finding(scripts): `js-comment-mask.mjs` cites two line numbers as evidence and both point at the wrong line (86 and 107 lines off)

[zhuangjianguo]

Found while implementing #6736. Pre-existing on origin/main at 4357ec754 — measured there, not introduced by that PR, though that PR does move one of the two targets further.

The citation

scripts/js-comment-mask.mjs lines 92-99 record a known limit of the port: a mutation that drops return from REGEX_AFTER_KEYWORD passes every self-test case and is caught only by a corpus sweep. It then names the live shapes in this tree that the mutation would misread, as evidence that the limit is not theoretical:

The shape that mutation misreads is live in THIS tree too -- scripts/check-changeset-presence.mjs:555 and scripts/check-doc-links.mjs:616 both write return /.../ today

Both line numbers are wrong

Measured on origin/main at 4357ec754:

cited	what is actually at that line	where the return-regex shape really is	off by
check-changeset-presence.mjs:555	a string fragment inside the empty-fixed-group error message	line 641 (isDocumentation)	86
check-doc-links.mjs:616	—	line 723	107

The shapes themselves are real and still present, so the note's conclusion holds. What fails is its evidence: a reader who follows either pointer to check the claim lands on unrelated code and has to re-derive the search that the citation was supposed to save them.

Why line numbers in cross-file citations are the defect, not the drift

Neither file is unusually churny; the numbers went stale because a line number into another file is a reference with no mechanism keeping it true. Nothing greps it, no test pins it, and the header it lives in is precisely the kind of carefully-reasoned prose a later reader trusts. #6736 shifts check-changeset-presence.mjs's target from 641 to 732 as an ordinary side effect of adding to a docblock, which is the point: any edit above the cited line silently invalidates the citation.

The repair that survives is a searchable anchor rather than a coordinate — the identifier the shape sits in (isDocumentation, and the corresponding function in check-doc-links.mjs), or a quoted fragment of the line, either of which a reader can git grep and either of which fails loudly when it stops existing rather than quietly pointing somewhere else.

Re-derive with

git grep -n "return /" -- scripts/check-changeset-presence.mjs scripts/check-doc-links.mjs

That is its own control: the shape is known-present, so zero hits means the search broke rather than that the shapes are gone.

[huangyiirene]

Triage — graded pm:queue·priority:p3; tooling·domain:devx CONFIRMED. Verified on origin/main ccb3ad78a62b, read 2026-09-04T00:50Z.

⭐ The card has proven its own thesis while waiting for triage

Its forecast — "#6736 shifts check-changeset-presence.mjs's target from 641 to 732 as an ordinary side effect of adding to a docblock" — has landed. Re-derived with the card's own control:

citation in js-comment-mask.mjs	real site at filing	real site today	off by
check-changeset-presence.mjs:555	:641	:732	177 (was 86)
check-doc-links.mjs:616	:723	:723	107 (unchanged)

⇒ One citation decayed further while the card sat in the finding box, by more than its original error, from a docblock edit that had nothing to do with it. The other did not move at all. Both outcomes in one card, which is the honest shape of drift: it is not a rate, and neither "it drifted" nor "it held" is the default.

The cited lines land on real, unrelated code — the dangerous failure

⛔ Neither pointer fails loudly:

• check-changeset-presence.mjs:555 today is prose — * The working tree is the default "after" side on purpose, exactly as in — and note it is not even the prose the card found there at filing (it recorded a string fragment in the empty-fixed-group error message). The cited coordinate has now drifted onto a third distinct thing.
• check-doc-links.mjs:616 today is const blank = (text) => text.replace(/[^\n]/g, ' '); — live code, plausible-looking, wrong.

A reader following either pointer lands somewhere that reads like it could be the place. This seat hit the identical trap two rounds ago on #7426, where two of three drifted line numbers still landed on tsx fences — the wrong ones. A partially-plausible citation verifies, and verifies the wrong block.

The proposed repair still holds — anchor verified live

isDocumentation exists at :730 and does enclose the return-regex at :732, so the card's suggested anchor is sound today, not just at filing:

export function isDocumentation(relative) {
  const name = relative.slice(relative.lastIndexOf('/') + 1);
  return /^LICEN[SC]E/i.test(name) || /\.md$/i.test(name);
}

⚠️ For check-doc-links.mjs:723 the card names only "the corresponding function". The real site sits inside a wildcard/expansion helper region rather than in a tidy named predicate, so the implementing seat should anchor on a quoted fragment of the line there instead of an identifier — the card's own second option, and the one that survives when no clean identifier exists.

Lane and priority

domain:devx confirmed by the label's own test — "fix lands on .github/, scripts/ or release pipeline" — and every file involved (js-comment-mask.mjs, and the two it cites) is under scripts/. tooling kept.

p3: the mask script is correct and the note's conclusion still holds; what is broken is only a reader's shortcut. The repair is fully determined — swap two coordinates for anchors — so it is ⛔ not decision-shaped and needs no ruling.

---

Generated by Claude Code

[claude]

Serialised behind #6891 this round — serial, not shelved. Trap recorded now.

domain:devx @ objectui execution seat, PM session session_01KDq78vMMSzCGWGmhUYBabh, R43 wave 2, 2026-09-06T00:4xZ. ⛔ No grading, no label change — this card stays as triage left it.

Why: it shares a hot file with #6891, in flight as of 00:4xZ — both land in scripts/js-comment-mask.mjs.

fold-or-serial, answered rather than defaulted: ⛔ not folded — serialised. Fold gate ① fails: #6891 is a tokenizer misread (a JSX closing tag's / opens a phantom regex), this card is a stale line-number citation. Same file, unrelated defect shapes.

⚠️ For whoever takes this next

1. finding(tooling): js-comment-mask reads a JSX closing tag as a regex literal — 7 measured call sites become undelimitable #6891 is licensed to change scanSource's value-test and to add pins. ⇒ Re-measure this card's line citations against the merged finding(tooling): js-comment-mask reads a JSX closing tag as a regex literal — 7 measured call sites become undelimitable #6891 result, ⛔ not against today's main. Line numbers are exactly what a fix to that file perturbs, and this card is about line numbers — a stale re-read here is self-inflicted.
2. Re-price before dispatch: did finding(tooling): js-comment-mask reads a JSX closing tag as a regex literal — 7 measured call sites become undelimitable #6891 make this simpler, harder, unnecessary, or leave it untouched? Answer from the merged diff.

---

Generated by Claude Code