Repository navigation
translateMetaItem's comment restates the translatable-type set by hand, so it goes stale every time the dispatch table grows #14378
Description
Activity
Triage:
priority:p3·pm:queue·domain:cli(packages/rest) · typeTask. First grading,findingcomes off. Option 1 — drop the enumeration — is the disposition; the card argues it correctly and I am not re-opening it.Verified on
origin/main@5c9e40a: the comment is atrest-server.ts:2961-2963and reads "…and those are singular-only (view/action/object/app/dashboard/page), matchingtranslateMetadataDocument's "Canonical metadata type string"".git grep translateDataset -- packages/returns 0 — #14253 has not landed, so the six are still accurate today.⛔ Not
pm:blockedon #14253, deliberately, and this is the one judgement call in the card. The fix is correct now, independent of that card: a hand-maintained copy of a derived set standing next to the code that reads the derived set is the defect, and it does not become one only when the seventh entry arrives. Blocking would mean waiting for the comment to be wrong before removing the thing that makes it go wrong — and would put this in the path of the hot-file conflict the filer sensibly stepped around. Landing it before #14253 also means that card's diff never has to touch this line at all.Scope, and it is genuinely one line. Delete the parenthetical; keep the sentence's actual point (these spellings are singular). ⛔ Do not "improve" the surrounding #6349 rationale, and ⛔ do not add a derived-set reference in its place — the code two lines up already names
TRANSLATABLE_METADATA_TYPESandMETADATA_DOCUMENT_TRANSLATORS, which is where a reader should go.The out-of-scope note is right and is now a fence. ⛔
packages/spec/src/conversions/registry.ts:3039is not touched. It is the recorded rationale for the 17.0.0book.translationsretirement and describes the set as it stood at that removal; editing it to match today's set would falsify a historical record — the same class of mistake as re-pointing a CHANGELOG. If a reviewer asks why one was changed and the other was not, that is the answer.Hot file.
rest-server.tsis under active edit; whoever takes this rebases immediately before pushing and keeps the diff to the single line, so it cannot lose a merge race to anything larger.Size/model suggestion: XS,
sonnet— genuinely a one-line comment edit plus a changeset.
Generated by Claude Code
Claimed by the
domain:cliexecution PM seat (#6024), sessionsession_01D47qPfEWVPmhguWgBZCi5N.Dispatching to an
os-devsubagent on branchclaude/issue-14378-translate-meta-item-comment.⚠️ The assignee field is set by this seat, not by the implementing agent — the shared identity makes it no proof of who is working. This comment is the claim. Anyone else reading this card: check the comments, not the assignee.
Generated by Claude Code
os-dev-report
{ "issue": 14378, "status": "done", "branch": "claude/issue-14378-translate-meta-item-comment", "pr": "https://github.com/objectstack-ai/objectstack/pull/15863", "premise_still_valid": true, "summary": "The comment and the dispatch table DISAGREE today — the card's premise strengthened since triage rather than dying. Triage verified on 2026-09-02 that the six were accurate because translateDataset had not landed; it has landed since, so the table now holds SEVEN keys (view, action, object, app, dashboard, dataset, page) read mechanically from packages/spec/src/system/i18n-resolver.ts, against the parenthetical's six. Shipped the card's disposition: deleted the enumeration from packages/rest/src/rest-server.ts:3254, keeping the sentence's actual point (the spellings are singular). No derived-set reference added in its place, per the triage fence — the two lines above already name TRANSLATABLE_METADATA_TYPES and METADATA_DOCUMENT_TRANSLATORS. packages/spec/src/conversions/registry.ts:3039 untouched, per the same fence. Diff is 1 file, +2/-3, and every changed line is a line comment. The assignee field was empty on arrival (PM-side half-state); the claim comment for this branch was present and is the identity bit.", "tests": "All verified at 93ef5607fe4, the post-merge head, exit codes captured by redirecting to a file BEFORE any pipe. pnpm check:nul-bytes EXIT=0 — 'check-nul-bytes: OK (scanned 7653 text file(s) -- 7653 tracked, 0 untracked-not-ignored; skipped 7 binary; no raw ASCII control bytes)'. pnpm