Repository navigation
feat(website): name the installer on the download button - #19505
benceruleanlu merged 3 commits into
Conversation
|
The latest updates on your projects. Learn more about Vercel for GitHub. 2 Skipped Deployments
|
|
Navigate logical layers of code changes, visualize relationships, and explore their blast radius. Note Reviews pausedIt looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the Use the following commands to manage reviews:
Use the checkboxes below for quick actions:
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: Repository: Comfy-Org/ComfyUI_frontend/.coderabbit.yaml Review profile: ASSERTIVE Plan: Advanced Run ID: 📒 Files selected for processing (3)
Included review availability: This review used your included allowance. Your plan provides up to 8 included reviews per hour; 7 remain after this review. 📝 WalkthroughWalkthroughThe local download button now displays installer-specific labels. The hero groups the download and GitHub buttons. Component and browser tests cover installer links, labels, button alignment, and rendered text. ChangesWebsite installer downloads
Priority: ⬇️ Low Estimated code review effort: 3 (Moderate) | ~20 minutes Change: Feature Suggested reviewers: Merge Risk: ⚪ Minimal · up to The change identifies the selected installer and aligns the download and GitHub buttons. No actionable merge-blocking risk is established; merge after normal checks pass. Architecture SummaryArchitecture risk: 🔵 Low · up to The change affects 1 system. Changed systems: Architecture concerns Review detailsSystems and components
Before / after behavior
🚥 Pre-merge checks | ✅ 7✅ Passed checks (7 passed)
✨ Finishing Touches 💡 1🛠️ Fix failing CI checks 💡
📝 Generate docstrings
🧪 Generate unit tests (beta)
Comment |
🧩 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 🌐 Website E2ETip All tests passed.
🌾 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. No new findings in the changed files. Reproduce locally# the base the gate compared against
npx fallow audit --root . --quiet --format json \
--changed-since 3585660f7a0c662ae30173aadd47b0fe2c177158
Full docs: ADR-DEVEX-LINT-0015 🔗 Website PreviewWebsite Preview: https://comfy-website-preview-pr-19505.vercel.app Models is built and gated by PostHog (workshop-enabled), using staging Cloud. This commit: https://website-frontend-7wgus8yaz-comfyui.vercel.app Last updated: 2026-10-01T09:50:55Z for |
Codecov Report❌ Patch coverage is
❌ Your patch check has failed because the patch coverage (33.33%) is below the target coverage (80.00%). You can increase the patch coverage or adjust the target coverage. @@ Coverage Diff @@
## bl/download-installer-links #19505 +/- ##
============================================================
Coverage 84.53% 84.53%
============================================================
Files 3318 3318
Lines 160231 160231
Branches 41566 41565 -1
============================================================
Hits 135458 135458
Misses 22966 22966
Partials 1807 1807
Flags with carried forward coverage won't be shown. Click here to find out more.
🚀 New features to boost your workflow:
|
|
🔗 Commit SHA: 944f35c | Docs | View more details | Give us feedback! |
DrJKL
left a comment
There was a problem hiding this comment.
Verdict: request changes. The new second line breaks two website e2e specs that pass on the base branch, and one of them is a real font regression on /download.
CI: Website E2E failed at 1856af4 with two non-screenshot failures. Every website-e2e shard passed at the base 3030ad2.
Pre-merge blockers:
- P1
e2e/font-preload.spec.tsfails on /download withPP Formula 500 → /fonts/PPFormula-Medium.woff2. The installer line'sfont-mediumpulls in a face thatBaseLayout.astrodoes not preload, so underfont-display: optionalit renders in the fallback font on slow loads. Inline onDownloadLocalButton.vue:53. - P1
e2e/button-height.spec.tsfails withDOWNLOAD DESKTOP Windows x64, Expected: 20, Received: 36. Either the spec or the markup has to change. Inline onDownloadLocalButton.vue:51.
The rest are P2 and P3 inline comments on the tests and the composable's return shape.
