today,
where the standards doc names two CONTEXT.md headings that have never existed.

Refs #2721

* feat(#2721): add the gsd-regen merge driver and regen:derived

The golden parity manifests and the two size baselines are pure functions of
the source tree, so their only correct merge is "recompute" -- something git's
ours/theirs interface cannot express. 140 of 143 conflicted-file instances
across the open PR queue are these files.

The driver deliberately does NOT regenerate. Four probes established that at
merge-driver time neither the working tree nor the index reflects the merge:
both hold the ours side, a file added by theirs does not exist yet, and
MERGE_HEAD is unwritten. Git also invokes the driver once per conflicted path
(20 here). A regenerating driver would therefore read the ours-side tree and
emit a plausible-but-wrong hash manifest -- worse than a conflict, because a
conflict is visible. So it accepts %A, runs zero subprocesses, records the
resolved paths, and prints one notice pointing at npm run regen:derived.
Staleness stays caught where it already was, by golden-install-parity in CI.

Every failure path degrades toward today's behaviour (a normal conflict).
install-tree is deliberately excluded per ADR-2719 section 7.

Also folded in, per the no-defer rule: workflow-size.cjs claimed .md files have
no eol=lf in .gitattributes; git check-attr shows eol: lf, set by .gitattributes
line 2 since #1088.

Refs #2721

* docs(#2721): document regen:derived and the gsd-regen merge driver

Adds the how-to a contributor actually reaches for when the generated parity
manifests or size baselines conflict, in both places they would look: the
merge-conflict path in CONTRIBUTING.md and the full guide in TESTING-SUITES.md,
including what the driver deliberately does not do (it does not clear GitHub's
CONFLICTING label, and it does not regenerate mid-merge).

Also scopes the new contributor-standards parity assertion to the doc's own
CONTEXT.md section. Its first run flagged `## Decision`, `## Consequences` and
`## Standards followed`, which the doc attributes to an ADR body and a PR body
rather than to CONTEXT.md -- a doc-wide extractor would have demanded CONTEXT.md
grow headings that do not belong to it.

Refs #2721

* fix(#2721): stop passing %P to the merge driver — shell injection

The isolated adversarial review found, and I independently reproduced, local
arbitrary command execution.

Git does not invoke a merge driver with an argv array. It substitutes %O %A %B
%L %P textually into the configured string and runs the whole thing through a
shell, and $(...) executes inside POSIX double quotes -- so quoting the
placeholder does not neutralise it. %O/%A/%B are git-generated temp names and
%L is an integer, but %P is the file's own path, chosen freely by any
contributor. A branch renaming a covered fixture to
evil$(touch PWNED_SENTINEL).json executed that command on the machine of every
maintainer who merged it, and the merge still reported success.

Fix removes the input rather than filtering it: %P is no longer registered, so
the driver receives no attacker-controlled argument at all. The marker records
a count instead of path names. A metacharacter filter would have been a guess
about shell grammar; passing nothing is a property. Re-ran the identical
exploit against the fixed command: nothing executed, conflict still resolved.

Two regressions guard it -- a platform-independent assertion that the
registered command carries no %P, and a real merge driven by the actual
planInstall output with a $(...) filename.

Also from review: CLI dispatch had no coverage at all (CONTRIBUTING's
"CLI and command routing" matrix), which is why runInstall/runStatus now take
{repoRoot} -- hardcoding REPO_ROOT was what made them untestable. Renamed
planResolution to resolveAndRecord since the plan* prefix promised purity it
did not have. Reconciled the eleven-vs-twelve generator count across
CONTEXT.md, CONTRIBUTING.md and the changeset.

Refs #2721

* test(#2721): scope safe.directory for the check-attr helper

The 66f4d85a run failed 11 assertions, all in the .gitattributes scoping block,
with "fatal: detected dubious ownership in repository at '/work'". The test
container checks the repo out at a path its user does not own, so git refuses
check-attr outright. Everything else passed (27,185).

`check-attr` is a pure read of .gitattributes -- no hooks, no filters -- so the
exemption is scoped to that one invocation. It is deliberately NOT applied to
the driver's own production `git config` calls, which run in the user's own
clone and should keep the protection.

Refs #2721

* test(#2721): delete the stale assertion that the driver command carries %P

The plex2 run on bdfd0856 left exactly two failures, both this test: it still
asserted the pre-fix command string, i.e. the vulnerable behaviour. Deleted
rather than relaxed, per RULESET.TESTS.delete-bad-tests -- its useful half is
already covered, in both directions, by
registeredDriverCommandNeverPassesThePlaceholderForTheFilePath.

Refs #2721

* test(#2721): drive the end-to-end merges from the real planInstall output

The e2e helper hand-rolled its own driver registration, and still carried %P.
That meant the five real-git tests were not exercising the production command
string at all -- planInstall could drift and they would keep passing. They now
register exactly what a contributor gets from npm run setup:merge-driver.

Refs #2721

* chore(#2721): backfill changeset pr number to 2730
This commit is contained in:
Tom Boucher
2026-07-27 19:55:37 -04:00
committed by GitHub
parent e48eb44003
commit a613caaeef
11 changed files with 1491 additions and 8 deletions

