Repository navigation
chore: merge main into FE-3583 group projection - #21967
christian-byrne wants to merge 31 commits into
Conversation
<!-- ccr-slack-attribution --> _Requested via [Slack thread](https://comfy-organization.slack.com/archives/C0A4XMHANP3/p1791325190704329?thread_ts=1791325190.704329&cid=C0A4XMHANP3)_ ## Summary On Cloud, generated videos and audio in the Assets tab can fail to play. The server sends no `preview_url`/`thumbnail_url` for them, so `MediaAssetCard` falls back to `/api/assets/{id}/content?disposition=inline` as the `<video>`/`<audio>` `src`. Media elements send cookies but no headers, and the backend looks up `/content` assets in the session's default workspace for cookie requests. An asset in any other workspace therefore 404s. Image cards work because they load through `/api/view?filename=<hash>`, which is resolved across the user's workspaces. ## Changes - New `getAssetInlineMediaUrl(asset)` in `assetUrlUtil.ts`. On Cloud, an asset with a `hash` is loaded from `/view?filename=<hash>&type=<type>` built through `api.apiURL` (same `type` resolution as `getAssetUrl`). With no hash, or off Cloud, it falls back to the existing inline `/content` URL. - `MediaAssetCard` uses it for the video/audio fallback. `MediaAudioTop` passes the same `src` to `WaveAudioPlayer`, whose waveform fetch already strips the API base and calls `api.fetchApi` on the route, so it works with the `/view` URL unchanged. - Download and drag paths still use `getAssetFileUrl` and are untouched. - Also kept from earlier commits: `resolveMediaSrc` scopes server-supplied `thumbnail_url`/`preview_url` media routes (including fragments and same-origin absolute URLs) with the web-session workspace. Fixes #20346 ## Verification - Unit tests: `getAssetInlineMediaUrl` (hash gives `/view`, type from tags, no hash falls back to `/content`, not used off Cloud), and `MediaAssetCard` cases for video and audio with empty preview/thumbnail (hash gives a `/view` URL, no hash keeps `/content`). `src/platform/assets`, `src/components/sidebar/tabs`, `src/platform/auth/session`, `useWaveAudioPlayer` and `src/components/common` suites pass; typecheck, oxlint, eslint and oxfmt are clean. - Not verified against the Cloud backend. The fix assumes the assets API returns `hash` for these job-output assets; an asset without one still uses the `/content` URL and would still 404 outside the default workspace. --------- Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: Christian Byrne <cbyrne@comfy.org>
## Summary Carries #21113 onto exact current `main` while preserving its commits and behavior. This companion supersedes #21113 as the merge carrier because the teammate branch is conflict-bound and this program has no active authority to push to it. Fixes the defect in **PM-1270** and **PM-1257** (both P0): switching workflow tabs does not bring up the Agent chat belonging to that tab. The carrier also resolves the two still-valid review findings: tab-follow history refreshes now use the deployed `workflow_id` server filter from Comfy-Org/cloud#14128 with race-safe cache merging, and the restored-hash regression test uses the workflow store API instead of overwriting a computed property. --- ## Reviewer's map This PR is 63 files / +6,479. That number is misleading and it has cost it three days. **Of the 6,479 added lines, 5,210 (80%) are tests and fixtures.** The code you actually have to judge is 26 files / +1,222, and the core of it is three files. What follows is the honest decomposition so you can review it in one pass instead of budgeting a day. ### Already split out — review this first, it is 342 lines **#21791 — `refactor(workflow): report user-initiated tab activations`.