Skip to content

fix: the comments round on Drop's own pages: one comment box, ViteHub's thread card, relative times, shared copy buttons, a doc header that copies at once, who dropped it, the folder's drops, and comark's side outline - #228

Open
onmax wants to merge 28 commits into
mainfrom
fix/drop-comments-round

Conversation

@onmax

@onmax onmax commented Nov 17, 2026 •

Copy link
Copy Markdown
Member

The comments left on Drop's own pages (#1 to #14 in the comments list) point at the same few problems. The new-comment box drew a frame inside its bubble, its wrapped lines started after the + button, the floating thread card cropped images to 4:3, saved areas were blue, times only read "3 min. ago" or a UTC date, copy buttons snapped back without motion, the pill's + was blurry at 1x, the toolbar could show you to yourself, the doc header's Copy was slow and changed width, "Dropped by" named the MCP client, the folder was plain text, and the outline stopped short of the bottom and never followed the section being read. Most of the fixes landed in ViteHub and comark-ui. This PR picks them up and deletes Drop's copies.

Upstream

Comments and the pill

Times and copy buttons

The doc header (Drop comments #6, #8, #9, #12)

  • Copy is built once and toggles data-copied and data-pending, with ViteHub's icon swap and the label fixed at "Copy". Before, every header message rebuilt it, so it changed width, lost focus, and could lose a press.
  • The Markdown loads once per version when the pointer or focus reaches Copy (doc-intent, answered with doc-markdown), and the click writes it from the frame. A click before it has arrived writes a promise of it. It says "Markdown copied" to screen readers, shows "Couldn't copy" in the frame on failure, and sends the error to the client-error log. The M shortcut and the menus copy the same text, and show the check on the button too.
  • "Dropped by" reads [avatar] Max · [logo] Claude Code, with "via Executor", the model and the reasoning level on hover. dropDetail now carries owner { name, image }, from the owner row the lookup already joins.
  • Signed-in readers can open the folder: a list of its newest drops with this one marked and focused, then Open folder. It's asked for on hover (doc-folder, answered with doc-folder-items) from GET /api/drops?folder=&limit=.
  • The save menu and the folder list are popover="auto" with popovertarget and CSS anchor positioning, so Escape, outside clicks and focus return are the browser's. The arrow keys stay. An Escape a popover or modal dialog takes no longer reaches Drop.

Why Copy failed in Firefox (#12)

Still unknown. Clipboard writes work from the doc's frame in Chromium and in Firefox 157 and 160 (stock), from an opaque-origin srcdoc frame and from a frame with its own origin, written at once, after 1.5 s, through a ClipboardItem promise of 1.5 s or 6.5 s, and for 3 MB of text. Main's own header copied in stock Firefox 160 too: a plain click, a hover first, a parent textarea focused first, and a double click. What's left is the macOS build or an extension, which I couldn't run. Production's client-error log should say more: with this PR, a refused write in the header reports its error name and message there.

One question before merge: dropDetail.owner (name and avatar) goes to everyone who can open the drop, public-link readers included, so a public doc shows who dropped it. Keep that, or show it only to signed-in members?

Saved areas are grey (Drop comment #1) and the pill's card matches Drop's (#2) through the ViteHub bump. #3 and #4 were already live.

The outline (Drop comment #5)

  • A Markdown doc's outline is comark-outline's side outline (fix(outline): run the side outline to the bottom of the window and keep the current link in view onmax/comark-ui#452), and Drop's OUTLINE_SCRIPT is gone. It runs to the bottom of the frame, marks and scrolls to the section being read, fades while more sits below, and makes the last heading current at the end of the page.
  • Only a reader's render adds it. The linter and the live editor parse without it, so it never reaches a doc's source or decides whether a doc is editable. comark-outline is a direct dependency at 5ed5886, like comark-properties and comark-time, and its script joins RENDERER_SCRIPTS so the /f/ pages' CSP allows it. The markdown cache integrity moves to drop.26.
  • --drop-inset reaches it as --comark-inset-end, and a wide block keeps clear of its column through --comark-rail-width, which it sets with data-comark-rail. The top stays at comark's default, 3.5rem, which was Drop's. While review comments hold the margin (data-drop-rail="open"), the article keeps that room as a transparent border instead of padding, so the outline lands past the cards as before. A click on its links still tells the viewer, so the address keeps the section.