View File

@@ -81,6 +81,139 @@ describe('docs/contributor-standards.md', () => {
});
});
/**
* Parity: docs/contributor-standards.md tells contributors which CONTEXT.md sections to
* write into. If it names a heading CONTEXT.md does not have, the instruction is
* unfollowable — and that is not hypothetical: on 2026-07-27 it named `## Domain terms`
* and `## AI Ops Memory`, neither of which has ever existed (#2721). Two surfaces, one
* truth; this asserts they cannot diverge again.
*/
const CONTEXT_MD = path.join(REPO_ROOT, 'CONTEXT.md');
/**
* The body of one `## ` section, exclusive of the next `## `. Split on /\r?\n/ so a CRLF
* checkout parses identically. Shared by every assertion below — two copies of the same
* section parser is exactly the silent divergence RULESET.SHARED-HELPERS-LINT-VS-TEST warns of.
*/
function sectionBody(content, heading) {
const lines = content.split(/\r?\n/);
const start = lines.findIndex((l) => l.trim() === heading);
if (start === -1) return null;
const rest = lines.slice(start + 1);
const end = rest.findIndex((l) => /^##\s/.test(l));
return (end === -1 ? rest : rest.slice(0, end)).join('\n');
}
/**
* Backticked heading references that the standards doc attributes to CONTEXT.md.
*
* Scoped to the doc's own `## CONTEXT.md` section on purpose. The doc also names headings
* belonging to *other* documents — `## Decision` and `## Consequences` describe an ADR
* body, `## Standards followed` describes an issue/PR body. Extracting doc-wide would
* demand CONTEXT.md grow headings that have nothing to do with it.
*
* `<Placeholder>` templates like `### <Module Name>` are skipped: they are shapes to
* follow, not headings to find.
*/
function extractContextHeadingRefs(standardsContent) {
const scope = sectionBody(standardsContent, '## CONTEXT.md');
if (scope === null) return [];
const refs = new Set();
for (const m of scope.matchAll(/`(#{2,6}\s+[^`]+)`/g)) {
const heading = m[1].trim();
if (heading.includes('<')) continue;
refs.add(heading);
}
return [...refs];
}
/** Headings actually present in CONTEXT.md. Anchored /m — CRLF-safe without a `\n` split. */
function actualHeadings(contextContent) {
return new Set([...contextContent.matchAll(/^#{2,6}\s+.*$/gm)].map((m) => m[0].trim()));
}
describe('docs/contributor-standards.md ↔ CONTEXT.md heading parity', () => {
test('everyContextHeadingNamedByTheStandardsDocExistsInContextMd', () => {
const refs = extractContextHeadingRefs(readStandardsDoc());
const actual = actualHeadings(fs.readFileSync(CONTEXT_MD, 'utf-8'));
assert.ok(refs.length > 0, 'the standards doc must name at least one CONTEXT.md heading');
const missing = refs.filter((r) => !actual.has(r));
assert.deepEqual(
missing,
[],
`docs/contributor-standards.md directs contributors to heading(s) that do not exist in ` +
`CONTEXT.md: ${JSON.stringify(missing)}. Fix the standards doc (or add the heading).`
);
});
test('contextMdStillHasTheGlossaryHeadingTheStandardsDocNamed', () => {
const actual = actualHeadings(fs.readFileSync(CONTEXT_MD, 'utf-8'));
assert.ok(
actual.has('## Glossary — Domain modules and seams'),
'the glossary heading is the one the standards doc points Module authors at'
);
});
// Negative space for the extractor itself. The RED run of this suite flagged
// `## Decision`, `## Consequences` and `## Standards followed` — all headings the doc
// attributes to an ADR body or a PR body, not to CONTEXT.md. A doc-wide extractor would
// demand CONTEXT.md sprout headings that do not belong to it.
test('doesNotTreatAdrOrPrBodyHeadingsAsContextMdClaims', () => {
const refs = extractContextHeadingRefs(readStandardsDoc());
for (const foreign of ['## Decision', '## Consequences', '## Standards followed']) {
assert.ok(
!refs.includes(foreign),
`${foreign} describes another document's structure and must not be read as a CONTEXT.md claim`
);
}
});
test('matchesAHeadingReferenceRegardlessOfLineEndingStyle', () => {
const crlf = '## Test rules and lint\r\n\r\n### Emitted Artifact Provenance\r\n';
const found = actualHeadings(crlf);
assert.ok(found.has('## Test rules and lint'), 'a CRLF checkout must not defeat the match');
assert.ok(found.has('### Emitted Artifact Provenance'));
});
});
/**
* The Emitted Artifact Provenance naming deliverable (#2721). Without these, the artifact
* family that half the open PR queue collides on still has no name a contributor can look
* up — which ADR-2719 identifies as a direct cause of the problem.
*/
describe('CONTEXT.md names the emitted-artifact family', () => {
test('emittedAttributionRulesetIsUnderTheTestRulesAndLintSection', () => {
const body = sectionBody(fs.readFileSync(CONTEXT_MD, 'utf-8'), '## Test rules and lint');
assert.ok(body, 'CONTEXT.md must have a `## Test rules and lint` section');
assert.ok(
body.includes('RULESET.EMITTED_ATTRIBUTION='),
'RULESET.EMITTED_ATTRIBUTION must be a sibling of the other test rules, not floating elsewhere'
);
});
test('emittedArtifactProvenanceIsRegisteredInTheGlossary', () => {
const body = sectionBody(
fs.readFileSync(CONTEXT_MD, 'utf-8'),
'## Glossary — Domain modules and seams'
);
assert.ok(body, 'CONTEXT.md must have the glossary section');
assert.ok(
body.includes('### Emitted Artifact Provenance'),
'the emitted-artifact family must be registered in the glossary'
);
});
test('emittedArtifactProvenanceIsAConceptNotAModule', () => {
const headings = actualHeadings(fs.readFileSync(CONTEXT_MD, 'utf-8'));
assert.ok(headings.has('### Emitted Artifact Provenance'));
assert.ok(
!headings.has('### Emitted Artifact Provenance Module'),
'it ships nothing, so it takes no `Module` suffix — follows the `### Resolution Provenance` precedent'
);
});
});
describe('CONTRIBUTING.md links contributor-standards.md', () => {
test('CONTRIBUTING.md contains link to contributor-standards.md', () => {
let contributing;

