OpenConceptLab/ocl_issues#2917 | Remove 223 unused i18n keys, and add the 4 the code uses to every catalog

[paynejd]

What changed and why

The Mapper's catalogs had drifted from its code (OpenConceptLab/ocl_issues#2917, a follow-up to OpenConceptLab/ocl_issues#2880). On main, 209 of the 791 English keys were unreferenced, and the code used four keys no catalog had.

• Four keys added to en, es, pt-BR and zh. common.none (MatchDetailsPanel.jsx), common.saved (MultiAlgoSelector.jsx) and search.save_default_filters (SearchFilters.jsx) showed the raw key in every language. map_project.candidates_legacy_shape (MapProject.jsx) showed its English default. common.none and search.save_default_filters take oclweb3's wording in all four locales, and common.saved follows the catalogs' existing "saved" strings. The English for candidates_legacy_shape is the default the call already passes, so English reads the same.
• The 209 unreferenced keys removed from all four catalogs. This is the ticket's list, still the same on current main. The bookmarks namespace, left empty, goes too.
• The optional item: the 14 keys only the non-concept column sets named. ResultConstants.js keeps only ALL_COLUMNS.concepts. TableResults.jsx reads ALL_COLUMNS[resource], and its only parent, SearchResults, is rendered by map-projects/Candidates.jsx and Search.jsx, both with resource='concepts'. So the mappings, repos, orgs, users and url_registry sets never rendered. They go, with their 14 keys and the three pieces only they used: mappings/ConceptCell.jsx, mappings/FromAndTargetSource.jsx and formatWebsiteLink() in common/utils.js.
• es's two keys that English lacks, common.or_go_to_your and concept.history, go too. Nothing references them, and no other catalog has them.

Each catalog now has the same 572 keys (791 − 209 − 14 + 4).

How each removed key was checked

• The list. The static audit from OCL Mapper: Audit unused i18n keys and Portuguese translation coverage ocl_issues#2880, run locally from its unmerged branch, reports the same 209 unreferenced keys and the same four missing ones on current main.

• Last segment. Every string literal under src/ that contains a key's last segment as a word was listed, and none of them builds a key. Only three key patterns built at run time span a whole namespace, at seven call sites, and each was read by hand:

  • `concept.${prop}` (ConceptProperties.jsx) yields only concept_class and datatype;
  • `search.${resource}` (SearchResults.jsx) yields only concepts;
  • 'map_project.' + bucket (MapProject.jsx, ScoreBucketButton.jsx) yields only the three score buckets.

  Five t() calls take their key from a variable. Four get it from literals that aren't on the list: announcement links, the usage chip's lines, target-version badges and column labels. The fifth translates an algorithm whose name is a <Trans> element, and there is none: API algorithms are JSON, and the two built-in ones name themselves with t().

• History. Each full key's history, the commits git log -S lists but with the key matched as a whole word, ends with the commit that removed its last use. 52 keys were never used in this repo's source. Of the other 157, 113 went with the oclweb3 leftovers removed in 4ae1a18 and 4c310cc. The three most recent removals (64511e9, 2806eef, 346cab7) drop the call outright rather than build the key another way.

• After the change, the audit reports no unreferenced keys and nothing needing attention, and no locale has a missing, blank or extra key.

Tests

src/i18n/__tests__/catalogKeys.test.js checks that:

• no locale has a key English lacks;
• every literal key in the source (t('a.b'), this.props.t('a.b'), t(`a.b`), i18nKey='a.b') is in the English catalog. It reads 598 uses of 479 keys. Keys built at run time aren't read.

Against main's catalogs, 2 of its 5 tests fail: es has two keys English lacks, and the literal-key check names exactly the four call sites above. With this branch's pt-BR and zh but main's English, those two locales fail as well, since they'd hold the four new keys and English wouldn't. aiAssistantNaming.test.js no longer lists the removed map_project.ai_assistant_limit_reached.

The gate, run as pr.yml runs it, on Node 22.23.2:

• the start-prod.sh secret check: passed;
• npm ci: ok;
• eslint: clean. I ran it with --no-eslintrc -c .eslintrc.json, the config CI uses: this checkout is nested in another, and ESLint would load that one's config too;
• npm run test:all: 511/511 unit tests and 6/6 integration tests pass;
• a development webpack build compiles.

PR checks: "Lint and test" passed on 4f2332d (run 39062561106).