Native first

  • Textareas grow with field-sizing: content between a min-height and a max-height in lh, so patches/@nuxt__ui@4.11.0.patch and its pnpm-workspace.yaml entry are gone. That was the patch's only hunk. Checked in Chromium, Firefox and WebKit: 3 lines empty, growing with the text, scrolling past the cap.
  • The drops list's Yesterday and This week count back calendar days with setDate, so a day with a daylight-saving change no longer moves them by an hour. app/utils/day-groups.ts has the rule and a test in Madrid's two change days.
  • review.js writes thread times with Intl.RelativeTimeFormat (narrow, English): "now", "5m ago", "yesterday", "2d ago". review.js has no version key; its one-hour Cache-Control picks it up.
  • AGENTS.md gets the browser support floor (latest two Chrome and Firefox, Safari 27.2 and 27.1) and the native-first rule, short, after docs: write down the browser support floor and the native-first rule onmax/comark-ui#458.

Plan: https://drop.vitehub.dev/d/q8p8asvvphav/drop-comments-plan

Before/after, Chromium and Firefox at 1x: https://drop.vitehub.dev/d/wb53k6p67j9n/before-after

Checked with pnpm typecheck, pnpm run doctor, pnpm knip, lint and format on the changed files, and pnpm install --frozen-lockfile. The Cloudflare preset builds with ViteHub's new driver names, and the built Worker under wrangler dev served a doc page with the side outline, the drops list, and email sign-up in Chromium and Firefox with no page errors. The full vp test run ran on a Slimbook at load 50: the failures were 5 s and 30 s timeouts in files this PR doesn't touch, plus this PR's Firefox copy test, which passes on its own. New browser tests cover a pin on a zoomed screenshot and on the page after it closes (Chromium and Firefox), and the header's copy, menu, Dropped by and folder list. CI now installs Playwright's Firefox as well as Chromium, so the Firefox tests run there. The screenshots come from local VPS builds of main and this branch.

🤖 Generated with Claude Code

…with the open fixes as patches

ViteHub's main renamed the backend keys, so the D1 database and the Cloudflare and memory Realtime use `driver: "cloudflare"` / `"memory"`, and Vercel Blob is `driver: "vercel"`. The fixes still open upstream ride along as patches built from their heads on the pinned main: vite-hub/vitehub#4785 and #4788 in @vite-hub/ui, #4794 in @vite-hub/comments, onmax/comark-ui#453 in comark-time and comark-properties. ViteHub Content now renders with the same comark-ui as Drop, so one copy carries the patches.
…, a crisp +, and each person once in the toolbar

The new-comment bubble is the box's frame, so the box passes `frame="none"` and the bubble shows the focus: the bordered one takes the accented border, the shadow-ring one a 22% ring (Drop comment #14). The dock's box and thread pass `size="sm"`, which brings the reply field, the 14px rows and the image shape from ViteHub, so the copies of those styles go (Drop comments #2, #7). The + of the pill is an inline SVG snapped to whole pixels while closed (Drop comment #13). The comments toolbar on Drop's own pages leaves out its faces and cursors, since the review dock already shows who's here (Drop comment #11). The frame's pop-out timing comes from ViteHub's `commentLeaveMs`.
… row reads relative

`TimeAgo` goes for ViteHub's `RelativeTime`: relative under 3 weeks, then the date, with the full date in a tooltip, short words in the lists. app.vue gives it Drop's one clock, so the words and Drop's countdowns move together (Drop comment #10). The copy buttons are ViteHub's `CopyButton`, whose icons cross-fade both ways, on Drop's 1.4 s (`DropCopyButton`); Drop's `icon-in` animation goes (Drop comments #6, #12).

The Updated row writes the version's full instant in comark-time's markup, and a box that had no date of its own now gets the relative-dates script too. The CSP allows comark-properties' `scripts` as they are, and the markdown cache integrity moves to drop.24.
…s render stable, and follow ViteHub's renamed drivers in tests

The parts' stylesheet loads only with the viewer, so DropCopyButton brings @vite-hub/ui/icon-swap.css itself. The Updated row's tooltip id comes from the version's time instead of a random one, so a warm render matches a cold one. The blob and rate limit tests use the `cloudflare` and `memory` drivers, and the area test checks ViteHub's grey saved areas (Drop comment #1).
…ed, and comark-ui main a410d81

vite-hub/vitehub#4788 (the 130 ms pop-out and commentLeaveMs) and #4794 (overlay presence: false, one face per person) merged, so their patches go and @vite-hub/comments runs unpatched. The @vite-hub/ui patch now carries #4785 alone, built on 20ab822. comark-properties' patch follows #453's latest head. comark-ui's pin is #456's head, the same tree as its main.
# Conflicts:
#	package.json
#	pnpm-lock.yaml
#	server/utils/markdown-cache.ts
…gs have, since it starts two Vite servers and a browser
…ded, and drop the comark-time and comark-properties patches