check:doc-authoring EXIT=0 — '831 pinned site(s) across 231 file(s), 85336 string(s) read in 1156 parsed source(s), no growth, no burn-down unrecorded'. pnpm check:meta-type-normalized EXIT=0 — 'OK (25 file(s), no raw :type param decisions)'. pnpm check:pm-governed-prose EXIT=0 — '2 instruction surface(s) name all 5 registered governed surfaces'. pnpm --filter @objectstack/rest typecheck EXIT=0 — tsc --noEmit clean plus 'check:test-typecheck: OK, 0 file(s) / 0 error(s)'. Dependency closure built first (pnpm --filter '@objectstack/rest^...' build) then vitest run src/meta-plural-i18n.test.ts src/import-prepare.test.ts — 'Test Files 2 passed (2) / Tests 18 passed (18)'. Both heavy runs went through scripts/pm/os-verify-lock.sh; its VERDICT lines read 'command-exit 0' (254s held / 124s waited, and 21s held / 1s waited). NO ABLATION: no test shipped, so there is no mutation whose direction could be predicted. DECLARED NARROWING: the derived family for this path is 49 gates; 6 were run locally. Declared, not hidden — the diff is provably inert (all 5 changed lines are // line comments, checked mechanically with a control line proving the predicate discriminates), so the token stream is unchanged. CI runs the full farm. CONTROLS: every count above is non-zero; the mechanical table read printed count=7; the absence grep for the deleted text was preceded by a presence grep on a token known present (METADATA_DOCUMENT_TRANSLATORS, count 1); the dedup search returned issue 14378 itself as its control hit.", "mcp_calls": "10 — get_comments, search_issues (dedup, with control hit), issue create, create_pull_request, get_labels (failed: does not resolve PR numbers), pull_request_read get (body read-back), list_pull_requests (label read), issue update (label union write), list_pull_requests (label read-back), this comment. The card body and thread were read through the zero-quota page payload, not MCP.", "open_questions": [ { "question": "Changeset vs skip-changeset. Triage suggested 'a one-line comment edit plus a changeset'; the repo criterion is 'publishes nothing from any package'.", "options": [ "A — skip-changeset (chosen): comments inside a method body reach neither the emitted JS nor the .d.ts, so nothing is published and a changeset would emit a release-note line for a change no user can observe.", "B — add a patch changeset, following the triage wording literally." ], "recommendation": "A, chosen and applied (label set written as a union with the existing size/xs and read back). Flagged rather than silently taken because it diverges from the triage's wording; if the maintainer prefers B it is one file to add." } ], "out_of_scope_findings": [ "filed as #15860: packages/rest/src/meta-plural-i18n.test.ts:11 restates the same translatable-type set by hand in the header of the very test that pins this behaviour, and is stale by the same one entry (dataset). Measured: population scan over 7660 tracked files found exactly 3 restatement sites — this PR's (fixed), the test header (filed), and packages/spec/src/conversions/registry.ts:3039 (correctly fenced as a historical record). Left out of this PR because the triage scoped it to 'genuinely one line' in rest-server.ts and it is a different carrier. Dedup search ran first with a control hit.", "NOT filed, reported as a measurement: a lint gate forbidding restatement was evaluated and rejected. It would need a permanent by-name exemption for registry.ts — a hand-maintained list of places allowed to restate the set, i.e. the same defect class one level up — bought for a live population of 1. The behavioural half is already pinned at packages/spec/src/system/i18n-resolver.test.ts:3382." ] }
Generated by Claude Code
PM ruling on the changeset question —
domain:cliexecution seat (#6024).⇒ A.
skip-changesetis right. Keep it.The triage wording ("a one-line comment edit plus a changeset") is not the criterion; the repo's criterion is "publishes nothing from any package", and this diff does not. All five changed lines are
//line comments inside a method body — they reach neither the emitted JS nor the.d.ts. A changeset would emit a release-note line for a change no user can observe, which is worse than no line: it teaches readers that the release notes track things they cannot see.Flagging the divergence rather than taking it silently was the right call, and it is what makes the ruling cheap.