** Every shared, non-Agent path this PR touched has been extracted there: `workflowService`, `workflowStore`, `subgraphNavigationStore`, `WorkflowTabs.vue`, `useCoreCommands`, `BaseWorkflowsSidebarTab.vue`, `MobileDisplay.vue` and their tests — 15 files, +342/−70, no user-visible behaviour, nothing subscribed to the new hook. It is the mechanism this PR consumes (`workflowStore.onUserActivation` is what triggers the follow). **Those 15 files are still in this diff and will disappear from it once #21791 merges** (via a merge from `main`, not a rebase, so no approval here is at risk). If you are reviewing before #21791 lands, skip them here and judge them there. ### Group 1 — multi-tab turn attribution (prerequisite refactor) | File | | | --- | --- | | `stores/agent/agentConversationStore.ts` | +171/−16 — **behaviour, read this** | | `stores/workflowTabActivityStore.ts` | +26/−9 — behaviour | | `components/topbar/WorkflowTab.vue` | +25/−21 — behaviour (new awaiting-approval dot) | | `composer/WorkflowSelectorChip.vue` | +4/−2 — mechanical (reads the new store field) | **Why it is here and not deferred:** the old code tracked a single `editingTabPath`. Once chats follow tabs, two tabs can each own a live turn at the same time, so one path is no longer expressible. This replaces it with a list of `AgentTurnTab`s, and a turn now carries a `TurnTab` whose `status` is `unknown` / `none` / `guessed` / `known` — the `guessed` state exists because a resumed turn's tab can only be inferred, and `guessed` is what `adoptThreadTab` is later allowed to overwrite. **The one thing to be skeptical about:** `turnTab()` in `AgentPanelRoot.vue` replaces the deleted `resumedTurnTabPath()`. The old function had a comment explaining that attributing an id-less turn to whatever tab is active lights the spinner on the wrong tab; the new `tabAt(..., 'guessed')` call is where that reasoning now lives, and the comment did not survive. Worth confirming the invariant did. ### Group 2 — the actual tab→chat follow (what PM-1270/PM-1257 describe) Read in this order: 1. **`composables/agent/useAgentWorkflowChatFollow.ts` (+297, new).** The whole feature. A reducer over `idle → loading → failed` with a `generation` counter; `isCurrent()` is the staleness check that every `await` boundary re-tests. **This is the file to review properly.** The interesting cases are `heldBack()` (a temporary workflow, an engaged user, or a chat already targeting this workflow suppresses the follow) and `restoreChatWorkflow()` (the followed chat adopts the tab instead of restoring its own recorded workflow). 2. **`composables/agent/useAgentChatHistoryRefresh.ts` (+127/−18).** The race-safe cache merge: `newestSighting` / `coveredAt` / `visibleSessions` let a narrow `workflow_id`-filtered listing land without evicting threads that only a full listing has seen. **Density is here, not in the feature file.** `services/agent/agentRestClient.ts` (+11/−3) is the matching query-param plumbing and is mechanical. 3. **`composables/agent/useAgentSession.ts` (+97/−44).** `loadThread` gains `keepOpenChatUntilLoaded`, which defers the destructive part of the thread switch (`openThread()`) until the new history has actually arrived — that is what stops a tab switch blanking the panel before it fills. `hydrateFromServer`'s positional args become an options object; `failureNotice` routes a follow failure to an inline notice instead of the blocking error overlay. The rest of this file's diff is that signature change. 4. **`AgentPanelRoot.vue` (+202/−73).** Wiring only, no new logic: constructs the composable, retires the follow on every competing user action (`onNewChat`, `onSelectHistory`, `onOpenHistory`, unmount), and renders the failure banner. ### Group 3 — mechanical, safe to skim `AgentPanel.vue` (+59/−9) and `Composer.vue` (+34/−11) add the opening-chat loading state: a 120 ms-delayed overlay, `inert` on the transcript and footer, and suppressed keydown/submit while opening. `RunNoticeBanner.vue` (+39/−16) and `RunNoticeAction.vue` (+25, new) add a `followFailed` context with a retry action and extract the duplicated button. `agentChatHistoryStore.ts` (+2), `agentComposerStore.ts` (+7, a `hasDraft` computed) and `useAgentWorkflowSelection.ts` (+5, an `adoptTarget` wrapper) are each one addition. ### What the tests prove - `browser_tests/tests/agent/agentTabActivationFollowsChat.spec.ts` (+94) and `agentTabChatTransitions.spec.ts` (+147) are the **black-box proof**: activating a tab surfaces that tab's chat, A/B/A with a late response, pinned-Send routing, skill-only draft retention, independent background completion indicators. These are what to trust for "does the reported defect actually stop happening". - `AgentPanelRoot.test.ts` (+2,024/−188) and `AgentPanelRoot.tabActivity.test.ts` (+826) are the bulk of the diff. The second is Group 1 only. Neither needs line-by-line review. - `useAgentChatHistoryRefresh.property.test.ts` (+120) is a property test over the cache-merge ordering — the right shape for that code, and the reason the merge logic is worth less of your attention than its size suggests. ### Known review debt in this diff This PR deletes a number of load-bearing explanatory comments in files it otherwise only lightly touches (`workflowService.ts`, `subgraphNavigationStore.ts`, `AgentPanelRoot.vue`, `agentConversationStore.ts`). That is churn, not intent. **The ones in #21791's files have been restored there**; the Agent-side ones are still deleted here and are a fair review comment. --- ## Verification - Parent-red black-box proof at `e910ac2e8ed05ebf5d5cd6fb79025e9e5d8fc625`: `PLAYWRIGHT_TEST_URL=http://127.0.0.1:6243 pnpm exec playwright test browser_tests/tests/agent/agentTabActivationFollowsChat.spec.ts --project=cloud --workers=1 -g "brings up the tab’s chat on activation"` failed at the visible-state assertion: the active Portrait tab still showed `Widen the landscape study` instead of `Tidy up the portrait study`. - Head-green black-box proof at `e9f4becdbb18d54f504d3397bc1a1b3c5ef18210`: `PLAYWRIGHT_TEST_URL=http://127.0.0.1:6242 pnpm exec playwright test browser_tests/tests/agent/agentTabActivationFollowsChat.spec.ts browser_tests/tests/agent/agentTabChatTransitions.spec.ts --project=cloud --workers=1` passed 7/7 with retries disabled. - Targeted post-rebase units passed 156/156 across `AgentPanelRoot.tabActivity`, `useAgentWorkflowChatFollow`, `workflowTabActivityStore`, and `workflowService`. - `pnpm typecheck:browser`, targeted ESLint/Oxlint, and the push gates (dependency-cruiser, Knip, changed-line Fallow audit) passed. - Hosted-CI follow-up: Chromium shard 13/16 twice exposed an unmocked release-note request in the user-selection fixture. Final head `c841ee4479` adds the same local empty-release mock used by the auth fixtures; commit and push gates passed, and the fresh hosted run is the authoritative multi-user check. - Exact base at final local verification: `e910ac2e8ed05ebf5d5cd6fb79025e9e5d8fc625`. ## Dependency and disposition - **#21791 is the base of this stack.** Land it first; these 15 shared-path files then drop out of this diff. - #20827 is merged. - Comfy-Org/cloud#14128 is merged and supplies the workflow-filtered thread endpoint. - Replaces #21113 as the current merge carrier; does not revive closed #21110 or #18402. - Related: PM-1270, PM-1257. **Why this still needs the frontend team rather than the greenlight bot:** 29 of the 63 paths fall outside the four Agent globs in `ProtectMain`'s `required_reviewers`, so the `comfy_frontend_devs` entry applies and the bot cannot satisfy it. #21791 removes 15 of those 29. The remainder are unavoidable for the feature — `src/locales/en/main.json` is deliberately unowned, and `browser_tests/fixtures/**` is CODEOWNERS-owned but absent from the ruleset's Agent globs.
## Summary - replace opaque unmatched Agent API status text with bounded HTTP status + operation context - keep authentication telemetry classification grounded only in backend-provided text - pin the visible failure message through the Agent panel black-box harness - carry the visible in-flight progress proof and the no-reload app-mode startup proof onto current `main` Fixes #20626. ## Proof - Red on parent `1a6324c76d`: `agentHttpError.spec.ts` failed because the panel exposed `No error message` instead of `Agent request failed (HTTP 503, post_thread_message)`. - Green on `709b23a906`: focused Agent Playwright matrix 7/7; video-mode pin 1/1; `agentRestClient.test.ts` 132/132. - `pnpm typecheck:app`, `pnpm typecheck:browser`, oxfmt, typed oxlint, dependency-cruiser, knip, fallow, and commit/push hooks pass. ## Review scope All five changed paths are covered by Agent CODEOWNERS globs. No shared fixture, locale, generic harness, or expected-failure baseline path changes. Cursor review found two low-severity edge cases; both are fixed in `709b23a906` with focused unit assertions and their threads are resolved.
…(PM-2219) (#21889) Justification: approved at exact head ed221f1 by dante01yoon (frontend reviewer) plus comfy-greenlight-bot and coderabbitai; required checks test, lint-and-format, e2e-status, website-e2e all SUCCESS at that head; the single failing job upload-e2e-coverage / package is not a required status check on ProtectMain. Remaining review threads are explicitly non-blocking and are being answered and routed to follow-ups.