5ed5886 is #458's head, the newest commit pkg.pr.new published; its packages match main 1ef8c09, since pkg.pr.new skips pushes to main.
… Escape a popover takes stays in the page

comark-ui#457 opens the zoom with showModal(), which made Drop's comment layer inert: no pin could land on a zoomed screenshot. The layer moves into the dialog while it's open, hides the page's own marks there, and returns on close.
…son and their agent, and the folder opens a list of its drops

The header's Copy is built once and toggles data-copied and data-pending with ViteHub's icon swap, so a press is never lost to a rebuild. Its Markdown loads once per version when the pointer or focus reaches it (doc-intent), and the click writes it in the frame; it says Markdown copied, or Couldn't copy in the frame, and reports a refusal to the client-error log. The shortcut and menus copy the same text, and show on the button too.

Dropped by reads [avatar] Max · [logo] Claude Code, with what it came through, the model and the reasoning level on hover (dropDetail now carries owner). Signed-in readers get the folder's newest drops in a popover (GET /api/drops?folder=&limit=). The save menu and the folder list are native popovers placed with anchor positioning.
@onmax onmax changed the title fix: the comments round on Drop's own pages: one comment box, ViteHub's thread card, relative times, shared copy buttons, outline, folder popover, and who dropped it fix: the comments round on Drop's own pages: one comment box, ViteHub's thread card, relative times, shared copy buttons, a doc header that copies at once, who dropped it, and the folder's drops Nov 17, 2026
@onmax onmax changed the title fix: the comments round on Drop's own pages: one comment box, ViteHub's thread card, relative times, shared copy buttons, a doc header that copies at once, who dropped it, and the folder's drops fix: the comments round on Drop's own pages: one comment box, ViteHub's thread card, relative times, shared copy buttons, a doc header that copies at once, who dropped it, the folder's drops, and comark's side outline Nov 18, 2026
@onmax
onmax marked this pull request as ready for review November 18, 2026 00:11
@chatgpt-codex-connector

Copy link
Copy Markdown
Codex Review Summary: ❌ Failed

This comment shows the latest Codex review activity on this pull request.

Review Status Commit Review trigger
📝 Code Review ❌ Failed 2026-11-18T00:11:36.651675Z
We couldn’t complete this review. Comment @codex review to try again.
375c787 Draft marked ready
ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

@chatgpt-codex-connector

Copy link
Copy Markdown

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

@pullfrog pullfrog Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Important

The folder picker can navigate to the previous folder after a move, and zoomed Marker feedback loses its on-page stroke preview after Done. Both were reproduced in the built app.

Approach: sound — consolidating on ViteHub and Comark removes duplicated UI behavior, but the new persistent header state and zoom visibility rules need the fixes below.

Reviewed changes — reviewed the full PR, including upstream integration, server responses, frame messaging, and browser regressions.

  • Upstream integration: upgrades ViteHub and Comark, updates backend driver names, and adds the RelativeTime patch.
  • Comment UI: adopts shared small composers and thread styling, removes the pin composer's inner frame, and adjusts toolbar presence and icon rendering.
  • Times and copying: replaces local time/copy controls, adds calendar-day grouping across DST, and writes document-header copies from the frame with preloading and error feedback.
  • Document facts and folders: exposes owner profiles, displays person/agent attribution, and adds a limited folder listing through native header popovers.
  • Outline and zoom: replaces the local outline with Comark's reader-only rail, updates CSP/cache integration, and relocates the comment layer into modal zoom dialogs.
  • Native controls and coverage: replaces textarea autoresizing with CSS field sizing and adds Firefox/header/zoom regression coverage.

Validation: typecheck, both focused test runs (47 and 265 passing tests), the default Cloudflare build, and the VPS/D1 HTTP build passed. Browser checks also covered the reviewed-site popover and HTML-frame tooltip comment paths.