Findings without an anchorable line
- P3 PR description. "The download button showed only an OS logo" overstates the old state. The base button already rendered
{{ label }}("DOWNLOAD DESKTOP") beside the logo (baseDownloadLocalButton.vue:78). What it lacked was the installer name. Something like "The download button named the platform only through its OS logo" would be accurate. - P3 PR description.
apps/website/AGENTS.mdasks for before and after screenshots on every visitor-visible change. The Screenshots section has a preview link and a prose description, and the preview cannot show the before state. One screenshot of each would also make the 52px to 68px height change easy to judge.
Existing findings confirmed or contested
None. The PR had no reviews or inline comments when I reviewed it. The bot issue comments carry no findings, and Fallow reported nothing new.
Checked and cleared
- Analytics still records the platform, not the installer.
captureDownloadClickgetsinstallers[key].platform, so the fallback buttons still sendwindowsandmacas the oldbtn.keydid. I ran a scratch component test that clicked both fallback buttons. data-astro-prefetch="false"now also applies to Windows ARM, since the check reads.platform. On the base branch the recognized-platform path also usedkey: platform.value, which waswindowsfor ARM, so nothing changed.- No consumer reads the removed
downloadUrlreturn.rgoverapps/website/srcfindsuseDownloadUrlonly inDownloadLocalButton,InstallerMenu, andMobileDownloadEmailForm. The droppedexternalLinks.githubfallback was unreachable, because the old template rendered it only whenplatformwas set. - All five installer label keys have
enandzh-CNentries, and a zh-CN render of the fallback gave spaced accessible names. - The layout claim holds from the classes. The inner
lg:flex-rowkeeps the defaultalign-items: stretch, so the GitHub button matches the download button's height. The outerlg:items-endstill bottom-aligns that row withMobileDownloadEmailForm.EcoSystemSectionalso stretches its sibling, andclasspassthrough reaches both fallback links. - The whitespace claim holds. Compiling the template shows sibling spans produce no text node between them, while the text-plus-nested-span form keeps the space. Swapping in sibling spans makes all three
DownloadLocalButton.test.tsrows fail withDOWNLOAD DESKTOPWindows x64, so the test does pin the spaced name. - The line counts in the change breakdown match
git diff --numstat(product +48/-69, tests +79/-5). navigatorstubs cannot leak between rows becauseapps/website/vitest.config.tssetsunstubGlobals: true.- The diff adds no
any, assertions, suppressions,dark:,!important, array classes, or untranslated template text.
Coverage
Depth standard. I ran five lanes: correctness, maintainability, idiomaticity, tests (including the Playwright spec procedure), and documentation, each over its own set of the six changed files and the consumers outside the diff (InstallerMenu, EcoSystemSection, BrandButton, posthog.ts, both download pages). At this head I ran vitest run on DownloadLocalButton.test.ts and useDownloadUrl.test.ts (26 passed), eslint and oxfmt --check on the changed files (clean), and pnpm typecheck:website (0 errors). I also ran three mutation experiments on the unit tests and playwright test --list e2e/download.spec.ts (45 tests found). I did not run Playwright locally. The two blockers come from the CI job logs for shards 1 and 2 of the run linked above.
| /> | ||
| <span class="text-left"> | ||
| {{ t('download.hero.downloadLocal', locale) }} | ||
| <span class="block text-xs font-medium tracking-normal"> |
There was a problem hiding this comment.
P1: font-medium renders PP Formula Medium (500) above the fold on /download, and nothing preloads that face. BaseLayout.astro preloads Regular, Light, Bold, Semibold, and NarrowSemibold, and download.astro passes no preloadFonts. Under font-display: optional, this line falls back to the system font on any slow load. e2e/font-preload.spec.ts catches it: shard 2 of the Website E2E run fails with "PP Formula 500 → /fonts/PPFormula-Medium.woff2", and the same spec passes on the base branch.
There are two fixes. Use a weight that is already preloaded (font-normal or font-semibold), or pass preloadFonts={['Medium']} from both src/pages/download.astro and src/pages/zh-CN/download.astro, as routes/models/index.astro does. The first adds no extra font request.
There was a problem hiding this comment.