Visual check

A throwaway harness, not committed, rendered two components side by side with main's catalogs and with this branch's. It ran in headless Chromium through Playwright, with each component in a 360px column, once per locale: en, es, pt-BR and zh. The ticket names no widths.

• MatchDetailsPanel, for a match with a highlight but no raw score. On main the Raw Scores row shows common.none in all four languages. On this branch it shows None, Ninguno, Nenhum and 无.
• SearchFilters, as a nested panel whose user may save default filters. On main the button shows search.save_default_filters. On this branch it shows Save as Default Filters, Guardar como Filtros Predeterminados, Salvar como filtros padrão and 另存为默认筛选条件. The longest, es, fits on one line.

The page logged no errors. common.saved and map_project.candidates_legacy_shape weren't rendered: one shows only after a custom algorithm's details are saved, the other only when a project loads in the old candidates format. The literal-key test covers their English entries.

Acceptance criteria

• Add common.none, common.saved, search.save_default_filters and map_project.candidates_legacy_shape to en, es, pt-BR and zh. Done in 37ef20e. catalogKeys.test.js ("every literal key in the source is in the English catalog") fails without the English entries, and the visual check above shows two of the keys in all four languages.
• Remove the unreferenced keys from all four locales at once, each checked by hand first (its last segment, and git log -S). Done in 09d5f6b; the check is described above. catalogKeys.test.js ("has no key English lacks") fails if a key leaves English but stays in another locale.
• Optional: remove the 14 keys in the non-concept column sets of ResultConstants.js, along with those sets. Done in 5e4f2e7.

Docs

Nothing to update. No README, doc or comment describes the catalogs' keys or the removed column sets, and the one comment that names ALL_COLUMNS['concepts'] (Candidates.jsx) still holds.

Not done

• No existing key is reworded; only the four new keys are translated.
• map_project.candidates_legacy_shape's call keeps its English default, as the Mapper's other calls with defaults do (map_project.target_repo_version_required, for one).

Noticed, not fixed

• addAlgo in MultiAlgoSelector.jsx still handles an algorithm whose name is a <Trans> element. No algorithm defines its name that way, so that branch never runs.
• HIGHLIGHT_ICON_WHITELISTED_FILEDS in ResultConstants.js is imported nowhere. FACET_ORDER and SORT_ATTRS there still have entries for other resources, and SearchResults.jsx branches on those resources too. None of them holds i18n keys, so they're outside this ticket.
• formatWebsiteLinkTruncated() in common/utils.js has no caller, here or on main.
• OpenConceptLab/ocl_issues#2907 | suggesting field mapping for column headers #106 also edits the four catalogs. It uses none of the removed keys, but whichever of the two merges second will need its catalogs rebased.

Review

• Codex, pass 1 (5e4f2e7): clean.
• Simplification review (5e4f2e7): 4 items. S1 and S2, both in catalogKeys.test.js, were applied in 4f2332d: a stdlib directory listing, and no per-locale test of the four keys. S3 and S4 were kept, since that code was dead before this PR and holds no i18n keys; both are listed under "Noticed, not fixed".
• Codex check of the cuts (4f2332d): clean, so no cut was reverted.

No P0 or P1 is open.

closes OpenConceptLab/ocl_issues#2917

🤖 Generated with Claude Code

https://claude.ai/code/session_01M8452pAtpyMrv7QRKXbUsj

Written by an Octoad session worker (worker-v0.11, recipe lean-loop, flow public>public, run openconceptlab-ocl-issues-2917-20261017-145015).

[paynejd]

Codex review, pass 1 (codex-cli 0.163.0, commit 5e4f2e7)

Clean.

findings.json

{
  "reviewer": "codex-cli 0.163.0",
  "perspective": "defects",
  "mode": "full",
  "commit": "5e4f2e7",
  "verdict": "clean",
  "prior": [],
  "findings": []
}

[paynejd]

Response to Codex pass 1

No findings, so nothing to fix. No P0 or P1 is open, so the defect loop stops after this pass.

[paynejd]

Simplification review (commit 5e4f2e7)

4 items.