Pullfrog  | Fix all ➔ | Fix 👍s ➔ | View workflow run | Using gpt-6.1-sol(xhigh) | 𝕏

Comment thread shared/doc-header.ts Resolved
Comment thread shared/plan-runtime.ts Resolved

@pullfrog pullfrog Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

ℹ️ No new P0 or P1 findings in this delta; approval remains pending the existing unresolved review feedback.

Reviewed changes — reviewed 13645b8, the dependency update since the previous review at 375c787.

  • Upgraded ViteHub: moved the application packages, workspace overrides, and the Nuxt package's Forge dependency to ccf6574, checking the affected UI, auth, migration, and deployment contracts.
  • Removed the local RelativeTime patch: verified that the upstream component and clock-provider exports replace the patched implementation without requiring changes to Drop's callers.

Validation passed: frozen-lockfile installation and postinstall, typecheck, Cloudflare and VPS/D1 HTTP builds, 101 focused server/deployment tests, 23 browser regression tests, and the Nuxt package's seven tests plus packaging/type checks. Two environment-gated runtime tests were skipped.

Pullfrog  | Fix it ➔ | View workflow run | Using gpt-6.1-sol(xhigh) | 𝕏

… drawn on a zoom stays visible while the comment is written
# Conflicts:
#	package.json
#	pnpm-lock.yaml
#	pnpm-workspace.yaml

@pullfrog pullfrog Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

ℹ️ No P0 or P1 findings remain. One P2 regression is noted inline.

Reviewed changes — reviewed 8c1df87 since the previous Pullfrog review at 13645b8.

  • Keyed the folder picker by folder ID: verified that an in-place move refreshes the rows and Open folder destination, even when the old folder’s response arrives late.
  • Restored the Marker preview in zoom: verified that the strokes stay visible where they were drawn after Done, and checked what happens when the reader switches screenshots while writing the comment.

pnpm typecheck and 20 focused browser tests passed. A separate check reproduced the wrong-screenshot preview in Chromium and Firefox.

Pullfrog  | Fix all ➔ | Fix 👍s ➔ | View workflow run | Using gpt-6.2-sol(high) | 𝕏

Comment thread shared/plan-runtime.ts Resolved

@pullfrog pullfrog Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

ℹ️ No P0 or P1 findings remain. One P2 regression is noted inline.

Reviewed changes — reviewed 8c1df87..6a43cc3, including the merge from main.

  • Scoped retained Marker ink: checked what happens when a reader leaves and returns to the screenshot they marked while composing a comment.
  • Merged main: verified that the pinned ViteHub Comments package preserves full-link filters and Console admission without the local patch.

Typecheck and 86 focused tests passed. An additional two-screenshot check reproduced the inline finding in Chromium and Firefox.

Pullfrog  | Fix all ➔ | Fix 👍s ➔ | View workflow run | Using gpt-6.2-sol(high) | 𝕏

Comment thread shared/plan-runtime.ts Resolved

@pullfrog pullfrog Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

ℹ️ No P0 or P1 findings remain. One P2 Marker preview regression is noted inline.

Reviewed changes — reviewed 6a43cc3, the Marker visibility change since the prior Pullfrog review.

  • Scoped retained Marker ink to the active screenshot: checked the preview after Done, after switching screenshots and returning, and after closing and reopening the zoom. The ink no longer appears on the second screenshot, but it also fails to reappear on the original.

All 17 existing browser tests in the zoom, HTML-frame tooltip, and reviewed-site popover suites passed. A separate two-screenshot check reproduced the regression in Chromium and Firefox and passed in both when keyed by zoom.item instead.

Pullfrog  | Fix all ➔ | Fix 👍s ➔ | View workflow run | Using gpt-6.2-sol(high) | 𝕏

Comment thread shared/plan-runtime.ts Resolved

@pullfrog pullfrog Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

✅ No new issues found. No P0 or P1 findings remain.

Reviewed changes — reviewed c4197f9 since the previous Pullfrog review at 6a43cc3.

  • Keyed retained Marker ink to the original screenshot: verified that the preview stays hidden on other screenshots and outside zoom, and returns when the reader navigates back or reopens the original.
  • Added browser regression coverage: checked that the new two-screenshot test verifies both keyboard navigation transitions in Chromium and Firefox.

Typecheck and all 17 focused browser tests passed, including the reviewed-site popover and HTML-frame tooltip regressions.

Pullfrog  | View workflow run | Using gpt-6.2-sol(high) | 𝕏

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant