test(formula): correct the stale #6133 comment in parse-cel-to-ast.test.ts (#6678) - #7203

Closes #6678.

packages/formula/src/parse-cel-to-ast.test.ts carried two comments describing classifyError's pre-#6202 behaviour as if it were current. The first does not merely go stale, it instructs: it tells the next author that asserting ((record.a) -> parse "would enshrine it", i.e. that today's correct verdict is a bug not to be pinned.

What I measured, before rewriting anything

Per the card, the current grading was established by running the case, not by paraphrasing PR #6202. Temporary probe on this branch calling celEngine.compile over the whole SYNTAX_REJECTED list (probe removed before the commit):

MEASURED | "((record.a)"     -> kind=parse | message[0]="Expected RPAREN, got EOF"
MEASURED | "record.budget >" -> kind=parse | message[0]="Unexpected token: EOF"
MEASURED | "record.a $$ 1"   -> kind=parse | message[0]="Unexpected character: $"
MEASURED | "record.a ?? 3"   -> kind=parse | message[0]="Unexpected token: QUESTION"

 Test Files  1 passed (1)
      Tests  4 passed (4)

Two things fall out of that, and only one of them was in the card:

1. The grading claim in the comment is false — ((record.a) grades parse, not runtime.
2. The wording claim in the same comment is still true — cel-js does still phrase it Expected RPAREN, got EOF. What changed is that the phrasing no longer decides anything: classifyCelFault now branches on err instanceof ParseError (celEngine 的 classifyError 把「括号/方括号不配对」这类真语法错答成 kind: 'runtime',而该 kind 会原样出现在作者可见的拒写消息与 REST 响应体里 #6133 / PR fix(formula): classifyError 把 cel-js 的 parse 错误按错误类归为 syntax/parse,不再误报 runtime (#6133) #6202), and the keyword table was deleted outright by classifyError 的 type/runtime 两支仍按文案分类,而 cel-js 把作者源码嵌进 message —— 字段名叫 parse_status 的记录,求值期故障被答成 kind: 'parse' #6223 / PR fix(formula): grade CEL faults by error class + code, not by the message (#6223) #6677. The rewritten comment keeps the measured wording and corrects the mechanism, rather than deleting both.

Confirming the regex the comment quotes is gone: the only occurrence of parse|unexpected|syntax anywhere under packages/formula/src/ was the stale comment itself.

Were the surrounding assertions stale too?

No — checked, and this is the one part of the card that resolves in the reassuring direction. Nothing in the file asserts runtime for any source; the only runtime on line 142 was inside the prose. The old grading was never pinned, it was omitted — the test asserted record.budget > and deliberately said nothing about ((record.a). So there is no second half to file: the fix is text plus closing the omission the text was rationalising.

Changes

• :115-121 — the parity note now says classification is independent of wording and cites the two PRs that made it so.
• :135-144 — the "NOT asserted / would enshrine it" note is replaced by an it.each that asserts both wordings grade parse, promoting ((record.a) to an ordinary assertion now that its verdict is stable. The comment records both measured messages and points at cel-error-classification.test.ts as the exhaustive per-wording matrix, so the two suites do not read as rival sources of truth.

Changeset

None added; this PR needs the skip-changeset label. Read from the gates rather than assumed:

• changeset-check in pr-automation.yml has no path-based exemption — it counts .changeset/*.md added against the merge base whatever the diff touches. Its only two exemptions are the skip-changeset label and the changeset-release/main PR.
• So a test-only change has exactly two routes, and an empty-frontmatter changeset is not one of them: check:empty-changeset (空 frontmatter changeset 相对 skip-changeset 标签零收益、单向风险 —— 「禁止空 changeset 进 .changeset/」的决策证据(#5292 结案后无处存放) #5471) rejects an empty changeset a PR newly introduces, and lint.yml is explicit that the label and an empty file are not equals downstream — the label produces no input for changesets/action, an empty changeset produces a real one.
• This PR releases no package, which lint.yml:372-375 calls the textbook skip-changeset case.

Worth noting for the lane: taking that label means changeset-check is skipped wholesale, so the changeset-family self-tests do not run from there. They are covered anyway — lint.yml's Changeset-family gate self-tests step is deliberately unconditional (#6509).

Verification

pnpm --filter @objectstack/formula exec vitest run \
  src/parse-cel-to-ast.test.ts src/cel-error-classification.test.ts

 Test Files  2 passed (2)
      Tests  82 passed (82)

Related: #6678, #6133 / PR #6202, #6223 / PR #6677, #4812.

Generated by Claude Code