Skip to content

fix(git-merge-regen): anchor merge=os-regen on whitespace tokens (#15701) #15748

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
baozhoutao merged 1 commit into from
Sep 5, 2026
Merged

fix(git-merge-regen): anchor merge=os-regen on whitespace tokens (#15701) #15748

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
130 changes: 117 additions & 13 deletions scripts/git-merge-regen.mjs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -344,7 +344,7 @@ function drive(argv) {
// ── Why the CALLEE NAME is the battery ──
//
// This file has no `selfTest()` entry function and no named section banners:
// the `--self-test` dispatch at the bottom invokes TWELVE named callees, each
// the `--self-test` dispatch at the bottom invokes THIRTEEN named callees, each
// printing its own line and returning a boolean. So the roster's unit is the
// CALLEE, and its label is the one the SOURCE ALREADY CARRIES — the function's
// own name. Nothing is invented and nothing is judged per comment, and a set
Expand All @@ -359,10 +359,10 @@ function drive(argv) {
// (PR #15271, `check-sdui-manifest`) makes a table row a battery. It does so
// for a file whose SELF-TEST *is* the table: one literal table, one driving
// loop over it, and a sink that writes only when a row fails. Here the table is
// a local of ONE callee among twelve, its rows are evaluated eagerly into
// a local of ONE callee among thirteen, its rows are evaluated eagerly into
// booleans before anything loops, and the callee already reduces them to a
// single printed verdict of its own. Flooring those rows would floor one
// callee's internals while the other eleven stayed at callee granularity — a
// callee's internals while the other twelve stayed at callee granularity — a
// roster whose unit changes per entry. The rule: the battery is the unit the
// DISPATCH names.
//
Expand All @@ -374,6 +374,7 @@ function drive(argv) {
// reaches it exactly once per run.
const SELF_TEST_BATTERIES = Object.freeze({
reconcileAttributes: 1,
reconcileAttributeTokenAnchor: 1,
reconcileAttributeSemantics: 1,
reconcileScripts: 1,
reconcileGenerators: 1,
Expand All @@ -393,7 +394,7 @@ const SELF_TEST_BATTERIES = Object.freeze({
// key in the literal above, so the roster falls below this number; the
// roster ↔ dispatch cross-check in the floor block is the other half, and it
// names WHICH callee was listed twice.
const SELF_TEST_BATTERY_FLOOR = 12;
const SELF_TEST_BATTERY_FLOOR = 13;

// The key a registration is filed under when a callee registers no name at all.
// It is not a declared battery, so it reds by the same set difference rather
Expand All @@ -403,7 +404,7 @@ const UNATTRIBUTED_BATTERY = '(no callee named)';
// The battery ledger, read by `batteryFloorFailures()` from the dispatch block
// at the very bottom of this file. It is MODULE-level rather than local to a
// self-test body because this file HAS no self-test body: the registrations
// happen inside twelve separate callees and the floor is read at the dispatch's
// happen inside thirteen separate callees and the floor is read at the dispatch's
// verdict site, so the ledger has to outlive every one of those frames.
//
// ⚠️ Named for the roster's role, deliberately NOT with a self-test spelling:
Expand All @@ -415,7 +416,7 @@ const batterySeen = new Map();
/**
* Record that a self-test callee RAN.
*
* Called as the FIRST statement of each of the twelve callees the `--self-test`
* Called as the FIRST statement of each of the thirteen callees the `--self-test`
* dispatch invokes — above any early return, so a callee that bails out early
* still reports that it ran, and the floor is never met by a frame that
* returned before doing anything.
Expand All @@ -428,7 +429,7 @@ function registerCase(name) {
/**
* The floor: every declared callee RAN (#13489).
*
* Evaluated at the dispatch's verdict site — after all twelve callees have had
* Evaluated at the dispatch's verdict site — after all thirteen callees have had
* their chance and immediately before the success line — and reached only from
* the `--self-test` branch, so a production merge-driver run never reads the
* ledger at all.
Expand Down Expand Up @@ -510,16 +511,56 @@ function fail(msg) {
return false;
}

/**
* A `.gitattributes` row routed to THIS driver, anchored on the WHITESPACE
* that delimits an attribute token (#15701).
*
* ⛔ NOT `/\bmerge=os-regen\b/`. `\b` sits between a word character and a
* non-word one, and `-` and `.` are both non-word, so the word-boundary
* spelling admitted rows git routes to a DIFFERENT driver:
*
* x merge=os-regen \b: true token: true git: os-regen
* x merge=os-regen-v2 \b: TRUE token: false git: os-regen-v2
* x merge=os-regen.2 \b: TRUE token: false git: os-regen.2
* x merge=os-regenX \b: false token: false git: os-regenX
*
* Latent rather than live: no sibling driver with this prefix exists today, so
* on the current file both spellings return the same 18 paths. It is worth
* anchoring anyway because `merge=os-regen-v2` is exactly how a second
* generation of this driver would be introduced beside the first — and that is
* the day `reconcileAttributes` reads the sibling's row as its own and reds
* with `mapped to merge=os-regen but not declared in regen-artifacts.mjs`: a
* rule the row does not break, on a path this driver does not own, whose
* suggested repair (declare it here) is the wrong one.
*
* An attribute value is a whitespace-delimited token, which is why whitespace
* is the thing to anchor on. `scripts/pm/os-regen-merge.sh`'s awk reader is
* the same predicate, character for character — the two readers are consulted
* about the same file and must not disagree about what it says.
*/
const DRIVER_ATTRIBUTE_ROW = /(^|[ \t])merge=os-regen([ \t]|$)/;