…ed (#21926) ## Summary A saved chat that cannot be opened made Chat history inescapable. Back now returns to the chat screen whether or not the chat it reopens loads. Back deliberately reloads the chat the user came from instead of revealing a transcript that only partially hydrated underneath it (`AGENT-HISTORY-0038`). What was missing is what happens when that reload fails too: the history screen stayed up with a "Could not open. Select this chat to retry." row, so Back answered Back with another failed attempt. New Chat was the only way out, and it costs the user the chat they were in. This changes Back only. A row the user picks still reports its failure in place and keeps the list up to retry, which is the behaviour `AGENT-HISTORY-0038` asks for and which `AgentPanel.test.ts` already pins. Reported by a beta tester on [PM-2044](https://linear.app/comfyorg/issue/PM-2044): no saved chat would open — not even the chat just started — and they could not leave history afterwards. A hard refresh did not help, because the chat fails to restore on the next boot for the same reason. **Why the open fails is a separate defect and is not addressed here**; #20555 covers the unresolved-workflow-target cause of it. That PR does not touch this exit path, so the two do not overlap: it reduces how often a user reaches this state and does not change what Back does once they are in it. The trap is surface behaviour, so it would be equally wrong on Cloud as on Local, where it was reported. Where Back lands when the reload fails is not a new state to reason about: it is the same chat screen a failed *startup* restoration already leaves you on — thread identity set, transcript empty, composer and history button both live. `AGENT-HISTORY-0038` already describes that case. So this makes Back agree with startup rather than inventing a third outcome, and the ADR's reason for reloading at all — not exposing a transcript that only partially hydrated underneath — is untouched, because the reload still happens. One doc gap, stated rather than silently left: `AGENT-HISTORY-0038` specifies that Back reloads the original chat during an unfinished or failed selection but does not say what happens when that reload fails. This fills that gap. The one-sentence ADR amendment is deliberately not in this PR — `docs/adr/` falls outside the Agent CODEOWNERS globs and would hold a P0 fix behind separate review routing. Happy to land it alongside or right after, whichever a reviewer prefers. ## Verification - **New black-box case**, `browser_tests/tests/agent/agentHistoryBackEscapesFailedChat.spec.ts`: a stored chat whose transcript request always fails, opened from history, then Back. It asserts visible panel state only — the "Could not open" alert, the history heading, and that the composer is reachable again. - **Red on the parent** (`1a6324c76d`): fails at the Back step with `expect(historyHeading).toBeHidden()` — `14 × locator resolved to <h2 …>Chat history</h2>`. The steps before it pass, so the spec also reproduces the reporter's first symptom through the harness. - **Green on this commit**, `--repeat-each=2`. - Run against a Cloud production-mode dev server, headless on Linux, `--project=cloud --workers=1`. - **New unit case** in `AgentPanel.test.ts` covering the same transition at the component level, red on the parent for the same reason. - `vitest run src/workbench/extensions/agent --retry=0`: 186 files, 4395 passed, 1 expected fail. - Neighbouring history specs re-run green unchanged: `agentChatHistoryTitleSync`, `agentChatHistoryPagination`, `agentChatRefreshPersistence`. - `typecheck:app`, `typecheck:browser`, oxlint (type-aware), eslint, stylelint and oxfmt pass.
## Summary - Surface exhausted workflow-sync retries in the Agent panel instead of failing silently. - Let the user retry synchronization without losing the current chat or replacing the canvas. - Rebuilt the formerly stacked PR on current `main`; the old implementation did not match the graph-API follower, but the user-facing failure mode still existed. ## Changes - Make retry-budget exhaustion enter a distinct `retry_exhausted` terminal state. - Expose a persistent semantic alert beside the composer with a Retry action. - Reset lifecycle timers and retry counters before resubscribing. - Keep permanent protocol refusals on their existing non-retryable path. ## Verification - Black-box regression: `browser_tests/tests/agent/agentRefusedDocSubscribe.spec.ts` failed on unchanged `main` because no alert appeared, then passed on this head and proved the retry accepts a new subscription, materializes the queued node, and dismisses the alert. - Full affected Playwright spec: 3 passed. - Focused unit/component suite: 215 passed. - `pnpm typecheck` and `pnpm typecheck:browser` passed. - Commit and push hooks passed, including formatting, stylelint, type-aware lint, app typecheck, dependency-cruiser, knip, and fallow. ## Review focus - Transient refusals remain silent while retries are still available. - Manual retry restarts the bounded lifecycle rather than creating an unbounded loop. - All changed paths are covered by the Agent CODEOWNERS globs. Related: #18181
## Summary Follow-up to #20098 (merged). Keep each selection operation to one `onSelectionChange` report from every entry point, and stop in-app node removal from reporting a deselection. Rebased onto `main` after #20098 landed; this PR is now a single commit holding the review-fix work that did not make it into #20098. ## Changes - **In-app removal of a selected node no longer reports.