• S1 · stdlib · src/i18n/tests/catalogKeys.test.js:42 · the recursive sources walker → readdirSync(SRC_DIR, { recursive: true }), filtered to .js/.jsx files outside the test folders (-4)
• S2 · dup-test · src/i18n/tests/catalogKeys.test.js:24 · the ADDED list and its per-locale test, which repeat the literal-key test for en and pin four one-off translations in the other locales → nothing replaces it (-8)
• S3 · delete · src/common/utils.js:31 · formatWebsiteLinkTruncated, which had no caller before this PR either, with the React, Tooltip and merge imports only it uses → nothing replaces it (-26)
• S4 · delete · src/components/search/ResultConstants.js:27 · HIGHLIGHT_ICON_WHITELISTED_FILEDS, which nothing imports → nothing replaces it (-9)

findings.json

{
  "reviewer": "claude subagent",
  "perspective": "simplify",
  "mode": "full",
  "commit": "5e4f2e7",
  "verdict": "findings",
  "prior": [],
  "findings": [
    {
      "id": "S1",
      "severity": null,
      "tag": "stdlib",
      "file": "src/i18n/__tests__/catalogKeys.test.js",
      "line": 42,
      "claim": "The recursive `sources` walker (6 lines) does what fs.readdirSync(dir, { recursive: true }) already does. Node has shipped that option since 20.1, and CI and the Dockerfile run Node 22. On this tree both list the same 171 files.",
      "scenario": null,
      "action": "const SOURCES = readdirSync(SRC_DIR, { recursive: true })\n  .filter(file => /\\.jsx?$/.test(file) && !/(^|\\/)__(tests|integration)__\\//.test(file))\nThe test then reads readFileSync(path.join(SRC_DIR, file), 'utf8') and reports `${file}: ${key}`. The paths come back already relative, so the path.relative call goes too.",
      "lines_delta": -4,
      "confidence": "high"
    },
    {
      "id": "S2",
      "severity": null,
      "tag": "dup-test",
      "file": "src/i18n/__tests__/catalogKeys.test.js",
      "line": 24,
      "claim": "For en, the ADDED list and its per-locale test repeat the literal-key test. All four keys are literal t() calls that LITERAL_KEY matches (checked), so if one goes missing from en, 'every literal key in the source is in the English catalog' already fails. For es, pt-BR and zh, the test pins four translations chosen only by this ticket's history. No other translation is pinned, and i18next falls back to English when one is missing.",
      "scenario": null,
      "action": "Nothing replaces it. The loop keeps only the 'has no key English lacks' test.",
      "lines_delta": -8,
      "confidence": "medium"
    },
    {
      "id": "S3",
      "severity": null,
      "tag": "delete",
      "file": "src/common/utils.js",
      "line": 31,
      "claim": "formatWebsiteLinkTruncated, the twin of the formatWebsiteLink this PR deletes, has no caller on HEAD or on origin/main. It holds the only JSX and the only merge call in utils.js, so import React (line 3), import { Tooltip } (line 5) and merge in the lodash import (line 8) can go with it. It was dead before this PR, so whether to include it here is the author's call.",
      "scenario": null,
      "action": "Nothing replaces it.",
      "lines_delta": -26,
      "confidence": "medium"
    },
    {
      "id": "S4",
      "severity": null,
      "tag": "delete",
      "file": "src/components/search/ResultConstants.js",
      "line": 27,
      "claim": "Nothing imports HIGHLIGHT_ICON_WHITELISTED_FILEDS, on HEAD or on origin/main. It is dead code in the file this PR trims down to the concept columns. It holds no i18n keys, though, so it is outside the ticket's scope.",
      "scenario": null,
      "action": "Nothing replaces it.",
      "lines_delta": -9,
      "confidence": "low"
    }
  ]
}

[paynejd]

Response to the simplification review

• S1 applied in 4f2332d
• S2 applied in 4f2332d
• S3 kept: formatWebsiteLinkTruncated had no caller before this PR either, and has nothing to do with the catalogs or the column sets, so it's outside OCL Mapper: remove unused i18n keys, and add the 4 English keys the code uses ocl_issues#2917.
• S4 kept: HIGHLIGHT_ICON_WHITELISTED_FILEDS holds no i18n keys and was dead before this PR, so it's outside the ticket. The PR's "Noticed, not fixed" lists it.

[paynejd]

Codex check of the cuts (commit 4f2332d)

Clean.

findings.json

{
  "reviewer": "codex-cli 0.163.0",
  "perspective": "defects",
  "mode": "cuts",
  "commit": "4f2332d",
  "verdict": "clean",
  "prior": [],
  "findings": []
}