Fixed in 167a761: the subtitle uses font-normal, whose Regular face is already preloaded. The font test now waits for the hydrated download button before measuring; all six font-preload cases pass.
| alt="" | ||
| class="inline-block size-5" | ||
| /> | ||
| <span class="text-left"> |
There was a problem hiding this comment.
P1: This two-line block makes BrandButton's wrapping <span class="flex items-center"> 36px tall while its computed line height stays 20px (text-sm). e2e/button-height.spec.ts measures exactly that span on every nowrap flex link on /download. Shard 1 of the Website E2E run fails with Error: DOWNLOAD DESKTOP Windows x64 ... Expected: 20, Received: 36, and the spec passes on the base branch.
The spec guards against a label box reserving descender space on a one-line label. A deliberate second line is a different case, so I think the spec needs to learn about it rather than the button dropping the line. For example, it could skip labels that contain a display: block child, or compare the height against the sum of the line boxes. Please update the spec's comment to match whichever way you go.
There was a problem hiding this comment.
Fixed in 167a761: the height check sums the visible text lines and still compares them with the enclosing label box. It waits for the download button first. The check passes with the intended two-line label and fails with +4px excess when the original inline-block spacing regression is restored temporarily.
| const offered = computed<Installer[]>(() => { | ||
| if (installer.value) return [installer.value] | ||
| if (showFallback.value) return ['windows', 'macArm'] | ||
| return [] | ||
| }) |
There was a problem hiding this comment.
P3: offered holds keys, so the template goes back to the table on every binding: installers[key].url, .platform twice, ICONS[installers[key].platform], and .label. If useDownloadUrl returned the selected entry (installers.windowsArm and so on) and this computed held entries ([installers.windows, installers.macArm] for the fallback), the template could read entry.url, entry.platform, and entry.label directly. The Installer key type would then have no consumer. :key could use entry.url.
There was a problem hiding this comment.
Fixed in 167a761: useDownloadUrl returns the selected installer entry, and the button list holds entries directly. URL, platform, and label now come from that entry.
| } | ||
| } as const satisfies Record<string, DesktopInstaller> | ||
|
|
||
| export type Installer = keyof typeof installers |
There was a problem hiding this comment.
P3 Nit: Installer is keyof typeof installers, but the entry shape a few lines up is called DesktopInstaller. So computed<Installer | null> reads as if it returns an installer, when it returns an index into the table. If the key stays (see the comment on DownloadLocalButton.vue), InstallerKey would say what it holds.
There was a problem hiding this comment.
The key type is gone in 167a761. The computed value now returns a DesktopInstaller entry or null.
| }) | ||
|
|
||
| return { downloadUrl, platform, showFallback, isMobileUa } | ||
| return { installer, platform, showFallback, isMobileUa } |
There was a problem hiding this comment.
P3: With installer returned, platform has one remaining consumer. InstallerMenu.vue:25 uses it only as v-if="platform || showFallback", and installer is non-null exactly when platform is. Switching that check to installer || showFallback would let this return drop platform and keep a single representation of the detected choice.
There was a problem hiding this comment.
Removed platform from the public return in 167a761. InstallerMenu already shares its parent button gate after the dropdown cleanup, so it no longer runs device detection or consumes platform.
|
|
||
| import DownloadLocalButton from './DownloadLocalButton.vue' | ||
|
|
||
| vi.mock(import('../../../scripts/posthog')) |
There was a problem hiding this comment.
P2: This mocks posthog but no row clicks a link or checks captureDownloadClick. This PR changed the argument from btn.key to installers[key].platform. I hard-coded captureDownloadClick('windows') in the component and all three rows still passed. One case that clicks a non-Windows link (the Linux row, or the fallback's macOS button) and asserts expect(captureDownloadClick).toHaveBeenCalledExactlyOnceWith('linux') would pin the analytics platform. The table can stay as it is for names and URLs.
There was a problem hiding this comment.