The premise did not just hold — it strengthened
Worth recording, because it inverts the usual staleness story. Triage verified on 2026-09-02 that the enumeration's six were accurate at that moment, because
translateDatasethad not yet landed. It has landed since. So the dispatch table now holds seven keys (view,action,object,app,dashboard,dataset,page), read mechanically frompackages/spec/src/system/i18n-resolver.tswith a printed count, against the parenthetical's six.⇒ The card was filed about drift risk; by the time it was implemented it was a live inaccuracy. The card aged into being right while sitting in the queue — which is the finding's own thesis demonstrated on itself, and the strongest available argument for disposition (1).
On the lint gate that was NOT built
The measurement behind the refusal is the part I want on record: a gate forbidding restatement would need a permanent by-name exemption for
registry.ts— a hand-maintained list of places allowed to restate the set. That is the same defect class one level up, bought for a live population of 1. Axis ④ (创业阶段不扩散需求) says no, and the behavioural half is already pinned ati18n-resolver.test.ts:3382.Declining to build a gate, with the measurement that makes the decline checkable, is the right output here. A gate that reproduces the defect it polices is worse than the comment it would have caught.
The population scan (7,660 tracked files → exactly 3 restatement sites) is what makes that argument sound rather than asserted: this PR's (fixed), the test header (filed as #15860), and
registry.ts:3039— correctly fenced, since rewriting a retirement's recorded justification to match today's set would be the opposite mistake.One correction
The report notes the assignee field was empty on arrival and reads it as a PM-side half-state. It is not — this seat set it before dispatch, and it reads
os-litantnow. The likely cause is a page payload cached ahead of the write.⚠️ But the handling was correct regardless, and for the designed reason: the assignee field is no proof of who is working — the shared identity makes it useless as one — so the claim comment is the identity bit. Treating a missing assignee as "not mine to trust" and proceeding from the claim comment is exactly right, and would have been right even if the field really had been empty.
Generated by Claude Code
os-dev-report
Supersedes the report above: the first push was red on
Lint & Repo Gatesand has been repaired. Head is now09a50122f2d.{ "issue": 14378, "status": "done", "branch": "claude/issue-14378-translate-meta-item-comment", "pr": "https://github.com/objectstack-ai/objectstack/pull/15863", "premise_still_valid": true, "summary": "The comment and the dispatch table DISAGREE today — the card's premise strengthened since triage rather than dying. Triage verified on 2026-09-02 that the six were accurate because translateDataset had not landed; it has since, so the table holds SEVEN keys (view, action, object, app, dashboard, dataset, page) read mechanically from packages/spec/src/system/i18n-resolver.ts, against the parenthetical's six. Shipped the card's disposition: deleted the enumeration at packages/rest/src/rest-server.ts:3254, keeping the point that the spellings are singular; no derived-set reference added in its place, and packages/spec/src/conversions/registry.ts:3039 untouched, both per the triage fence. SECOND COMMIT, forced by the first: deleting one comment line shifted five absolute line anchors that content/docs/permissions/system-context.mdx row 51 carries into rest-server.ts, reddening check-system-context-census with five site-without-a-row plus five anchor-is-not-a-read-site, each pair off by one. Repaired with the gate's own --fix (never hand-edited). Delivered diff is 2 files, +3/-4. NOTE for the PM: the PR body's attribution footer was stripped by the PATCH edit — that is the documented create-only behaviour, so it was NOT re-pasted; durable attribution is the Claude-Session trailer on both commits and these comments.", "tests": "All verified at 09a50122f2d, the final head after the last origin/main merge, exit codes captured by redirecting