/**
* The paths a `.gitattributes` TEXT routes to this driver.
*
* Text in, paths out — not a read of the live file — so `reconcileAttributeTokenAnchor`
* can hold the SAME code the live reader runs against a fixture. The live
* `.gitattributes` carries no sibling-driver row, so it cannot measure this.
*/
function routedPaths(text) {
return text
.split('\n')
.filter((l) => l.trim() && !l.trim().startsWith('#'))
.filter((l) => DRIVER_ATTRIBUTE_ROW.test(l))
.map((l) => l.trim().split(/\s+/)[0]);
}

/** `.gitattributes` and the table must name the same paths — in both directions. */
function reconcileAttributes() {
registerCase('reconcileAttributes');
const file = join(REPO_ROOT, '.gitattributes');
if (!existsSync(file)) return fail('.gitattributes is missing — the driver is mapped to nothing.');
const mapped = readFileSync(file, 'utf8')
.split('\n')
.filter((l) => l.trim() && !l.trim().startsWith('#'))
.filter((l) => /\bmerge=os-regen\b/.test(l))
.map((l) => l.trim().split(/\s+/)[0]);
const mapped = routedPaths(readFileSync(file, 'utf8'));

const declared = REGEN_ARTIFACTS.map((e) => e.path);
const missing = declared.filter((p) => !mapped.includes(p));
Expand All @@ -540,6 +581,68 @@ function reconcileAttributes() {
return ok;
}

/**
* The row filter itself, held against a fixture (#15701).
*
* `reconcileAttributes` above reads the LIVE `.gitattributes`, and no row in it
* is routed to a sibling driver — there is no such driver yet. So the loose
* predicate and the anchored one return the same 18 paths on this tree and
* every assertion up there passes either way: the case that DISCRIMINATES them
* cannot be built from the live file, only from a fixture. Same reason
* `reconcileOwnership` exists beside `reconcileScripts`.
*
* Both directions are pinned. An anchor tightened until it matched nothing
* would satisfy every negative case here and route no artifact at all, which is
* the failure the positive rows exist to catch.
*/
function reconcileAttributeTokenAnchor() {
registerCase('reconcileAttributeTokenAnchor');
// What git itself says the sibling row routes to, measured on a scratch repo
// (git 2.43.0) whose `.gitattributes` held exactly `x merge=os-regen-v2`:
// $ git check-attr merge -- x
// x: merge: os-regen-v2
// A different driver. This reader must not count it among ours.
const fixture = [
'# packages/spec/commented.json merge=os-regen',
'',
' ',
'packages/spec/plain.json merge=os-regen',
'\tpackages/spec/indented.json\tmerge=os-regen',
'packages/spec/flanked.json text merge=os-regen -diff',
'packages/spec/sibling-dash.json merge=os-regen-v2',
'packages/spec/sibling-dot.json merge=os-regen.2',
'packages/spec/sibling-word.json merge=os-regenX',
'packages/spec/elsewhere.json merge=union',
].join('\n');
const routed = routedPaths(fixture);
const has = (name) => routed.includes(`packages/spec/${name}`);

const cases = [
['a plain row routes', has('plain.json')],
['a tab-delimited row routes', has('indented.json')],
['a row carrying other attributes on both sides routes', has('flanked.json')],
['merge=os-regen-v2 does NOT route — it is a different driver', !has('sibling-dash.json')],
['merge=os-regen.2 does NOT route — it is a different driver', !has('sibling-dot.json')],
['merge=os-regenX does not route', !has('sibling-word.json')],
['an unrelated merge driver does not route', !has('elsewhere.json')],
['a comment line naming the driver does not route', !has('commented.json')],
// Exactness, not membership: a predicate that also admitted something this
// list forgot to name would still satisfy all eight cases above.
['exactly the 3 routed rows come back', routed.length === 3],
];

const failures = cases.filter(([, ok]) => !ok).map(([name]) => name);
if (failures.length) {
return fail(`.gitattributes row filter ${DRIVER_ATTRIBUTE_ROW}:\n ${failures.join('\n ')}\n`
+ ' An attribute value is a WHITESPACE-delimited token, so whitespace is the anchor.\n'
+ ' `\\b` is not: `-` and `.` are non-word characters, so it admits sibling drivers\n'
+ ' and reports their paths as undeclared rows of THIS driver.');
}
console.log(`✓ .gitattributes row filter: ${cases.length} case(s) pinned,`
+ ' sibling drivers (merge=os-regen-v2, merge=os-regen.2) refused');
return true;
}

/** Where an owner's manifest lives, repo-relative, given a resolved directory. */
function manifestFor(dir) {
return dir === '.' ? 'package.json' : `${dir}/package.json`;
Expand Down Expand Up @@ -1302,13 +1405,14 @@ function endToEndMixed() {

if (process.argv.includes('--self-test')) {
console.log('git-merge-regen --self-test\n');
// The twelve callees as a literal LIST rather than twelve bare calls, so the
// The thirteen callees as a literal LIST rather than thirteen bare calls, so the
// names this block invokes are data the floor below can cross-check the
// roster against, in both directions. The names are read off the function
// declarations themselves (`fn.name`), so a renamed callee moves this list
// with it and cannot drift from the roster in silence.
const callees = [
reconcileAttributes,
reconcileAttributeTokenAnchor,
reconcileAttributeSemantics,
reconcileScripts,
reconcileGenerators,
Expand Down
Loading