Added a Linux click case in 167a761 that asserts captureDownloadClick is called exactly once with linux. The focused component/composable suite passes all 37 cases.
| { | ||
| label: 'Linux', | ||
| userAgent: UA.linux, | ||
| buttons: [ | ||
| [ | ||
| 'DOWNLOAD DESKTOP Linux x64 (AppImage)', | ||
| 'https://download.comfy.org/linux/appimage/x64' | ||
| ] | ||
| ] | ||
| }, |
There was a problem hiding this comment.
P3: No test here or in useDownloadUrl.test.ts feeds a Mac user agent, so nothing checks that useDownloadUrl.ts:114 picks macArm. The fallback's macOS button is hard-coded, so it does not cover that branch. I changed that line to return 'windows' and all 26 tests in both files still passed. A Mac row in this table (Macintosh UA, maxTouchPoints: 0, expecting DOWNLOAD DESKTOP macOS (Apple Silicon) and the dmg URL) would cover it.
There was a problem hiding this comment.
Added a Macintosh UA row with zero touch points in 167a761. It asserts the macOS installer name and arm64 DMG URL through the rendered button.
| {{ t('download.hero.installGithub', locale) }} | ||
| </span> | ||
| </BrandButton> | ||
| <div class="flex flex-col gap-4 lg:flex-row"> |
There was a problem hiding this comment.
P3: This wrapper exists so the GitHub button stretches to the taller download button while the outer lg:items-end keeps the email form bottom-aligned. From the classes it works. But no test would fail if someone removed the wrapper or the outer lg:items-end, because download.spec.ts's desktop hero test checks only visibility and hrefs. A bounding-box check in that test at the desktop viewport would guard it: equal heights for the two links, and matching bottom edges for the row and the form.
There was a problem hiding this comment.
Added rendered height and bottom-edge equality assertions for the desktop download and GitHub links in 167a761. The current device gate shows the email form only on phones, where the desktop download button is absent; that existing behavior remains covered. All 31 focused desktop/mobile download, font, and height checks pass.
There was a problem hiding this comment.
Addressed in 681e95c: the Windows desktop hero test now reads both links' bounding boxes and asserts the GitHub button has the same height and bottom as the download button (download.spec.ts:132-140).
1856af4 to
051a434
Compare
051a434 to
167a761
Compare
167a761 to
681e95c
Compare
DrJKL
left a comment
There was a problem hiding this comment.
Verdict: ready to approve. Both P1s and every other finding from my first review are fixed at 681e95c. Three P3 comments and one question remain, all non-blocking.
I reviewed 167a761 and then its rebase, 681e95c. git range-diff shows the rebase adds only whitespace-normal on the installer subtitle. At 681e95c I rebuilt the site and ran button-height, font-preload, and download.spec.ts on the desktop and mobile projects: 32 passed. The /download overflow guards in visual-responsive.spec.ts also passed (4 of 4). Website E2E in CI was still running when I posted.
Non-blocking items (inline):
- P3
download.spec.ts:132-140takes two separate layout snapshots and compares them as exact floats. The same file already has a retryingexpect.poll+boundingBox()pattern for this. - P3 The equal-height check never runs on the two-button fallback, the layout that wraps at 1024 and 1280px.
- P3 Nit:
labelinDownloadLocalButton.vuenow has one use. - Question:
whitespace-normalmakes the subtitle wrap to two lines atlgwidths below 1440px, so the buttons become 84px tall there. Is that intended?
Findings without an anchorable line
- P3 The commit message is stale in three places, even though the PR body is now accurate. It says the button "showed only an OS logo", but the base already rendered DOWNLOAD DESKTOP beside the logo. It gives "Windows x64" as an example, but the rendered text is "Windows x64 (including Snapdragon)". It says
useDownloadUrl"returns the chosen installer key", but it now returns the installer entry. This only matters if the commit message survives the merge.