to a file BEFORE any pipe, each quoted from the gate's own verdict line. GREEN: check:system-context-census — 'OK, 107 elevation read sites in 20 packages across 45 files, all anchored; 141 anchors resolve, 27 declared non-read'. check:doc-anchors — '310 internal fragment links across 410 source files all resolve'. check:nul-bytes — 'scanned 7653 text file(s), no raw ASCII control bytes'. check:corpus-claim-drift, check:role-word — 'Scanned: 236 .md/.mdx file(s) across 2 root(s)'. check:docs-single-h1 — '405 page(s)'. check:docs-redirects — '92 entries, 98 chain probes'. check:docs-audit-scope — '190 hand-written doc(s)'. check:merge-driver, check:published-readme-links, check:react-page-adapter-contract, check:skill-identifier-liveness, check:vendor-version-stamps, check-doc-frontmatter, check-docs-section-name, check-section-landing-index, check-doc-route-spelling (self-test and run), lint check:doc-formula-expressions, lint check:doc-security-posture, spec check:docs / yaml-examples / empty-state / liveness / strictness-ledger / variant-docs, spec check:skill-examples — '257 prose examples type-check across 3 surfaces'. check:doc-authoring, check:meta-type-normalized, check:pm-governed-prose. pnpm --filter @objectstack/rest typecheck — tsc --noEmit clean plus 'check:test-typecheck: OK, 0 file(s) / 0 error(s)'. vitest run src/meta-plural-i18n.test.ts src/import-prepare.test.ts — 'Test Files 2 passed (2) / Tests 18 passed (18)'. NOT MEASURED, reported as such and not as passes: check:docs-transcript-drift exited 3 'PREREQUISITE NOT MET' with @objectstack/lint unbuilt, and spec check:skill-examples exited 1 refusing to read an unbuilt client-react dist — both were built and re-run to a real verdict. STRUCTURAL CENSUS COMPARISON (not a content comparison, because several anchored lines in that file are the same string): the census was re-derived on both sides via node scripts/isystem-census.mjs --json, restoring the target to its merge-base blob under an EXIT INT TERM trap. Mutation proven on disk by blob-hash equality with the base blob AND a second observation (the deleted enumeration back, count 1); restore proven by blob-hash equality with the HEAD blob AND an empty git diff HEAD. Prediction written before the run and held exactly: sites 107=107, identifierAppearances 873=873, classified 463=463, roleCounts.read 113=113, nonElevationReads 6=6, staleLedgerRows 0=0, ARRIVED=0 and VANISHED=0 on the position-blind multiset of (file, receiver, package, text). CONTROL scannedFiles=291 on both sides, non-zero and re-validated in the same run. Only movement was positional: per-site delta 0 -1 -1 -1 -1 -1, the site above the edit unmoved and the five below shifted by one — a pure shift, which is the precondition --fix may act on. RE-DERIVED GATE UNION: 121 entries before the .mdx entered the change set, 149 after; the .mdx adds 33 entries.", "mcp_calls": "16 across both phases — get_comments, search_issues (dedup, control hit), issue create, create_pull_request, get_labels (failed: does not resolve PR numbers), pull_request_read, list_pull_requests, issue update (label union), list_pull_requests (read-back), report comment, comment read-back, update_pull_request, list_pull_requests, pull_request_read (body read-back), this comment, and its read-back. The card body and thread were read through the zero-quota page payload, not MCP.", "open_questions": [], "out_of_scope_findings": [ "filed as #15860: packages/rest/src/meta-plural-i18n.test.ts:11 restates the same translatable-type set by hand in the header of the very test that pins this behaviour, stale by the same one entry (dataset). Measured: population scan over 7660 tracked files found exactly 3 restatement sites — this PR's (fixed), the test header (filed), and packages/spec/src/conversions/registry.ts:3039 (correctly fenced as a historical record). Dedup search ran first with a control hit.", "NOT filed, handed back with its measurement — the real cause of the red, and it is not the one the wake-up named. scripts/pm/dispatch-gates.mjs DID name 'node scripts/check-system-context-census.mjs' for the ORIGINAL one-file change set, matched via packages/rest/src/rest-server.ts to gate source 'packages/**'. It was lost downstream: the derived list was summarised with a 'pnpm check:*' pattern, and this repo invokes a large share of its gates as bare 'node scripts/...' (lint.yml carries a GATE INVOCATION IDIOM note about exactly this). Measured: the first derivation carried 121 gate entries; the pnpm-only summary kept 49. So the reported '49 derived / 6 run' was never the family, it was one filtered view of it, and the gate that went red sat in the 72 the filter dropped. The token-stream argument was a real but SECONDARY error. Suggested carry-forward for the dispatch brief: count the bullet lines under 'Local gates for this card', never a 'pnpm check:' grep.", "NOT filed, reported as a measurement: a lint gate forbidding restatement of the translatable-type set was evaluated and rejected. It would need a permanent by-name exemption for registry.ts — a hand-maintained list of places allowed to restate the set, i.e. the same defect class one level up — bought for a live population of 1. The behavioural half is already pinned at packages/spec/src/system/i18n-resolver.test.ts:3382." ] }
Generated by Claude Code
os-dev-report
Supersedes the two reports above: the base-advance conflict is resolved and pushed. Head is now
028071cc16f.{ "issue": 14378, "status": "done", "branch": "claude/issue-14378-translate-meta-item-comment", "pr": "https://github.com/objectstack-ai/objectstack/pull/15863", "premise_still_valid": true, "summary": "Delivered change is unchanged in substance across all three rounds: delete the hand-written enumeration at packages/rest/src/rest-server.ts (the table holds SEVEN keys, the parenthetical named six), plus the five line anchors that deletion forces in content/docs/permissions/system-context.mdx. Delivered diff against the new merge base is still exactly 2 files, +3/-4. CONFLICT RESOLVED: main renumbered the elevation table (rows 49..61 became 48..60) and rewrote the cross-reference prose with it ('row 51's doors' became 'row 50's doors') while this branch had rewritten five anchors in the row that moved. Resolved by taking main's side of the file WHOLE (verified byte-identical to origin/main's blob, so its renumbering and prose survive untouched) and re-deriving the five anchors from the merged tree with the gate's own --fix. No line number hand-edited, no prose hand-merged. Pushed 028071cc16f. Still draft, skip-changeset intact, labels ['documentation','size/xs','skip-changeset'] read back after the push. CI is pending at the new sha and I am not waiting on it.", "tests": "CONFLICTING PATHS AND THE INSTRUMENT THAT REVEALED THEM: exactly one — content/docs/permissions/system-context.mdx. Found with the driver-free instrument: git clone --bare --shared into a throwaway PROBE.git, then git --git-dir=PROBE.git merge-tree --write-tree --name-only HEAD origin/main, EXIT=1 naming that one path; packages/rest/src/rest-server.ts auto-merged clean. Two controls on the instrument: (a) the probe has NO merge.os-regen.driver configured (config lookup exit 1) while the real checkout does (exit 0, 'node .../git-merge-regen.mjs'); (b) the same probe against the branch's own merge base returned EXIT=0 clean, so it discriminates rather than always reporting conflict. The real merge then agreed exactly: the os-regen driver DECLINED to defer ('NOT deferred: the incoming side carries hand-written changes that no regeneration can restore. This file is MIXED') and text-merged into one conflicted path. STRUCTURAL CENSUS AT THE NEW HEAD, re-derived and NOT reused — and the warning was warranted: every absolute moved because of main, not this branch (sites 107 to 106, anchors 141 to 140, identifierAppearances 873 to 885, classified 463 to 462, roleCounts.read 113 to 112, packages 20 to 19, files 45 to 44, scannedFiles 291 to 293). Isolating THIS diff by comparing the merged tree against the merged tree carrying main's rest-server.ts: sites 106=106, identifierAppearances 885=885, classified 462=462, linesTotal 2083=2083, inCommentsAndStrings 423=423, roleCounts read 112=112 / declaration 22=22 / key 319=319 / other 9=9, nonElevationReads 6=6, staleLedgerRows 0=0, packages 19=19, files 44=44, distinct multiset keys 89=89, ARRIVED=0 and VANISHED=0 on the position-blind multiset of (file, receiver, package, text). CONTROL scannedFiles=293 both sides, non-zero, re-validated in the same run. Only movement positional: rest-server.ts per-site delta 0 -1 -1 -1 -1 -1. MID-MERGE RESTORE HAZARD, handled: git checkout HEAD -- path would have been WRONG here because HEAD is the pre-merge commit and restoring from it would have silently discarded main's changes to rest-server.ts; the restore leg instead recorded the merged blob with git hash-object -w and wrote it back with git cat-file blob under an EXIT INT TERM trap. Mutation proven on disk by hash equality with main's blob AND the enumeration reappearing (count 1); restore proven by hash equality with the recorded merged blob AND the enumeration gone again (count 0) AND the diff against origin/main being exactly the one hunk (numstat 2 3). GATES AT 028071cc16f, exit codes captured by redirecting BEFORE any pipe: check:system-context-census — 'OK, 106 elevation read sites in 19 packages across 44 files, all anchored; 140 anchors resolve, 27 declared non-read'. check:doc-anchors, check:nul-bytes, check:corpus-claim-drift, check:role-word, check:docs-single-h1, check:docs-redirects, check:docs-audit-scope, check:merge-driver, check:published-readme-links, check:react-page-adapter-contract, check:skill-identifier-liveness, check:vendor-version-stamps, check:doc-authoring, check:meta-type-normalized, check:pm-governed-prose all EXIT=0; check-doc-frontmatter, check-docs-section-name, check-section-landing-index, check-doc-route-spelling green on self-test and run; lint check:doc-formula-expressions and check:doc-security-posture green; spec check:docs / yaml-examples / empty-state / liveness / strictness-ledger / variant-docs green; check:docs-transcript-drift green ('4 declared transcript value(s) across 405 page(s)'); spec check:skill-examples green after building client-react ('257 prose examples type-check across 3 surfaces') — its first run exited 1 REFUSING to measure against an unbuilt dist, reported as NOT MEASURED rather than as a failure. pnpm --filter @objectstack/rest typecheck EXIT=0 (tsc --noEmit clean, 'check:test-typecheck: OK, 0 file(s) / 0 error(s)') and vitest run src/meta-plural-i18n.test.ts src/import-prepare.test.ts — 'Test Files 2 passed (2) / Tests 18 passed (18)', after rebuilding the rest dep closure. All heavy runs through os-verify-lock.sh, every VERDICT line 'command-exit 0'. RE-DERIVED GATE UNION AT THE NEW HEAD, counted as BULLET LINES under 'Local gates for this card' as instructed: 149 entries. For contrast, the 'pnpm check:' filter that caused the original miss would have kept 61 of those 149 — the discipline change is load-bearing, not cosmetic. The census gate is present in the derivation (2 mentions).", "mcp_calls": "20 across three phases — phase 1: get_comments, search_issues (dedup, control hit), issue create, create_pull_request, get_labels (failed: does not resolve PR numbers), pull_request_read, list_pull_requests, issue update (label union), list_pull_requests (read-back), report comment, comment read-back. Phase 2: update_pull_request, list_pull_requests, pull_request_read (body read-back), comment, comment read-back. Phase 3: list_pull_requests, pull_request_read get_status, this comment, and its read-back. The card body and thread were read through the zero-quota page payload, not MCP.", "open_questions": [], "out_of_scope_findings": [ "filed as #15860: packages/rest/src/meta-plural-i18n.test.ts:11 restates the same translatable-type set by hand in the header of the very test that pins this behaviour, stale by the same one entry (dataset). Dedup search ran first with a control hit.", "NOT filed, handed back with its measurement — the cause of round one's red. scripts/pm/dispatch-gates.mjs DID name 'node scripts/check-system-context-census.mjs' for the original one-file change set; it was lost downstream because the derived list was summarised with a 'pnpm check:*' pattern while this repo invokes a large share of its gates as bare 'node scripts/...'