Repository navigation
refactor(website): render split headings as complete messages - #21034
benceruleanlu wants to merge 2 commits into
Conversation
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
Five headings were translated as before/after fragments around a styled span, so each locale had to split its sentence where English did. Each is now one message with a named placeholder, rendered with vue-i18n's I18nT slots, so word order belongs to the locale. Existing Chinese and Japanese copy is joined into the new messages rather than retranslated. The website catalogs are typed as string-keyed message dictionaries so the per-locale composer satisfies I18nT's i18n prop.
|
Important Review skippedReview was skipped due to path filters ⛔ Files ignored due to path filters: 13
CodeRabbit blocks several paths by default. You can override this behavior by explicitly including those paths in the path filters. For example, including ⚙️ Run configuration
You can disable this status message by setting the Use the checkbox below for a quick retry:
Comment |
fbcf793 to
5b9f4bb
Compare
🧩 Custom Nodes Ecosystem Matrix: 🟡 Withheld — could not measureExecutes the frontend JS of every registry pack that ships any, against this change, in both renderers. The eight This candidate changes nothing the matrix measures, so every expensive stage was skipped. Reproduce locally# build one shard's fixture, then run it
python3 scripts/registry-census/build_matrix.py --shard 1/4
MATRIX_RENDERER=vue MATRIX_VUE=1 \
npx vitest run --config vitest.matrix.config.mts
Full docs: scripts/registry-census/README.md 🌾 Fallow: ✅ PassedFallow audits the diff for duplication, complexity and dead code. It gates on new findings only, so when the audit completes, a red is something this PR introduced or moved — never inherited backlog. A red can also mean the audit itself did not finish; the details below say which happened. Fallow gates on new findings in changed files only, so everything below was introduced or moved by this PR (17 changed files).
Pre-existing findings are baselined in Reproduce locally# the base the gate compared against
npx fallow audit --root . --quiet --format json \
--changed-since f3bf32f94c0a0d0bc350b2500427a49b4439f1fb
Full docs: ADR-DEVEX-LINT-0015 🌐 Website E2ETip All tests passed.
🔗 Website PreviewWebsite Preview: https://comfy-website-preview-pr-21034.vercel.app Models is built and gated by PostHog (workshop-enabled), using staging Cloud. This commit: https://website-frontend-98848qq77-comfyui.vercel.app Last updated: 2026-10-14T23:57:42Z for |
Codecov Report✅ All modified and coverable lines are covered by tests. @@ Coverage Diff @@
## main #21034 +/- ##
==========================================
- Coverage 86.40% 86.34% -0.07%
==========================================
Files 3687 3582 -105
Lines 175499 172258 -3241
Branches 48434 47332 -1102
==========================================
- Hits 151636 148732 -2904
+ Misses 21812 21619 -193
+ Partials 2051 1907 -144
Flags with carried forward coverage won't be shown. Click here to find out more.
... and 113 files with indirect coverage changes 🚀 New features to boost your workflow:
|
There was a problem hiding this comment.
Verdict: no blocking defects; two P2 test gaps and some P3 items to settle before human review
I compared the preview deployment with comfy.org for the changed headings. Text matches on /about, /careers and /cloud in en and zh-CN, and on /zh-CN/models. Element screenshots of the careers "Why join" card, the cloud audience heading and the zh-CN about values heading are pixel-identical (ImageMagick AE 0). The only text difference is the careers title losing a space before the <br>, which does not render. So the refactor holds up at runtime. What it lacks is protection. keypath is typed string, so a typo or a renamed slot in three of the five headings passes typecheck and the whole website suite. The other items are a translation-pipeline consequence of reusing about.careers.whyTitle, overlap with three of your open PRs, and test and style nits.
Findings without an anchorable line (6)
- P2, open PR overlap with #20943. Draft #20943 adds a
messageGroupsentry that validatesmodels.list.heroTitleBeforeandheroTitleAfteras one joined string. It also drops theheroTitleAfterknown violation and rewrites the jaheroTitle*values. This PR deletes both keys. If this lands first, that group points at missing keys and its check returns[]without an error, and the manifest and jamodels.listhunks conflict. #20943 should drop that group entry and the ja hunk. - P3, open PR overlap with #20585. Draft #20585 rewrites ja copy for the split
headingBefore/headingAfterkeys this PR deletes, and editsOurValuesSection.vueandStorySection.vue. Rebased after this merges, its ja heading copy lands on deleted keys and gets pruned. No ja about page is live yet (src/pages/ja/has onlyindex.astro), so nobody sees it today. - P3, FE-3069 is not fully resolved. The body says "Resolves FE-3069". That issue's acceptance criteria say no translated message is split.
