knowledge: insert unauthored text with a String.replace replacer function ($$, $& are templates)

[choiyounggi]

Knowledge flush — 1 insight(s)

Headless autoflush run. 16 queue rows were claimed. 4 were found and judged: 1 ingested, 3 released as pending duplicates of open PRs. The other 12 were released untouched, because this run could not find them (see the Open-PR check).

Verified best-practice

Candidate 85e196ab8171f1d5 — when a JS/TS script inserts text it did not write with String.prototype.replace, pass that text through a replacer function.

• Claim: a replacement string is a template even when from is a plain string. $$ becomes $. $&, $` and $' insert the match, the text before it and the text after it. $n and $<name> insert capture groups (regex patterns only). The value a replacer function returns is inserted as-is. replaceAll and regex replacements follow the same rules.
• Sources checked:
  • https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects/String/replace
    • The replacement-patterns table.
    • "Specifying a function as the replacement": the return value is used as the replacement, and the special replacement patterns do not apply to strings it returns.
    • Confirmed through two web searches. One of them also turned up the same note in OpenJDK JDK-8016542 (https://bugs.openjdk.org/browse/JDK-8016542), where "abc".replace("c", "$&") with a function replacement must keep the literal $&.
  • https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects/String/replaceAll
    • The replacement has the same semantics as replace().
  • https://tc39.es/ecma262/#sec-getsubstitution
    • The spec operation behind both replace and RegExp.prototype[Symbol.replace]. A WebKit changeset in the search results cites it.
• Field evidence from the candidate's session: a plpgsql AS $$ … $$ fixture was spliced in with str.replace(from, to). Postgres then failed with 42601 syntax error at or near "$". Redoing the splice so the text went in verbatim made the suite pass 6/6.
• This headless run has no Node execution tool, so no new local repro was run. The page leans on the MDN text and the session's field evidence.
• Confidence: verified.

Existing-layer check

• wiki_search was run with the candidate's trigger and with a paraphrase (k=8 each). Top hits:
  • testing-quality-anchor-cases-for-allowlist-regexes
  • platforms-shells-escapes-in-shell-string-literals
  • testing-quality-schema-additions-under-a-golden-gate
  • platforms-shells-unset-versus-empty-parameters
  • infrastructure-config-config-values-emitted-as-source-text
  • frontend-security-xss-safe-rendering
  • testing-quality-json-manifest-edit-gates
  • testing-quality-ci-workflow-step-behavior-tests
• Opened and compared "When this applies":
  • anchor-cases: regex ^/$ anchors in allowlist tests. Different trigger.
  • escapes-in-shell-string-literals: backslash escapes in shell literals. Different trigger.
  • config-values-emitted-as-source-text: escaping a value a build step writes into generated source. This is the adjacent case, and it now has a related: link both ways.
• No existing page covers JS replace replacement patterns, so this is a new page. No conflicts were flagged.
• wiki/backend/node/index.md had no language category. wiki/backend/python/index.md does have one, and it served as the model.

Pages read: infrastructure-config-config-values-emitted-as-source-text, platforms-shells-escapes-in-shell-string-literals, testing-quality-anchor-cases-for-allowlist-regexes, infrastructure-agent-orchestration-worktree-isolated-workers

Open-PR check

Open knowledge/* heads: #280, #279, #275, #274, #273, #272, #271, #270, #269, #268, #266, #265. This run cannot git-diff them, so overlap was judged from each PR's title and file list (gh pr list … --json number,title,files).

Candidate	Overlapping open PR(s)	Verdict
85e196ab8171f1d5 JS String.replace $ patterns	None. No open PR touches backend/node/language/ or a JS string-replacement topic.	new, ingested
12aa086463fd4e2a macOS BSD grep: $ in a BRE with | alternation is literal	#273 (platforms/tools/bsd-vs-gnu-cli.md, platforms/shells/portable-shell-scripts.md) and #265 (wrapped grep)	pending duplicate, claim released
66f97671dfa1b324 Prisma 7 driver adapter: COMMIT-time errors arrive as DriverAdapterError and slip past an instanceof Prisma.*Error wrapper	#280 (a wrapper's own error label pinned by a boundary fault) and #269 (backend/common/orm/prisma-7-…)	pending duplicate, claim released
447c9601617b87ca Assertion-count detector misses unittest self.assert*	#270, #274 (testing/quality/checks-that-cannot-pass.md, harness-reverse-controls.md)	pending duplicate, claim released

The 12 other claimed ids could not be located:

97c37473b53e63c2, 51b969b2ddf1a0e4, eb432a16d29c7f6b, 059bcdc4893c6cb6,
9308b554babe5a27, a8587555b58e7c6e, 3c9c55ed8769b312, 3e9ce6cd2ed89d38,
96c6e1721a7f2aaf, a2ca98b37ab9ef8f, 7ec41d4bae57db6e, 056027e7c9098793

• Cause: this headless run's permissions do not allow listing ~/.dev-loop/queue/, and the session files holding these rows could not be found from session ids.
• They were released (set back to pending), not retired, so the next manual /dev-loop:knowledge-flush will process them.

Routing decision

• 85e196ab8171f1d5 → backend/node/language/string-replace-replacement-patterns.md, id backend-node-language-string-replace-replacement-patterns.
  • The candidate's domain: testing hint was not followed. The lesson is a JavaScript language rule that applies to any script splicing text, not a test-design rule.
  • This adds a new category language to the backend/node subtree. Why: the node subtree only has runtime (process/runtime environment), async and boundaries (validating external data). None of them covers built-in language semantics. The python subtree already keeps its equivalent traps under language/.
• Plumbing:
  • wiki/backend/node/index.md gets a new ## language section at the end of the file. It goes at the end so it does not sit next to rows that open PRs knowledge: dead first nameserver, plugin deps after update, split-task DoD, request-body reader cancel #266 and knowledge: claude --json-schema draft-7, scraped-text noise cut, fs-mock default export, count restatements #268 add to that index.
  • log.md gets an ingest line.
  • related: link added both ways with infrastructure-config-config-values-emitted-as-source-text.
• INDEX.md and wiki/backend/index.md were left unchanged. Their node rows already send any Node.js/TypeScript code question to the node subtree. Open PRs knowledge: wrapped grep, review diff base, frontmatter quotes, GITHUB_TOKEN chaining, blind LLM judges (+4 amends) #265/knowledge: no-html-link-for-pages misses static App Router routes, tsx top-level await without type:module, Server Action refresh() of the current route #274/knowledge: blur-driven layout shift that swallows the next click, fixtures whose id/position/order coincide #279 (INDEX.md) and knowledge: wrapped grep, review diff base, frontmatter quotes, GITHUB_TOKEN chaining, blind LLM judges (+4 amends) #265/knowledge: dead first nameserver, plugin deps after update, split-task DoD, request-body reader cancel #266/knowledge: claude --json-schema draft-7, scraped-text noise cut, fs-mock default export, count restatements #268/knowledge: stream pull probe, pooled lock control, jsdom select survivor, Korean bigram dedupe, clipboard copy, Next.js request-time reads, photo downscale #269/knowledge: refusing a subcommand when the selected backend lacks the capability it needs (scheme-only stderr line, usage-class exit code decided on entry) #272 (backend/index.md) edit those files.

Local-layer candidates

none