Checked and cleared
- The
button-heightspec still catches the regression it was written for. WithBrandButton.vue's label box set back toinline-block(the markup #19023 fixed), it fails withexcess: 4on both INSTALL FROM GITHUB buttons. The two-line download button reads 0 under that mutation. I think that is correct: its inline-flex child extends further below the baseline than the strut does, so the strut adds no height. - I checked a claim that the new line counter treats same-line text in two font sizes as two lines, and so hides extra height. That does not happen. With the subtitle made inline, the count doubles, but the spec fails with
excess: -16because it requires exactly 0. - Mutation checks on the unit tests. Hard-coding
captureDownloadClick('windows')fails the Linux click test. Returning the Windows entry for macOS fails the macOS row. Removing the whitespace between the label and the subtitle fails 4 rows, and removing the subtitle fails 5 tests. - No consumer reads the removed
downloadUrlorplatformreturns.InstallerMenureads only the staticinstallerstable.cn()produceslg:p-4 flex-1 rounded-r-none lg:min-w-0on the split button, so the wrapper'slg:min-w-60sets the combined width. - Analytics still records the platform, and
data-astro-prefetchbehaves as on the base branch. The zh-CN and English accessible names both come from the same keys thatInstallerMenuuses. - The diff adds no
any, assertions, suppressions,dark:,!important, array classes, or untranslated text.
Coverage
Depth standard. I ran correctness, maintainability, idiomaticity, tests (with the Playwright spec procedure), and documentation lanes against 167a761, then checked the 681e95c delta myself. At 681e95c: vitest run on DownloadLocalButton.test.ts, InstallerMenu.test.ts, and useDownloadUrl.test.ts (37 passed), eslint and oxfmt on the changed component (clean), an astro build, and the Playwright runs above. At 167a761: pnpm typecheck:website (0 errors) and eslint and oxfmt on all 9 changed files. I measured the layout with a scratch Playwright script at 390, 1024, 1280, and 1440px, using a Windows UA and an unrecognized UA.
Review record
Round 1: review at 1856af4, request changes.
- P1 Subtitle used
font-medium, a face that is not preloaded (font-preload.specfailed). Addressed in681e95cwithfont-normaland a hydration wait in the spec. Author reply. The spec passes locally. - P1 The two-line label failed
button-height.spec(36px against 20px). Addressed in681e95c: the spec now sums the visible text lines. Author reply. It passes locally and still fails on the #19023 regression. - P3 The composable returned a key and the template re-indexed
installers. Addressed in681e95c: it now returns the entry. Author reply. - P3 Nit:
Installernamed a key union. Addressed in681e95c: the alias is removed. Author reply. - P3
platformreturn was redundant. Addressed in681e95c: it is removed from the return. Author reply. - P2 Click analytics was never asserted. Addressed in
681e95cwith a Linux click test. Author reply. Mutation-proven. - P3 No Mac case. Addressed in
681e95cwith a Macintosh row. Author reply. Mutation-proven. - P3 The hero stretch had no falsifier. Addressed in
681e95cwith a height and bottom comparison in the Windows hero test. My reply. The follow-ups are the P3 inline comments in this review. - P3 The PR description said "only an OS logo". Addressed by the rewritten PR body.
- P3 No before and after screenshots. Addressed: the PR body now has one before image and two after images.
Round 2 (this review) at 681e95c: three P3 inline comments, one question, and one P3 on the commit message, all non-blocking.
Checks at 681e95c: unit tests 37 passed; eslint and oxfmt clean; website build succeeded; Playwright button-height, font-preload, and download 32 passed on the desktop and mobile projects; /download overflow guards 4 passed. CI Website E2E was running when I posted. Typecheck: 0 errors at 167a761, and 681e95c changes only a class string.
| const { downloadUrl, platform, showFallback } = useDownloadUrl() | ||
| const { installer, showFallback } = useDownloadUrl() | ||
|
|
||
| const label = computed(() => t('download.hero.downloadLocal', locale)) |
There was a problem hiding this comment.
P3 Nit: This PR removed the fallback aria-label, so label is now used only at line 75. It no longer does anything beyond aliasing one t() call. Rendering {{ t('download.hero.downloadLocal', locale) }} inline, as the subtitle already does, would let you drop both this computed and the computed import.
There was a problem hiding this comment.
Addressed in 937043cb: inlined the translated primary label. The computed import remains because buttons still uses it. I also rewrote the commit message to describe the existing Download Desktop text, the translated installer entry, and the final API accurately.
| const downloadBox = await downloadBtn.evaluate((element) => { | ||
| const { height, bottom } = element.getBoundingClientRect() | ||
| return { height, bottom } | ||
| }) | ||
| const githubBox = await githubBtn.evaluate((element) => { | ||
| const { height, bottom } = element.getBoundingClientRect() | ||
| return { height, bottom } | ||
| }) | ||
| expect(githubBox).toEqual(downloadBox) |
There was a problem hiding this comment.
P3: This takes two separate getBoundingClientRect() snapshots and then compares them with an exact, non-retrying toEqual. The two links sit in one stretch row, so at any given moment they match. The risk is a layout change between the two evaluate calls, for example a late font swap. The mobile check at the bottom of this file already polls boundingBox(). Something like this would retry and ignore sub-pixel noise:
await expect
.poll(async () => {
const [download, github] = await Promise.all([
downloadBtn.boundingBox(),
githubBtn.boundingBox()
])
return (
!!download &&
!!github &&
Math.abs(download.height - github.height) < 0.5 &&
Math.abs(download.y + download.height - (github.y + github.height)) < 0.5
)
})
.toBe(true)There was a problem hiding this comment.
Addressed in 937043cb: both rectangles are now read in one evaluateAll callback, with a retrying assertion and a 0.5px tolerance for height and bottom alignment. The same helper covers the fallback row.
|
|
||
| const macBtn = hero.getByRole('link', { | ||
| name: /DOWNLOAD DESKTOP — macOS/i | ||
| name: 'DOWNLOAD DESKTOP macOS (Apple Silicon)' |
There was a problem hiding this comment.
P3: The equal-height check runs only in the Windows test. This fallback is a different flex shape: two download wrappers plus GitHub in the new inner row, with the installer menu attached to the first. It is also the shape that wraps. With this UA, the Windows subtitle is two lines and all three buttons are 84px at both 1024 and 1280px. The same height assertion as the Windows test, applied to both download links against GitHub, would cover it.
There was a problem hiding this comment.
Addressed in 937043cb: added table-driven fallback cases at 1024px and 1280px. After hydration, each compares both download buttons and GitHub in the same layout snapshot. All 34 desktop/mobile download, font-preload, and button-height checks pass without retries; the 37 focused unit tests also pass.
| <span class="text-left"> | ||
| {{ label }} | ||
| <span | ||
| class="block text-xs font-normal tracking-normal whitespace-normal" |
There was a problem hiding this comment.
Question (P3): whitespace-normal lets the subtitle wrap, and it does wrap at lg widths, not only on narrow screens. I measured the built page with a scratch Playwright script:
| UA | 390px | 1024px | 1280px | 1440px |
|---|---|---|---|---|
| Windows | 1 line, 52px | 2 lines, 84px | 1 line, 68px | 1 line, 68px |
| Unrecognized (fallback) | 1 line, 52px | 2 lines, 84px | 2 lines, 84px | 1 line, 68px |
So on a 1280px laptop with the fallback, all three hero buttons are 84px tall, and "Windows x64 (including Snapdragon)" breaks after "(including". If that is intended, fine. If the goal was only to stop overflow on small screens, lg:whitespace-nowrap or a shorter label would keep the desktop row at 68px.
There was a problem hiding this comment.
Yes, allowing the installer name to wrap at constrained desktop widths is intentional. Keeping the full compatibility label visible prevents the clipping found at 320px; the adjacent GitHub button stretches with it. The fallback's 1024px and 1280px layouts now have explicit alignment coverage. I kept that behavior and clarified the PR description rather than forcing a desktop no-wrap rule.
681e95c to
937043c
Compare
DrJKL
left a comment
There was a problem hiding this comment.
Verdict: ready to approve. 937043c addresses every open item from my second review, and I have no new findings.