src/components/product/shared/ReasonSection.vue:42-48still rendersheadingKey+headingHighlightKey+headingSuffixKeyfragments.cloud.reasonanddownload.reasonuse it ("Why professionals choose " + "Cloud"), and zh-CNcloud.reason.headingis a locked-order fragment.ModelLaunchHeroSection.vue(titleKey+titleRestKey) andReasonsSplit01.vuehave the same shape. Either migrateReasonSectionhere or change "Resolves" to "Part of" and track the rest. The README Localization section also does not mention the I18nT slot convention yet, so the next heading will probably copyReasonSection. - P3 Nit, comments left in
WireNodeLayout.vueblocks the diff edits. The diff deletes the<!-- Left column … -->label but keeps its siblings<!-- Center column: Reasons card -->(L218) and<!-- Right column: slot for card content -->(L255), plus<!-- Desktop layout -->(L144) and<!-- Mobile layout -->(L282). Each restates the markup under it. ThevariantJSDoc at L26-30, in thedefinePropsblock the diff edits, lists consumer routes (/careers, /about,/forward-deployed-creatives) that go stale when another page adopts the layout. Cut it to one contract line. These lines sit outside the diff hunks, so I can't attach suggestions. - P3, visual baselines. No
toHaveScreenshotgolden covers any of the five headings.e2e/visual-responsive.spec.ts's About block captures only the hero, and the diff updates no goldens. An element-level screenshot of the values or story heading in that About block, and of the audience heading ine2e/cloud.spec.ts, would add a baseline without another page load. Context only: the existingcloud.spec.ts:68checkgetByText(/creators/i).first()also matches the "CREATORS" card label, so it passes even with the heading broken. - P3, test count in the description. "Website unit tests pass (3,141)". A full website run at this head reports 612 files and 11,805 tests.
Unverified (1)
- P3, #20366 merge fit. #20366 removes
messagesfromcreateI18nintranslations.ts, which conflicts with themessagesconstant added here. Without a typedmessages, the composer should still fit I18nT'si18n?: Composer, but I have not checked it. To settle it, merge #20366 onto5b9f4bband runpnpm typecheck:website.
Checked and cleared (5)
- Careers title whitespace. Base renders "Why join ?" / "为什么加入 ?" and head renders "Why join?" / "为什么加入?". The space sat before a forced
<br>, and preview vs production screenshots of the card are pixel-identical. - Hardcoded
about.careers.whyTitlein acommon/layout. On base the removedtitleBeforeKey/titleAfterKeyprops already defaulted toabout.careers.*, and no caller passed them.BuildersSectionusesvariant="node", which never renders the title. Deleting the props is right. - ja row in the models hero table. ja has the same order as zh-CN, but #21033 just fixed the ja hero order in the catalog. The row pins that fix, and it can regress without zh-CN changing.
- The
messagesannotation intranslations.ts. Without it, the composer is not assignable to I18nT'si18n?: Composer(TS2322 on the locale union), so the widening is needed.TranslationKeystill derives from the English JSON, and unknown keys compiled before too. cloud.spec.ts:68loose match. Pre-existing and untouched; noted under item 5.
Also checked with no finding: SSR of all five components in three locales (I18nT reads the i18n prop directly, no plugin needed); hydration of WireNodeLayout from server HTML with no mismatch warnings, against a control that does warn; a model name containing <b>, {name}, @: and | renders as literal text; nothing outside the catalogs reads any removed key; removing heroTitleAfter from knownViolations matches locale:check (0 violations).
Coverage
Depth standard. Lanes: correctness, maintainability, tests, each over all 12 files. Lenses: blast-radius, boundary-discipline, fix-root-causes, regression-risk blame (correctness); ponytail, no-comments (0 comments added, 1 removed, 5 left in edited blocks), reader-load, migrate-callers, type-system-discipline (maintainability); designing-test-suites, reviewing-unit-tests (partially applicable to apps/website), test-mock-quality-analysis (0 mocks in either file) (tests). No Playwright spec changed. Fallow: only the WireNodeLayout template complexity, unchanged from base. Run at 5b9f4bb: vitest on the changed tests plus about, careers, common (196 pass); mutation runs for keypath typos and slot renames; locale:check; base vs head SSR and hydration comparison; preview vs production text and screenshots. CI at this head is green. Existing feedback: CodeRabbit skipped the path and Codex left no findings, so nothing to confirm or contest.
| <I18nT | ||
| keypath="about.values.heading" |
There was a problem hiding this comment.