View File

@@ -0,0 +1,879 @@
/**
* Tests for the `gsd-regen` merge driver (#2721, epic #2719, ADR-2719 Phase 1).
*
* Design + behavior table: .gsd/phase/feat-2721-merge-driver-and-regen-derived/40-design.md
* Test matrix: .gsd/phase/feat-2721-merge-driver-and-regen-derived/50-test-matrix.md
*/
'use strict';
const { describe, test } = require('node:test');
const assert = require('node:assert/strict');
const fs = require('node:fs');
const path = require('node:path');
const cp = require('node:child_process');
const { createTempDir, cleanup } = require('./helpers.cjs');
const REPO_ROOT = path.join(__dirname, '..');
const DRIVER_PATH = path.join(REPO_ROOT, 'scripts', 'git-merge-regen-driver.cjs');
const {
ACTION,
REASON,
GITDIR_SOURCE,
NOTICE_WINDOW_MS,
resolveGitDir,
resolveAndRecord,
planInstall,
} = require(DRIVER_PATH);
// --- helpers ---------------------------------------------------------------
const GIT_TIMEOUT_MS = 30_000;
function git(cwd, args) {
return cp.spawnSync('git', args, {
cwd,
encoding: 'utf8',
timeout: GIT_TIMEOUT_MS,
env: { ...process.env, GIT_CONFIG_NOSYSTEM: '1', HOME: cwd, GIT_TERMINAL_PROMPT: '0' },
});
}
/**
* `git check-attr <attr> -- <path>` in the real repo. Returns the attribute value.
*
* `-c safe.directory=…` is required, not incidental: the test container checks this repo
* out at a path its user does not own, and git then refuses every command with
* "detected dubious ownership in repository at '/work'". `check-attr` is a pure read of
* `.gitattributes` — no hooks, no filters — so scoping the exemption to this one
* invocation is safe. It is deliberately NOT applied to the driver's own production
* `git config` calls, which run in the user's own clone and should keep the protection.
* Forward slashes unconditionally: git wants them in this value on every platform.
*/
function checkAttr(attr, relPath) {
const safeRoot = REPO_ROOT.replace(/\\/g, '/');
const r = cp.spawnSync('git', ['-c', `safe.directory=${safeRoot}`, 'check-attr', attr, '--', relPath], {
cwd: REPO_ROOT,
encoding: 'utf8',
timeout: GIT_TIMEOUT_MS,
});
assert.equal(r.status, 0, `git check-attr failed: ${r.stderr}`);
// Format: "<path>: <attr>: <value>" — the value is everything after the last ": ".
const line = String(r.stdout).trim();
const idx = line.lastIndexOf(': ');
return idx === -1 ? '' : line.slice(idx + 2);
}
/** Replace an fs method with a throwing stub for the duration of `fn`, then restore. */
function withFsFailure(method, fn) {
const original = fs[method];
fs[method] = () => {
throw Object.assign(new Error('injected'), { code: 'EACCES' });
};
try {
return fn();
} finally {
fs[method] = original;
}
}
/** Write an "ours" temp file and return its path — production always receives a real %A. */
function writeOurs(dir, content = 'ours-content\n') {
const p = path.join(dir, '.merge_file_OURS');
fs.writeFileSync(p, content);
return p;
}
/**
* The argv shape git actually supplies under the registered command: [%O, %A, %B, %L].
* `%P` is deliberately NOT registered — see planInstall's comment on shell interpolation.
* `extra` lets one test prove a stray 5th entry (an old registration) is ignored.
*/
function gitArgv(dir, ...extra) {
const o = path.join(dir, '.merge_file_ANC');
const b = path.join(dir, '.merge_file_THEIRS');
fs.writeFileSync(o, '');
fs.writeFileSync(b, 'theirs-content\n');
return [o, writeOurs(dir), b, '7', ...extra];
}
function readMarker(gitDir) {
return JSON.parse(fs.readFileSync(path.join(gitDir, 'gsd-regen-pending.json'), 'utf8'));
}
function seedMarker(gitDir, value) {
fs.mkdirSync(gitDir, { recursive: true });
fs.writeFileSync(
path.join(gitDir, 'gsd-regen-pending.json'),
typeof value === 'string' ? value : JSON.stringify(value),
);
}
const GOLDEN_DIR = path.join(REPO_ROOT, 'tests', 'fixtures', 'golden-install-parity');
const INSTALL_TREE_DIR = path.join(REPO_ROOT, 'tests', 'fixtures', 'install-tree');
function jsonFixturesIn(dir) {
return fs
.readdirSync(dir)
.filter((f) => f.endsWith('.json'))
.sort();
}
// --- .gitattributes scoping (rows 1-10) ------------------------------------
describe('.gitattributes declares the gsd-regen driver for exactly the churning artifacts', () => {
test('goldenParityFixturesDeclareGsdRegenMergeDriver', () => {
assert.equal(
checkAttr('merge', 'tests/fixtures/golden-install-parity/claude.json'),
'gsd-regen',
);
});
test('allNineteenGoldenFixturesDeclareTheDriver', () => {
const fixtures = jsonFixturesIn(GOLDEN_DIR);
assert.ok(fixtures.length > 0, 'expected golden-install-parity fixtures to exist');
for (const f of fixtures) {
assert.equal(
checkAttr('merge', `tests/fixtures/golden-install-parity/${f}`),
'gsd-regen',
`${f} must declare merge=gsd-regen`,
);
}
});
test('workflowSizeBaselineDeclaresTheDriver', () => {
assert.equal(checkAttr('merge', 'tests/workflow-size-baseline.json'), 'gsd-regen');
});
test('agentSizeBaselineDeclaresTheDriver', () => {
assert.equal(checkAttr('merge', 'tests/agent-size-baseline.json'), 'gsd-regen');
});
// NEGATIVE SPACE — ADR-2719 §7 keeps install-tree committed precisely so that
// "the installer stopped shipping X" stays an absolute failure. Capturing it
// with the driver would silently convert that absolute into an auto-resolve.