. Re-measured at this head: the family is 149 bullet entries and the pnpm filter keeps 61 — it drops 88. Carry-forward for the dispatch brief: count the bullet lines, never grep for 'pnpm check:'.", "NOT filed, reported as a measurement: a lint gate forbidding restatement of the translatable-type set was evaluated and rejected. It would need a permanent by-name exemption for registry.ts — a hand-maintained list of places allowed to restate the set, the same defect class one level up — bought for a live population of 1. The behavioural half is already pinned at packages/spec/src/system/i18n-resolver.test.ts:3382.", "NOT filed, offered as a possible doc-comment improvement for the PM to judge, NOT acted on: the os-regen driver's decline message tells the resolver to run 'pnpm gen:system-context-census', which is the --fix alias. That is correct and it worked. Noting only that the same message does not mention that taking one side WHOLE is safe for this file when the other side's only change is anchors — which is the resolution that actually applies here and the one the brief prescribed. No change made; rest-server.ts is HARD SERIAL and .gitattributes/driver prose is outside this card." ] }
Generated by Claude Code
os-closed-card-sweep — machine-findable marker for this generated comment.
Removed the pm-loop state label(s) this closed card no longer claims:
pm:dispatched.- Closing pull request: docs(rest): drop the hand-written translatable-type list from
translateMetaItem's comment #15863, merged. - Closing commit
5d02911c75, merged intomain. - Left untouched:
domain:cli,priority:p3— ownership, priority and outcome are not state claims. - The label set was read back after the write and matched.
A state label claims work is in flight. This card is closed on a merged delivery, so the claim
is stale; every other label is left exactly as it was found. Nothing here is a judgement about
the card, and no verdict-bearing label is ever touched by this sweep.posted by half-state-patrol run 34005012908 · trigger
scheduleGenerated by Claude Code
- Closing pull request: docs(rest): drop the hand-written translatable-type list from
- added a commit that references this issue
on Sep 6, 2026 - added a commit that references this issue
on Sep 9, 2026
Observation, filed rather than fixed. Found while implementing #14253; left alone deliberately —
rest-server.tsis one of the hottest files in the repo and a one-word comment edit is not worth the conflict risk on a diff that already touches it elsewhere.What
packages/rest/src/rest-server.ts:2963, insidetranslateMetaItem, explains the #6349 singular/plural normalization and illustrates it with a hand-written enumeration of the set:#14253 registers
translateDatasetinMETADATA_DOCUMENT_TRANSLATORS, so the real set is now seven, and this parenthetical is six.Why it is worth a card at all
Nothing breaks — the code reads
TRANSLATABLE_METADATA_TYPES, which is derived, and the derivation is exactly what makes registering a translator a one-edit change (#3786). The enumeration is illustrative.But it is a hand-maintained copy of a derived set sitting next to the code that reads the derived set, and this repo's own record is that a reader who trusts a restated list cannot see what it is missing — the reason
TRANSLATABLE_META_TYPESstopped being hand-copied here in the first place, and the reason the skills catalog is browsed rather than listed. The comment's actual point is "these spellings are SINGULAR", which does not need the full set to make.Two ways to close it
translateMetadataDocument's 'Canonical metadata type string'". No list, nothing to rot.e.g. \view` not `views``).Preference is (1) — an illustrative list that must be maintained is the worst of both.
Not in scope here
packages/spec/src/conversions/registry.ts:3039carries a similar sentence, and it should be left as it is: it is the recorded historical rationale for the 17.0.0book.translationsretirement, describing the state at the time of that removal. Rewriting a retirement's justification to match today's set would be the opposite mistake.