P2: three of the five converted headings have no test. keypath is typed string in vue-i18n 11.4.10, so typecheck can't catch a typo here. I renamed this slot to #hl, changed StorySection's keypath to about.story.headng, and changed WireNodeLayout's to about.careers.whyTitl. The heading then rendered "We believe , own their process…", the raw key about.story.headng, and the raw key twice in the careers card. The full website suite (11,805 tests) still passed, and no e2e spec asserts these headings.
One table covering every heading component catches all three. Each row would be { component, locale, heading, highlight } with the exact-text and highlight-element assertions AudienceSection.test.ts already makes, and AudienceSection can become a row in it. FE-3069's acceptance criteria ask for this component-level coverage too. Prose because the fix is a new test file.
There was a problem hiding this comment.
Added interpolatedHeadings.test.ts in a85eef5: one table over OurValuesSection, StorySection and AudienceSection in en and zh-CN (exact text plus the highlight as its own element), and a WireNodeLayout title-card case. Your three mutations, plus an Audience keypath typo, now each fail two rows. AudienceSection.test.ts folded into it.
| class="text-2xl font-light text-primary-comfy-canvas" | ||
| > | ||
| <template #logo> | ||
| <br /> |
There was a problem hiding this comment.
P3, question: the line break now lives inside the #logo slot, so it always renders right before the logo, wherever a locale puts {logo}. A translation that opens with the logo gets an empty first line in the card. That is a likely output. The ja whyTitle this PR replaces was machine-generated as "Comfyに参加する理由は?", brand first, and the next generation run will retranslate this key (see my comment on en/main.json). Can the break move out of the placeholder? For example, render the logo as its own line with CSS, or split the title so the break is not part of {logo}. The same applies to the copy at L316. Prose because where the break lives is a design call.
There was a problem hiding this comment.
Moved the break into its own {br} placeholder, so each locale places it: Why join{br}{logo}?, 为什么加入{br}{logo}?, なぜ{br}{logo}に参加するのか?. Rendering is unchanged for all three today.
| <I18nT | ||
| keypath="about.careers.whyTitle" | ||
| tag="p" | ||
| :i18n | ||
| class="text-2xl font-light text-primary-comfy-canvas" | ||
| > | ||
| <template #logo> | ||
| <br /> | ||
| <span | ||
| class="mb-0.5 inline-block h-5 w-16 bg-primary-comfy-yellow align-middle" | ||
| style="mask: url(/icons/logo.svg) no-repeat center / contain" | ||
| /> | ||
| </template> | ||
| </I18nT> |
There was a problem hiding this comment.
P3 Nit: this block is a 14-line copy of L176-189. The duplication was already there, but the diff grows each copy from 8 lines to 14, and the two have to change together (the keypath, the slot, the logo mask). The file already imports @vueuse/core. const [DefineWhyTitle, WhyTitle] = createReusableTemplate() lets the desktop and mobile layouts render one definition. Prose because the change touches both template sites and the script.
There was a problem hiding this comment.
Done: createReusableTemplate defines the title card once and both layouts render <WhyJoinTitle />.
| import { externalLinks } from '@/config/routes' | ||
| import { I18nT } from 'vue-i18n' | ||
|
|
||
| import { translationsFor } from '@/i18n/translations' |
There was a problem hiding this comment.
P3 Nit: the vue-i18n import sits between two @/ imports here. The other four migrated files put it in its own group above the @/ imports. No configured rule orders these, so lint passes either way.
| import { externalLinks } from '@/config/routes' | |
| import { I18nT } from 'vue-i18n' | |
| import { translationsFor } from '@/i18n/translations' | |
| import { I18nT } from 'vue-i18n' | |
| import { externalLinks } from '@/config/routes' | |
| import { translationsFor } from '@/i18n/translations' |
| const i18n = translationsFor(locale) | ||
| const { t } = i18n |
There was a problem hiding this comment.
P3 Nit: const i18n = translationsFor(locale) followed by const { t } = i18n repeats in all five files only to feed the :i18n shorthand. Keeping the base line const { t } = translationsFor(locale) and writing :i18n="translationsFor(locale)" on the I18nT drops the extra name and shrinks the diff. translationsFor returns a cached composer, so the second call costs nothing. Prose because the change spans the script and the template, and the same edit applies in OurValuesSection, StorySection, WireNodeLayout and ModelsHeroSection.
There was a problem hiding this comment.
Done in all five files.
| expect(title).toHaveTextContent(new RegExp(`^${heading}$`), { | ||
| normalizeWhitespace: false | ||
| }) |
There was a problem hiding this comment.
P3: the RegExp is built from the raw heading, so . matches any character. I changed the en catalog to "…control, and simplicity!" and both rows still passed. ^…$ with normalizeWhitespace: false amounts to exact text equality, which toBe states directly. I ran the replacement at this head and both rows pass.
| expect(title).toHaveTextContent(new RegExp(`^${heading}$`), { | |
| normalizeWhitespace: false | |
| }) | |
| expect(title.textContent).toBe(heading) |
There was a problem hiding this comment.
Applied (now toBe in the shared table).
| expect(title).toHaveTextContent(new RegExp(`^${heading}$`), { | ||
| normalizeWhitespace: false | ||
| }) | ||
| expect(within(title).getByText(highlight)).toHaveClass('text-white') |
There was a problem hiding this comment.
P2: toHaveClass('text-white') copies the production class string and uses it as a stand-in for "the highlight is styled", which docs/guidance/testing-principles.md asks us not to do. within(title).getByText(highlight) already fails when the highlight stops being its own element, because no element's full text is then "creators". The class check adds only a failure on a colour-token rename. I ran the replacement at this head and both rows pass.
| expect(within(title).getByText(highlight)).toHaveClass('text-white') | |
| expect(within(title).getByText(highlight)).toBeInTheDocument() |
There was a problem hiding this comment.
Applied; the table only asserts the highlight is its own element.
| { locale: 'zh-CN', title: 'ComfyUI 中的 Flux' }, | ||
| { locale: 'ja', title: 'ComfyUIのFlux' } | ||
| ] as const)( | ||
| 'places the brand where the $locale title puts it', |
There was a problem hiding this comment.
P3: the title says "places the brand", but the body only checks the heading text. I removed the yellow span from the #comfyui slot and all three rows still passed. One more line after the text assertion (L23-26) fails on that mutation 3/3 and passes at this head:
expect(screen.getByText('ComfyUI')).toBeInTheDocument()No other element in the section has the exact text "ComfyUI", so screen is enough here and within isn't needed. Prose because L23-26 is outside the diff hunk.
There was a problem hiding this comment.
Added; removing the span now fails all three rows.
| "whyTitle": "Why join Comfy?", | ||
| "whyTitleBefore": "Why join", | ||
| "whyTitleAfter": "?", | ||
| "whyTitle": "Why join{logo}?", |
There was a problem hiding this comment.
P3: this reuses about.careers.whyTitle, which already existed with the English "Why join Comfy?", so generation treats it as modified English. The README's Localization rule says that when English is modified, "generation replaces the matching eligible translations, including locale edits made in the same change". locale:check at this head lists about.careers.whyTitle as pending for zh-CN (the only zh-CN entry) and for ja. The next run will overwrite the hand-joined "为什么加入{logo}?" and "なぜ{logo}に参加するのか?" with machine output that only has to contain {logo} somewhere. The four other headings use new keys, so their joined copy is kept.
Nothing reads the old whyTitle on base, so a new key such as about.careers.whyJoinTitle (in all three catalogs and both keypaths in WireNodeLayout), with the dead whyTitle deleted, keeps the copy you verified. Prose because the change spans three catalogs and the component.
There was a problem hiding this comment.
Switched to a new about.careers.whyJoinTitle key and dropped whyTitle. locale:check no longer lists anything pending for zh-CN.
- Move the careers title to a new about.careers.whyJoinTitle key so the
hand-joined zh-CN and ja copy is kept instead of regenerated, and give
the line break its own {br} placeholder so locales that open with the
logo do not get an empty first line.
- Render the Why-join title card from one reusable template.
- Pass translationsFor(locale) straight to I18nT.
- Cover every highlighted heading in one table test that fails on keypath
typos and renamed slots, and assert the ComfyUI brand in the models hero.
- Document the I18nT slot convention in the website README.
There was a problem hiding this comment.
Verdict: one P2 test gap left (the careers card's logo and line break are untested); every other round-1 finding is addressed
a85eef5 addresses almost everything from my last review, and CI is green at this head. The new {br} placeholder, the whyJoinTitle key and the shared createReusableTemplate card hold up in practice. Server rendering at this head gives two title cards with the logo and break in en, zh-CN and ja, with no VueUse warnings. Hydrating that HTML produces no mismatch warnings, while a control with altered text does. The only difference in server HTML from 5b9f4bb is Vue's fragment comment markers. locale:check exits 0 with nothing pending for zh-CN. The open item is inline. The careers title-card case passes with the logo or the break removed, so the description's "renamed slots in each component fail it" does not hold for WireNodeLayout.
Round-1 findings at this head
- P2, three headings untested: addressed in
a85eef5for values, story and audience. A renamed#highlight, a keypath typo or an unwrapped highlight now each fail two rows. ForWireNodeLayout, a keypath typo and a missing mobile card now fail, but the slots don't (inline). - P2,
toHaveClass('text-white'): addressed ina85eef5. The table asserts the highlight is its own element. - P3,
about.careers.whyTitlereuse: addressed ina85eef5with the newwhyJoinTitlekey.whyTitleis gone from every catalog. - P3,
<br>inside#logo: addressed ina85eef5with the{br}placeholder. - P3, unescaped RegExp in the audience test: addressed. The table uses
toBe. - P3, models hero brand element: addressed. Unwrapping or renaming
#comfyuifails all three rows. - P3, FE-3069 overclaim: addressed. The description now says "Part of", names the three components still split, and the README documents the slot convention.
- P3, test count in the description: addressed (removed).
- P3 Nits (duplicated card,
AudienceSectionimport order,const i18ndestructure,WireNodeLayoutcomments): addressed ina85eef5. - P2, overlap with #20943: superseded. #20943 is closed.
- P3, overlap with #20585: still open, now with specifics. #20585 adds reviewed ja copy under the split keys this PR deletes:
about.careers.whyTitleBefore/After("なぜ" … "で働くのか"),about.story.heading*,about.values.heading*andcloud.audience.heading*. This head's ja catalog holds different joined copy ("なぜ{br}{logo}に参加するのか?", and different story and values wording). Whichever merges second has to port #20585's ja wording into the joined messages. Otherwise generation prunes the split keys and the reviewed copy is lost. No ja about or careers page is live yet. - P3, visual baselines: unchanged and optional. No golden covers the five headings. That matters less now that the rendered DOM is unchanged.
- Unverified, #20366 merge fit:
git merge-treeof this head and #20366 conflicts inapps/website/src/i18n/translations.ts. main and #20366 merge clean. #20366 keepstranslationsFor(locale)synchronous and returning acreateI18n().globalcomposer, so taking its side and droppingmessagesshould keep:i18n="translationsFor(locale)"working. Runningpnpm typecheck:websiteon that resolution settles it.
Checked and cleared
ModelsHeroSection.test.tsstill buildsnew RegExp(`^${title}$`). Those lines are unchanged from main, and none of the titles contains a metacharacter.interpolatedHeadings.test.tssits at thesrc/components/root and imports from three subfolders. That follows my round-1 request for one table across the heading components.StorySectionandOurValuesSectionkeep 12 template comments from main. The diff edits only the heading element, not the blocks those comments label.DefineWhyJoinTitleadds a template root, butWireNodeLayoutwas already multi-root and its callers pass only declared props, so attribute fallthrough is unchanged.
Coverage
Follow-up delta review of 5b9f4bb..a85eef5 (12 files, +141/-92) by one subagent applying the correctness, maintainability and tests briefs. Run at a85eef5: vitest on both changed test files (11 pass); 12 mutation runs (logo or br slot renamed, logo span deleted, br emptied, mobile card dropped, keypath typos, renamed highlight slots, unwrapped highlight and brand); SSR renderToString plus createSSRApp hydration at both heads, with a positive control; locale:check; git merge-tree against #20366. CI at this head: 50 success, 50 skipped, none failing. The author answered every round-1 thread, so I replied in none.
| it.for([ | ||
| { locale: 'en', title: 'Why join?' }, | ||
| { locale: 'zh-CN', title: '为什么加入?' } | ||
| ] as const)( | ||
| 'WireNodeLayout renders the $locale title card as one message', | ||
| ({ locale, title }) => { | ||
| render(WireNodeLayout, { props: { locale, reasons: [] } }) | ||
|
|
||
| const cards = screen.getAllByText(title) | ||
| expect(cards.map((card) => card.textContent)).toEqual([title, title]) | ||
| } | ||
| ) |
There was a problem hiding this comment.