test('installTreeFixturesAreNotCapturedByTheDriver', () => {
assert.equal(
checkAttr('merge', 'tests/fixtures/install-tree/claude.json'),
'unspecified',
'install-tree must keep normal merge semantics (ADR-2719 §7)',
);
});
test('noInstallTreeFixtureIsCapturedByTheDriver', () => {
const fixtures = jsonFixturesIn(INSTALL_TREE_DIR);
assert.ok(fixtures.length > 0, 'expected install-tree fixtures to exist');
for (const f of fixtures) {
assert.equal(
checkAttr('merge', `tests/fixtures/install-tree/${f}`),
'unspecified',
`${f} must NOT be captured by the driver`,
);
}
});
test('installTreeFixturesAreNotMarkedLinguistGenerated', () => {
assert.equal(
checkAttr('linguist-generated', 'tests/fixtures/install-tree/claude.json'),
'unspecified',
'ADR-2719 §7 keeps install-tree because its diffs are readable',
);
});
test('goldenParityFixturesAreMarkedLinguistGenerated', () => {
assert.equal(
checkAttr('linguist-generated', 'tests/fixtures/golden-install-parity/claude.json'),
'true',
);
});
test('sizeBaselineDeclarationsAreExactPathsNotAGlob', () => {
assert.equal(
checkAttr('merge', 'tests/other-size-baseline.json'),
'unspecified',
'the two baselines are declared by exact path, never by a tests/*-size-baseline.json glob',
);
});
test('driverPatternDoesNotCrossADirectorySeparator', () => {
assert.equal(
checkAttr('merge', 'tests/fixtures/golden-install-parity/sub/nested.json'),
'unspecified',
);
});
});
// --- resolveGitDir (rows 11-19) --------------------------------------------
describe('resolveGitDir', () => {
test('resolvesGitDirWhenDotGitIsADirectory', (t) => {
const dir = createTempDir('gsd-regen-');
t.after(() => cleanup(dir));
fs.mkdirSync(path.join(dir, '.git'));
const r = resolveGitDir(dir);
assert.equal(r.source, GITDIR_SOURCE.DIRECTORY);
assert.equal(r.gitDir, path.join(dir, '.git'));
});
test('resolvesGitDirFromAWorktreePointerFile', (t) => {
const dir = createTempDir('gsd-regen-');
t.after(() => cleanup(dir));
const target = path.join(dir, 'real-gitdir');
fs.mkdirSync(target);
fs.writeFileSync(path.join(dir, '.git'), `gitdir: ${target}\n`);
const r = resolveGitDir(dir);
assert.equal(r.source, GITDIR_SOURCE.GITFILE);
assert.equal(r.gitDir, target);
});
test('resolvesARelativeWorktreePointerAgainstCwd', (t) => {
const dir = createTempDir('gsd-regen-');
t.after(() => cleanup(dir));
fs.mkdirSync(path.join(dir, 'nested'));
fs.writeFileSync(path.join(dir, '.git'), 'gitdir: ./nested\n');
const r = resolveGitDir(dir);
assert.equal(r.source, GITDIR_SOURCE.GITFILE);
assert.equal(r.gitDir, path.resolve(dir, './nested'));
});
test('parsesAWorktreePointerWrittenWithCrlf', (t) => {
const dir = createTempDir('gsd-regen-');
t.after(() => cleanup(dir));
const target = path.join(dir, 'real-gitdir');
fs.mkdirSync(target);
fs.writeFileSync(path.join(dir, '.git'), `gitdir: ${target}\r\n`);
const r = resolveGitDir(dir);
assert.equal(r.source, GITDIR_SOURCE.GITFILE);
assert.equal(r.gitDir, target, 'a trailing CR must not become part of the path');
});
test('parsesAWorktreePointerWithNoSpaceAfterTheColon', (t) => {
const dir = createTempDir('gsd-regen-');
t.after(() => cleanup(dir));
const target = path.join(dir, 'real-gitdir');
fs.mkdirSync(target);
fs.writeFileSync(path.join(dir, '.git'), `gitdir:${target}\n`);
assert.equal(resolveGitDir(dir).gitDir, target);
});
test('treatsAnEmptyPointerFileAsUnresolved', (t) => {
const dir = createTempDir('gsd-regen-');
t.after(() => cleanup(dir));
fs.writeFileSync(path.join(dir, '.git'), '');
const r = resolveGitDir(dir);
assert.equal(r.source, GITDIR_SOURCE.UNRESOLVED);
assert.equal(r.gitDir, null);
});
test('treatsAGarbagePointerFileAsUnresolved', (t) => {
const dir = createTempDir('gsd-regen-');
t.after(() => cleanup(dir));
fs.writeFileSync(path.join(dir, '.git'), 'this is not a gitdir pointer\n');
assert.equal(resolveGitDir(dir).source, GITDIR_SOURCE.UNRESOLVED);
});
test('treatsAMissingDotGitAsUnresolved', (t) => {
const dir = createTempDir('gsd-regen-');
t.after(() => cleanup(dir));
const r = resolveGitDir(dir);
assert.equal(r.source, GITDIR_SOURCE.UNRESOLVED);
assert.equal(r.gitDir, null);
});
test('treatsAnUnreadablePointerFileAsUnresolved', (t) => {
const dir = createTempDir('gsd-regen-');
t.after(() => cleanup(dir));
fs.writeFileSync(path.join(dir, '.git'), 'gitdir: somewhere\n');
const r = withFsFailure('readFileSync', () => resolveGitDir(dir));
assert.equal(r.source, GITDIR_SOURCE.UNRESOLVED, 'must degrade, never throw');
});
});
// --- resolveAndRecord: happy & boundary (rows 20-29) -------------------------
describe('resolveAndRecord — resolution and the notice window', () => {
test('acceptsOursAndNoticesOnTheFirstResolution', (t) => {
const dir = createTempDir('gsd-regen-');
t.after(() => cleanup(dir));
const gitDir = path.join(dir, '.git');
fs.mkdirSync(gitDir);
const r = resolveAndRecord({ argv: gitArgv(dir), gitDir, now: 1_000_000 });
assert.equal(r.action, ACTION.ACCEPT_OURS);
assert.equal(r.reason, REASON.OK_RESOLVED);
assert.equal(r.exitCode, 0);
assert.equal(r.notice, true);
assert.equal(r.pendingCount, 1);
});
test('suppressesTheNoticeForASubsequentPathInTheSameOperation', (t) => {
const dir = createTempDir('gsd-regen-');
t.after(() => cleanup(dir));