Compared with 681e95c, this head inlines the primary label, rewrites the commit message, and adds expectAlignedButtons. That helper reads every matched button's rectangle in one evaluateAll, then polls until the height and bottom spread is under 0.5px. It now runs in the Windows hero test and in new fallback cases at 1024 and 1280px. I confirmed it catches a regression: with lg:items-start added to the hero's inner row, the Windows test and both fallback cases fail (spreads of 14 and 30px). With that mutation reverted, button-height, font-preload, and download.spec.ts pass 34 of 34 on the desktop and mobile projects. Website E2E in CI was still running when I posted.
Review record
Round 1: review at 1856af4, request changes.
- P1 Subtitle used
font-medium, a face that is not preloaded (font-preload.specfailed). Addressed withfont-normaland a hydration wait in the spec. Author reply. - P1 The two-line label failed
button-height.spec. Addressed: the spec now sums the visible text lines, and it still fails (+4px) on the #19023inline-blockregression. Author reply. - P3 The composable returned a key and the template re-indexed
installers. Addressed: it now returns the entry. Author reply. - P3 Nit:
Installernamed a key union. Addressed: the alias is removed. Author reply. - P3
platformreturn was redundant. Addressed: it is removed. Author reply. - P2 Click analytics was never asserted. Addressed with a Linux click test, which fails when the platform is hard-coded. Author reply.
- P3 No Mac case. Addressed with a Macintosh row, which fails if macOS gets the Windows installer. Author reply.
- P3 The hero stretch had no falsifier. Addressed with the equal-height assertion. My reply.
- P3 The PR description said "only an OS logo". Addressed by the rewritten PR body.
- P3 No before and after screenshots. Addressed: the PR body now has them.
Round 2: review at 681e95c, ready to approve with nits.
- P3 Nit:
labelcomputed had one use. Addressed in937043c: it is inlined. Author reply. The author also pointed out that my suggestion to drop thecomputedimport was wrong, becausebuttonsstill uses it. - P3 The equal-height check took two snapshots and compared exact floats. Addressed in
937043cwith oneevaluateAllper poll and a 0.5px tolerance. Author reply. - P3 The fallback layout had no equal-height check. Addressed in
937043cwith cases at 1024 and 1280px, which fail under thelg:items-startmutation. Author reply. - Question: does the subtitle wrap at
lgwidths on purpose? The author answered yes: wrapping keeps the full label visible and avoids clipping at 320px, and the GitHub button stretches with it. Author reply. I accept that. - P3 The commit message was stale. Addressed in
937043c: the message now describes the entry API and the existing Download Desktop text. - I withdrew one candidate before posting round 2: that the new line counter could hide extra height when one line mixes font sizes. Running it showed the spec fails (
excess: -16) instead of passing.
Checks at 937043c: 37 unit tests passed in DownloadLocalButton.test.ts, InstallerMenu.test.ts, and useDownloadUrl.test.ts. eslint and oxfmt are clean on the two changed files. The website build succeeded. Playwright button-height, font-preload, and download passed 34 of 34 on desktop and mobile, and the alignment helper fails under the lg:items-start mutation. In CI, lint, typecheck, build, test, and fallow passed. Website E2E and website unit were still running when I posted.
The merge-base changed after approval.
99b13e3 to
e6ac6da
Compare
e6ac6da to
49aa253
Compare
Show the selected installer below Download Desktop using its translated menu label. Return the installer entry from useDownloadUrl so its URL, platform and label stay together. Keep fallback downloads labeled and stretch the adjacent GitHub button to match wrapped subtitles. Cover installer selection, analytics, font preloads and button alignment, including the unrecognized-desktop fallback at laptop widths.
49aa253 to
944f35c
Compare
things got fixed but i don't actually know where this review is right now
Summary
Download buttons name the exact installer they open, with matching accessible names and aligned desktop hero buttons.
Changes
Review Focus
Stacked on #19364. Verify the primary and fallback button labels and their alignment on
/download/.Before the final restack, validation passed 14 component tests and 24 focused browser tests against a fresh website build. After removing the checkout prerequisite and integrating both installer fixes, the final stack passes 36 focused installer/composable unit tests. CI has not been checked after this final restack. Controlled layout checks reject descender-space and icon-only excess height while accepting quarter-pixel differences; overriding the accessible names makes all four platform table cases fail.
Screenshots (if applicable)
Preview