P2: this case passes when the card loses its logo or its line break. The logo span and the <br> contribute no text, so getAllByText('Why join?') and the textContent check match either way. Each of these mutations still passes all 11 tests at this head: renaming #logo to #logoo, renaming #br to #brr, emptying the #logo slot, and emptying #br. A keypath typo and a dropped mobile card do fail. So the description's "Keypath typos and renamed slots in each component fail it" holds for values, story, audience and models but not here. The case also can't tell the joined message from a split implementation written without whitespace, so "as one message" claims more than it checks.
The logo has no accessible name either (same as main), so screen readers already read this card as "Why join?" with the brand missing. That is also why no semantic query can reach it. Giving the span role="img" and an aria-label from a new key (for example about.careers.logoLabel: "Comfy" in all three catalogs) fixes both. The case can then assert within(card).getByRole('img', { name: 'Comfy' }) for each of the two cards. The {br} slot is presentational. Either drop it from the description's claim or accept that it is untested. Prose because the fix spans WireNodeLayout.vue, three catalogs and this test.
| - Styled words inside a sentence are a named placeholder, never a message | ||
| split around them: `"Built for {highlight} who need quality."`, rendered with | ||
| `<I18nT keypath="…" :i18n="translationsFor(locale)">` and a `#highlight` | ||
| slot, so each locale decides where the styled word sits. |
There was a problem hiding this comment.
P3 Nit: the convention covers styled words, but not the rule that fixed the careers title: an inline image or a line break is its own placeholder too, never part of another slot. Without it, the next author can bake a <br> into a slot again and lock the break for every locale. One clause and the whyJoinTitle example cover it.
| - Styled words inside a sentence are a named placeholder, never a message | |
| split around them: `"Built for {highlight} who need quality."`, rendered with | |
| `<I18nT keypath="…" :i18n="translationsFor(locale)">` and a `#highlight` | |
| slot, so each locale decides where the styled word sits. | |
| - Styled words, inline images and line breaks inside a sentence are named | |
| placeholders, never a message split around them: `"Built for {highlight} who | |
| need quality."` or `"Why join{br}{logo}?"`, rendered with | |
| `<I18nT keypath="…" :i18n="translationsFor(locale)">` and one slot per | |
| placeholder, so each locale decides where each one sits. |
dante01yoon
left a comment
There was a problem hiding this comment.
question: apps/website no longer exists on main. #21861 (merged 2026-10-26) removed the monorepo copy now that comfy.org deploys from Comfy-Org/comfy-website, which is also why this PR now conflicts. Could it be ported there (and this one closed)?
I did read the change at a85eef55, so here is what carries over to the port:
- Turning the five split headings into single messages with named
I18nTslots is the right fix. The{br}placeholder lets each locale place the careers line break, which addresses the earlier brand-first concern. - No remaining reader of the removed keys (
*.headingBefore/After,about.careers.whyTitle*,models.list.heroTitleBefore/After) or of the removedtitleBeforeKey/titleAfterKeyprops. Bothsplitconsumers (WhyJoinSection,ValuesSection) use the defaults. This matters becauset()accepts any string, so typecheck would not catch a stale key. TranslationKeyis stillMessageKey<typeof en>, so loosening themessagesrecord type forI18nTkeeps key checking for typed key props.
No blocking findings on the change itself. I would approve it in the new repo.
Summary
Five website headings were translated as before/after fragments around a styled span (
about.values,about.story,cloud.audience, the careers "Why join" title, and the models hero), so every language had to break its sentence where English did. Each is now one message with named placeholders ({highlight},{logo},{br},{comfyui}), rendered through vue-i18n'sI18nTslots, so word order and the careers line break belong to the locale. The existing Chinese and Japanese copy is joined into the new messages, not retranslated; every heading uses a new key, so generation keeps that copy. The README documents the slot convention.I18nT'si18nprop only accepts an untyped composer, so the website catalogs are now typed as string-keyed message dictionaries.t()already accepted any string key, so only the locale-name typing ont's options is lost.Part of FE-3069 and FE-2045.
ReasonSection,ModelLaunchHeroSectionandReasonsSplit01still render split fragments and follow separately.Verification
interpolatedHeadings.test.tschecks the values, story and audience headings in en and zh-CN (exact text, highlight as its own element) and the careers title card. Keypath typos and renamed slots in each component fail it. The models hero test covers en, zh-CN and ja order and the ComfyUI brand element.locale:checkreports 0 protected-token violations and nothing pending for zh-CN.