const gitDir = path.join(dir, '.git');
seedMarker(gitDir, { startedAt: 999_999, count: 1 });
const r = resolveAndRecord({ argv: gitArgv(dir), gitDir, now: 1_000_000 });
assert.equal(r.notice, false);
assert.equal(r.pendingCount, 2, 'the second conflicted path in the same operation');
});
test('treatsAMarkerJustInsideTheWindowAsTheSameOperation', (t) => {
const dir = createTempDir('gsd-regen-');
t.after(() => cleanup(dir));
const gitDir = path.join(dir, '.git');
const now = 5_000_000;
seedMarker(gitDir, { startedAt: now - (NOTICE_WINDOW_MS - 1), count: 1 });
assert.equal(resolveAndRecord({ argv: gitArgv(dir), gitDir, now }).notice, false);
});
test('treatsAMarkerAtExactlyTheWindowAsTheSameOperation', (t) => {
const dir = createTempDir('gsd-regen-');
t.after(() => cleanup(dir));
const gitDir = path.join(dir, '.git');
const now = 5_000_000;
seedMarker(gitDir, { startedAt: now - NOTICE_WINDOW_MS, count: 1 });
assert.equal(
resolveAndRecord({ argv: gitArgv(dir), gitDir, now }).notice,
false,
'the window is inclusive at the limit',
);
});
test('resetsAndRenoticesForAMarkerJustOutsideTheWindow', (t) => {
const dir = createTempDir('gsd-regen-');
t.after(() => cleanup(dir));
const gitDir = path.join(dir, '.git');
const now = 5_000_000;
seedMarker(gitDir, { startedAt: now - (NOTICE_WINDOW_MS + 1), count: 7 });
const r = resolveAndRecord({ argv: gitArgv(dir), gitDir, now });
assert.equal(r.notice, true, 'a later operation must not inherit the previous silence');
assert.equal(r.pendingCount, 1, 'the stale count must reset, not accumulate');
});
test('noticesOnceAcrossAllTwentyArtifactResolutions', (t) => {
const dir = createTempDir('gsd-regen-');
t.after(() => cleanup(dir));
const gitDir = path.join(dir, '.git');
fs.mkdirSync(gitDir);
const notices = [];
for (let i = 0; i < 20; i += 1) {
const r = resolveAndRecord({ argv: gitArgv(dir), gitDir, now: 2_000_000 + i });
notices.push(r.notice);
}
assert.equal(notices.filter(Boolean).length, 1, 'exactly one notice for the whole operation');
assert.equal(readMarker(gitDir).count, 20, 'this repo conflicts on 20 artifacts');
});
test('countsEveryResolutionInTheOperation', (t) => {
const dir = createTempDir('gsd-regen-');
t.after(() => cleanup(dir));
const gitDir = path.join(dir, '.git');
fs.mkdirSync(gitDir);
resolveAndRecord({ argv: gitArgv(dir), gitDir, now: 3_000_000 });
const r = resolveAndRecord({ argv: gitArgv(dir), gitDir, now: 3_000_001 });
assert.equal(r.pendingCount, 2);
});
test('resolvesAnAddAddConflictWhereTheAncestorIsEmpty', (t) => {
const dir = createTempDir('gsd-regen-');
t.after(() => cleanup(dir));
const gitDir = path.join(dir, '.git');
fs.mkdirSync(gitDir);
const argv = gitArgv(dir);
fs.writeFileSync(argv[0], ''); // %O is a 0-byte file in the add/add case
const r = resolveAndRecord({ argv, gitDir, now: 4_000_000 });
assert.equal(r.action, ACTION.ACCEPT_OURS);
assert.equal(r.exitCode, 0);
});
test('resolvesOnTheMinimumThreeArgumentForm', (t) => {
const dir = createTempDir('gsd-regen-');
t.after(() => cleanup(dir));
const gitDir = path.join(dir, '.git');
fs.mkdirSync(gitDir);
const [o, a, b] = gitArgv(dir);
const r = resolveAndRecord({ argv: [o, a, b], gitDir, now: 4_100_000 });
assert.equal(r.action, ACTION.ACCEPT_OURS);
assert.equal(r.pendingCount, 1);
});
// An old registration (or a hand-edited .git/config) may still pass %P. It must be
// inert data, never consumed — the driver's contract does not depend on it.
test('ignoresAStrayFifthArgumentFromAnOldRegistration', (t) => {
const dir = createTempDir('gsd-regen-');
t.after(() => cleanup(dir));
const gitDir = path.join(dir, '.git');
fs.mkdirSync(gitDir);
const r = resolveAndRecord({
argv: gitArgv(dir, 'tests/workflow-size-baseline.json'),
gitDir,
now: 4_200_000,
});
assert.equal(r.action, ACTION.ACCEPT_OURS);
assert.equal(r.pendingCount, 1);
assert.equal(r.realPath, undefined, 'no path is read from argv at all');
});
});
// --- resolveAndRecord: negative & hostile (rows 30-46) -----------------------
describe('resolveAndRecord — degrades toward a normal conflict, never toward a wrong resolution', () => {
const badArgvCases = [
['declinesRatherThanGuessingWhenArgvIsTooShort', (dir) => gitArgv(dir).slice(0, 2)],
['declinesOnEmptyArgv', () => []],
];
for (const [name, build] of badArgvCases) {
test(name, (t) => {
const dir = createTempDir('gsd-regen-');
t.after(() => cleanup(dir));
const gitDir = path.join(dir, '.git');
fs.mkdirSync(gitDir);
const r = resolveAndRecord({ argv: build(dir), gitDir, now: 6_000_000 });
assert.equal(r.action, ACTION.DECLINE);
assert.equal(r.reason, REASON.FAIL_BAD_ARGV);
assert.equal(r.exitCode, 1, 'a non-zero exit gives git a normal conflict — today’s behavior');
});
}
test('declinesWhenTheOursSideIsMissing', (t) => {
const dir = createTempDir('gsd-regen-');
t.after(() => cleanup(dir));
const gitDir = path.join(dir, '.git');
fs.mkdirSync(gitDir);
const [ancestor, , theirs] = gitArgv(dir);
const neverWritten = path.join(dir, '.merge_file_NEVER_WRITTEN');
const r = resolveAndRecord({
argv: [ancestor, neverWritten, theirs, '7', 'tests/workflow-size-baseline.json'],
gitDir,
now: 6_100_000,
});
assert.equal(r.action, ACTION.DECLINE);
assert.equal(r.reason, REASON.FAIL_OURS_UNREADABLE);
assert.equal(r.exitCode, 1);
});
const blankOursCases = [
['declinesOnAnEmptyOursPath', ''],
['declinesOnAWhitespaceOnlyOursPath', ' '],
];
for (const [name, oursPath] of blankOursCases) {
test(name, (t) => {
const dir = createTempDir('gsd-regen-');
t.after(() => cleanup(dir));
const gitDir = path.join(dir, '.git');
fs.mkdirSync(gitDir);
const [o, , b] = gitArgv(dir);
const r = resolveAndRecord({ argv: [o, oursPath, b], gitDir, now: 6_200_000 });
assert.equal(r.action, ACTION.DECLINE);
assert.equal(r.reason, REASON.FAIL_BAD_ARGV);
});
}
// Valid JSON that is not a usable marker object. Each must be treated as absent
// (reset + notice) rather than throwing or being read as state.
const hostileMarkers = [
['treatsANumericMarkerAsAbsent', '0'],
['treatsAStringMarkerAsAbsent', '"str"'],
['treatsAnArrayMarkerAsAbsent', '[]'],
['treatsANullMarkerAsAbsent', 'null'],
['treatsABooleanMarkerAsAbsent', 'true'],
['treatsAnEmptyMarkerFileAsAbsent', ''],
['treatsACorruptMarkerAsAbsent', '{not json at all'],
['treatsANonNumericStartedAtAsAbsent', '{"startedAt":"yesterday","count":1}'],
['treatsANonFiniteStartedAtAsAbsent', '{"startedAt":1e999,"count":1}'],
['treatsANonNumericCountAsAbsent', '{"startedAt":1,"count":"three"}'],
['treatsANonFiniteCountAsAbsent', '{"startedAt":1,"count":1e999}'],
['treatsANegativeCountAsAbsent', '{"startedAt":1,"count":-5}'],
['treatsAMissingCountAsAbsent', '{"startedAt":1}'],
];
for (const [name, raw] of hostileMarkers) {
test(name, (t) => {
const dir = createTempDir('gsd-regen-');
t.after(() => cleanup(dir));
const gitDir = path.join(dir, '.git');
seedMarker(gitDir, raw);
const r = resolveAndRecord({ argv: gitArgv(dir), gitDir, now: 7_000_000 });
assert.equal(r.action, ACTION.ACCEPT_OURS, 'a bad marker must never block the merge');
assert.equal(r.notice, true);
assert.equal(r.pendingCount, 1, 'an unusable marker resets rather than accumulating');
});
}
test('stillResolvesWhenTheMarkerCannotBeWritten', (t) => {
const dir = createTempDir('gsd-regen-');
t.after(() => cleanup(dir));
const gitDir = path.join(dir, '.git');
fs.mkdirSync(gitDir);
const argv = gitArgv(dir);
const r = withFsFailure('writeFileSync', () =>
resolveAndRecord({ argv, gitDir, now: 8_000_000 }),
);
assert.equal(r.action, ACTION.ACCEPT_OURS);
assert.equal(r.exitCode, 0, 'a diagnostic must never fail a merge');
});
test('resolvesAndNoticesEveryTimeWhenTheGitDirIsUnknown', (t) => {
const dir = createTempDir('gsd-regen-');
t.after(() => cleanup(dir));
const first = resolveAndRecord({ argv: gitArgv(dir), gitDir: null, now: 9_000_000 });
const second = resolveAndRecord({ argv: gitArgv(dir), gitDir: null, now: 9_000_001 });
assert.equal(first.action, ACTION.ACCEPT_OURS);
assert.equal(first.exitCode, 0);
assert.equal(first.notice, true);
assert.equal(second.notice, true, 'without a marker there is nothing to dedupe against');
});
});
// --- planInstall (rows 47-50) ----------------------------------------------
describe('planInstall', () => {
test('plansBothMergeDriverConfigEntries', () => {
const keys = planInstall({ repoRoot: '/repo' }).entries.map((e) => e.key);
assert.deepEqual(keys, ['merge.gsd-regen.name', 'merge.gsd-regen.driver']);
});
test('normalizesTheDriverCommandToForwardSlashesOnEveryPlatform', () => {
const { entries } = planInstall({ repoRoot: 'C:\\Users\\dev\\gsd-core' });
const driver = entries.find((e) => e.key === 'merge.gsd-regen.driver').value;
assert.ok(!driver.includes('\\'), `driver command must contain no backslash: ${driver}`);
});
// `driverCommandPassesEveryPlaceholderGitProvides` lived here and asserted that the
// command carried %P. That was the vulnerable behaviour, so the test has been deleted
// rather than relaxed — a test that asserts the old, now-wrong behaviour is worse than
// no test. Its useful half (the four git-generated placeholders ARE passed) is folded
// into `registeredDriverCommandNeverPassesThePlaceholderForTheFilePath`, which asserts
// both directions in one place.
test('plansIdenticalEntriesOnRepeatedInvocation', () => {
assert.deepEqual(planInstall({ repoRoot: '/repo' }), planInstall({ repoRoot: '/repo' }));
});
});
// --- CLI dispatch (rows 62-73) ---------------------------------------------
// CONTRIBUTING.md → "QA Matrix Requirements" / "CLI and command routing" requires a
// negative-input matrix for any command dispatcher: unknown subcommands, duplicate and
// conflicting flags, plus assertions on exit status and the absence of a stack trace.
describe('CLI dispatch', () => {
/** Run the driver as a real child process — its output never touches this test's stdout. */
function runCli(args, cwd = REPO_ROOT) {
return cp.spawnSync(process.execPath, [DRIVER_PATH, ...args], {
cwd,
encoding: 'utf8',
timeout: GIT_TIMEOUT_MS,
});
}
/**
* Swap process.stdout.write for the duration of `fn`. Same monkeypatch-and-restore-in-
* finally shape as withFsFailure — the run* functions print, and leaking that into the
* runner's stream is how a reporter ends up parsing a driver banner as a test result.
*/
function withStdoutSilenced(fn) {
const original = process.stdout.write;
process.stdout.write = () => true;
try {
return fn();
} finally {
process.stdout.write = original;
}
}
/** A stack frame looks like a line beginning with whitespace + "at ". */
function hasStackTrace(text) {
return /^\s+at\s/m.test(String(text));
}
const rejectedInvocations = [
['rejectsAnUnknownFlag', ['--bogus']],
['rejectsTwoConflictingFlags', ['--install', '--status']],
['rejectsADuplicatedFlag', ['--install', '--install']],
['rejectsAFlagCombinedWithAPositionalArgument', ['--status', 'extra']],
];
for (const [name, args] of rejectedInvocations) {
test(name, () => {
const r = runCli(args);
assert.equal(r.status, 2, `${args.join(' ')} must exit 2, got ${r.status}: ${r.stderr}`);
assert.ok(
!hasStackTrace(r.stderr),
`usage errors must not print a stack trace, got: ${r.stderr}`,
);
});
}
test('driverModeWithNoArgumentsDeclinesRatherThanCrashing', () => {
const r = runCli([]);
assert.equal(r.status, 1, 'too-few-args declines, which git reads as a normal conflict');
assert.ok(!hasStackTrace(r.stderr), `expected no stack trace, got: ${r.stderr}`);
});
test('statusEmitsParseableJsonWithTheDocumentedShape', () => {
const r = runCli(['--status']);
assert.equal(r.status, 0);
const report = JSON.parse(r.stdout);
assert.equal(typeof report.registered, 'boolean');
assert.equal(typeof report.pendingCount, 'number');
});
/** A scratch repo so registration never touches the developer's own .git/config. */
function scratchRepo(t) {
const dir = createTempDir('gsd-regen-cli-');
t.after(() => cleanup(dir));
git(dir, ['init', '-q', '.']);
return dir;
}
test('installRegistersBothConfigEntriesAndStatusReportsIt', (t) => {
const dir = scratchRepo(t);
const { statusOf, runInstall } = require(DRIVER_PATH);
assert.equal(statusOf({ repoRoot: dir }).registered, false, 'precondition: not registered');
assert.equal(withStdoutSilenced(() => runInstall({ repoRoot: dir })), 0);
assert.equal(statusOf({ repoRoot: dir }).registered, true);
const driver = git(dir, ['config', '--get', 'merge.gsd-regen.driver']).stdout.trim();
assert.equal(driver, planInstall({ repoRoot: dir }).entries[1].value);
});
test('installIsIdempotent', (t) => {
const dir = scratchRepo(t);
const { statusOf, runInstall } = require(DRIVER_PATH);
withStdoutSilenced(() => runInstall({ repoRoot: dir }));
const first = git(dir, ['config', '--get-all', 'merge.gsd-regen.driver']).stdout;
assert.equal(withStdoutSilenced(() => runInstall({ repoRoot: dir })), 0);
const second = git(dir, ['config', '--get-all', 'merge.gsd-regen.driver']).stdout;
assert.equal(second, first, 'a second install must not append a duplicate value');
assert.equal(statusOf({ repoRoot: dir }).registered, true);
});
test('uninstallRemovesTheRegistration', (t) => {
const dir = scratchRepo(t);
const { statusOf, runInstall, runUninstall } = require(DRIVER_PATH);
withStdoutSilenced(() => runInstall({ repoRoot: dir }));
assert.equal(withStdoutSilenced(() => runUninstall({ repoRoot: dir })), 0);
assert.equal(statusOf({ repoRoot: dir }).registered, false);
});
test('uninstallOnACleanRepoSucceedsRatherThanFailing', (t) => {
const dir = scratchRepo(t);
const { runUninstall, statusOf } = require(DRIVER_PATH);
assert.equal(withStdoutSilenced(() => runUninstall({ repoRoot: dir })), 0);
assert.equal(statusOf({ repoRoot: dir }).registered, false);
});
/**
* REGRESSION — arbitrary command execution via `%P` (isolated security review, #2721).
*
* Git does not invoke a merge driver with an argv array: it substitutes the placeholders
* textually into the configured string and runs the whole thing through a shell. Quoting
* does not save you — `$(…)` executes inside POSIX double quotes. `%O`/`%A`/`%B` are
* git-generated temp names and `%L` is an integer, but `%P` is the file's own path, which
* any contributor names freely. Registering `"%P"` let a branch that renamed a covered
* fixture to `evil$(touch PWNED).json` run that command on the machine of every maintainer
* who merged it — and the merge still reported success, so nothing looked wrong.
*
* The structural assertion is the real guard: it is platform-independent and fails the
* moment someone re-adds the placeholder.
*/
test('registeredDriverCommandNeverPassesThePlaceholderForTheFilePath', () => {
const { entries } = planInstall({ repoRoot: REPO_ROOT });
const driver = entries.find((e) => e.key === 'merge.gsd-regen.driver').value;
assert.ok(
!driver.includes('%P'),
'git shell-interpolates %P — passing it is arbitrary command execution. Do not re-add it.',
);
for (const safe of ['%O', '%A', '%B', '%L']) {
assert.ok(driver.includes(safe), `${safe} is git-generated and must still be passed`);
}
});
test('aFilenameCarryingShellSubstitutionCannotExecuteDuringAMerge', (t) => {
const dir = createTempDir('gsd-regen-inject-');
t.after(() => cleanup(dir));
git(dir, ['init', '-q', '.']);
git(dir, ['config', 'user.email', 'test@example.com']);
git(dir, ['config', 'user.name', 'test']);
// Register the REAL production command string — a hand-rolled one would not regress.
for (const { key, value } of planInstall({ repoRoot: REPO_ROOT }).entries) {
git(dir, ['config', key, value]);
}
fs.writeFileSync(path.join(dir, '.gitattributes'), 'evil*.json merge=gsd-regen\n');
// Written with fs, so this shell never expands it — the payload is the literal name.
const evil = 'evil$(touch PWNED_SENTINEL).json';
const sentinel = path.join(dir, 'PWNED_SENTINEL');
const write = (v) => fs.writeFileSync(path.join(dir, evil), `{"v":${v}}\n`);
write(0);
git(dir, ['add', '-A']);
git(dir, ['commit', '-qm', 'base']);
const base = git(dir, ['rev-parse', 'HEAD']).stdout.trim();
git(dir, ['checkout', '-qb', 'ours']);
write(1);
git(dir, ['add', '-A']);
git(dir, ['commit', '-qm', 'ours']);
git(dir, ['checkout', '-q', base]);
git(dir, ['checkout', '-qb', 'theirs']);
write(2);
git(dir, ['add', '-A']);
git(dir, ['commit', '-qm', 'theirs']);
git(dir, ['checkout', '-q', 'ours']);
git(dir, ['merge', 'theirs', '-m', 'merge']);
assert.equal(
fs.existsSync(sentinel),
false,
'a filename containing $(...) must never execute — see the regression note above',
);
});
});
// --- real-git end-to-end (rows 51-55) — #2721 AC1 --------------------------
describe('gsd-regen driver under real git operations', () => {
/**
* Build a repo whose `derived.json` is a stand-in for a golden fixture: both
* branches edit a DIFFERENT workflow file and both regenerate the artifact, so
* the artifact conflicts while the sources do not. That is exactly the #2721
* scenario (7 of 7 conflicting PRs collide on the identical artifact set).
*/
function buildScenario(t, { register }) {
const dir = createTempDir('gsd-regen-e2e-');
t.after(() => cleanup(dir));
git(dir, ['init', '-q', '.']);
git(dir, ['config', 'user.email', 'test@example.com']);
git(dir, ['config', 'user.name', 'test']);
if (register) {
// Register the REAL production entries. A hand-rolled command string would let the
// end-to-end tests keep passing while planInstall drifted — and would have kept
// registering the %P form these tests exist to prove is gone.
for (const { key, value } of planInstall({ repoRoot: REPO_ROOT }).entries) {
git(dir, ['config', key, value]);
}
}
fs.mkdirSync(path.join(dir, 'workflows'));
fs.writeFileSync(path.join(dir, '.gitattributes'), 'derived.json merge=gsd-regen\n');
fs.writeFileSync(path.join(dir, 'workflows', 'a.md'), 'A0\n');
fs.writeFileSync(path.join(dir, 'workflows', 'b.md'), 'B0\n');
fs.writeFileSync(path.join(dir, 'derived.json'), '{"a":"A0","b":"B0"}\n');
git(dir, ['add', '-A']);
git(dir, ['commit', '-qm', 'base']);
const base = git(dir, ['rev-parse', 'HEAD']).stdout.trim();
git(dir, ['checkout', '-qb', 'ours']);
fs.writeFileSync(path.join(dir, 'workflows', 'a.md'), 'A1\n');
fs.writeFileSync(path.join(dir, 'derived.json'), '{"a":"A1","b":"B0"}\n');
git(dir, ['add', '-A']);
git(dir, ['commit', '-qm', 'ours edits workflow a']);
git(dir, ['checkout', '-q', base]);
git(dir, ['checkout', '-qb', 'theirs']);
fs.writeFileSync(path.join(dir, 'workflows', 'b.md'), 'B1\n');
fs.writeFileSync(path.join(dir, 'derived.json'), '{"a":"A0","b":"B1"}\n');
git(dir, ['add', '-A']);
git(dir, ['commit', '-qm', 'theirs edits workflow b']);
git(dir, ['checkout', '-q', 'ours']);
return dir;
}
test('twoBranchesEditingDifferentWorkflowsMergeCleanlyWithTheDriver', (t) => {
const dir = buildScenario(t, { register: true });
const merge = git(dir, ['merge', 'theirs', '-m', 'merge']);
assert.equal(merge.status, 0, `merge should succeed: ${merge.stdout}${merge.stderr}`);
assert.equal(git(dir, ['ls-files', '-u']).stdout.trim(), '', 'no unmerged index entries');
assert.ok(
!fs.readFileSync(path.join(dir, 'derived.json'), 'utf8').includes('<<<<<<<'),
'the artifact must carry no conflict markers',
);
assert.equal(
fs.readFileSync(path.join(dir, 'workflows', 'b.md'), 'utf8'),
'B1\n',
'the source side of the merge must still be applied normally',
);
});
// Control: proves the test observes the DRIVER, not a trivially-mergeable artifact.
test('theSameTwoBranchesConflictWithoutTheDriver', (t) => {
const dir = buildScenario(t, { register: false });
const merge = git(dir, ['merge', 'theirs', '-m', 'merge']);
assert.notEqual(merge.status, 0, 'without the driver this scenario must conflict');
assert.notEqual(git(dir, ['ls-files', '-u']).stdout.trim(), '');
});
test('resolvesUnderRebaseNotJustMerge', (t) => {
const dir = buildScenario(t, { register: true });
git(dir, ['checkout', '-q', 'theirs']);
const rebase = git(dir, ['rebase', 'ours']);
assert.equal(rebase.status, 0, `rebase should succeed: ${rebase.stdout}${rebase.stderr}`);
assert.equal(git(dir, ['ls-files', '-u']).stdout.trim(), '');
});
test('leavesAOneSidedChangeToGitsTrivialMerge', (t) => {
const dir = buildScenario(t, { register: true });
git(dir, ['checkout', '-q', '-b', 'sideways', 'ours']);
fs.writeFileSync(path.join(dir, 'workflows', 'c.md'), 'C1\n');
git(dir, ['add', '-A']);
git(dir, ['commit', '-qm', 'unrelated']);
const merge = git(dir, ['merge', 'ours', '-m', 'merge']);
assert.equal(merge.status, 0);
assert.equal(
fs.readFileSync(path.join(dir, 'derived.json'), 'utf8'),
'{"a":"A1","b":"B0"}\n',
'a one-sided change is git’s trivial merge — the driver must not be involved',
);
});
test('resolvedArtifactIsExactlyTheOursSide', (t) => {
const dir = buildScenario(t, { register: true });
const ours = fs.readFileSync(path.join(dir, 'derived.json'), 'utf8');
git(dir, ['merge', 'theirs', '-m', 'merge']);
assert.equal(
fs.readFileSync(path.join(dir, 'derived.json'), 'utf8'),
ours,
'the driver invents nothing — it takes ours verbatim and defers to regen:derived',
);
});
});