** `canvasStore`'s `node:before-removed` listener deselected with a report, so every `graph.remove` of a selected node reported once and Convert to Subgraph reported N+1 times. That contradicts ADR `CANVAS-SELECTION-0028` ("graph removal … without emitting"). It now deselects with `{ notify: false }`, matching `LGraph` removal. - Every `onSelectionChange` call in `LGraphCanvas` goes through one private depth-gated `notifySelectionChange()`. On `main`, five entry points check the depth and five (`resetTransientState`, cancelled `finalizeGhostPlacement`, `finalizeLiveSelect`, `_handleMultiSelect`, `deleteSelected`) call the callback directly, so a selection hook that starts one of those operations still adds a report. - Tests fail if any `select` / `deselect` / `selectItems` suppression scope is removed, if the depth counter is not restored after a throwing hook, or if the group menu reports twice. Removal silence covers nodes, groups and reroutes; the options form of `deselectAll` is covered. - The CANVAS-SELECTION-0028 amendment records the hook-nesting rule and the `canvasStore` listener. ## Test evidence - `LGraphCanvas.selection.test.ts`, `useGroupContextMenu.test.ts`, `canvasStore.test.ts`: 144 passed. - With `main`'s `canvasStore.ts` restored under this branch, the removal test fails (`onSelectionChange` called once); with this branch it passes. - Pre-commit lint, format and `typecheck:app` pass; pre-push depcruise and fallow audit pass. ## Breaking changes - Removing a selected node no longer reports a deselection to `onSelectionChange`. - A selection hook that starts a reset, ghost-placement cancel, marquee, or delete operation inside an outer selection operation no longer adds its own report. ## Review focus `notifySelectionChange()` is a private helper on `LGraphCanvas`. If the selection reporting moves out of `LGraphCanvas` (noted on #20098), this helper is the single call site to move. Part of #19707. ## Screenshots Not applicable. Unit regressions cover the selection event contract.
…410) (#21931) ## Summary [PM-1410](https://linear.app/comfyorg/issue/PM-1410) reports that the Agent still describes a link after the user removes it. This test-only PR pins the one live arm left after #21888 and cloud #15379: - another tab owns the workflow ID remembered by the thread; - the selected unsaved tab therefore cannot adopt that ID and remains unbound; - the user severs a visible canvas link; and - the continuing turn sends neither a semantic op nor the selected tab's updated draft. The production-path unit test and black-box Cloud panel test both expected-fail at that missing-draft boundary. They deliberately do not choose the product fix: resolving which workflow the turn targets is an identity-contract decision, and always posting an unidentified canvas could replace the wrong server document. Test-only; no production or fixture changes. ## Verification - `NODE_OPTIONS=--no-experimental-webstorage pnpm test:unit src/workbench/extensions/agent/composables/agent/useAgentSession.test.ts` — 274 passed, 1 expected failure. - Removing the unit expected-failure marker makes `(h12)` fail because the second turn's `draft` is `undefined`. - `PLAYWRIGHT_LOCAL=1 ... pnpm exec playwright test browser_tests/tests/agent/agentUnboundTabLinkSeverance.spec.ts --project=cloud --workers=1` — 1 passed (expected failure). - Removing the browser expected-failure marker makes the final assertion fail with `Expected: false, Received: null`, after the canvas visibly reaches zero links. This proves the continuing turn omitted the updated draft rather than failing a setup precondition. - `pnpm typecheck:browser`, `oxfmt --check`, `oxlint`, `git diff --check`, and push-time dependency/dead-code checks passed. ## Remaining gate Product/architecture must select an identity path that keeps one safe sync route attached when an unsaved selected tab receives a workflow ID already owned by another tab. The fix should resolve or explicitly target identity; it should not reopen document-wins behavior or always post an unidentified canvas.
…en they arrive (#21915) ## TL;DR Behind the server flag `late_custom_nodes_enabled` (off by default), on an editor that may be on a deployment, the editor waits at most 2 s for custom node scripts and then opens. Extensions that register after that are turned on in one batch once every script has settled. ## **Why** A deployment's CPU server serves the custom node scripts. While it is still starting, one slow pack keeps the whole editor on its loading screen, often 20 s or more, with nothing saying why. ## Before / After - **Before:** the loading screen stays up until the slowest pack's script arrives. - **After:** the editor opens about 2 s after the core extensions. Extensions that register later are held (none of their functions run). Once every script has settled, startup has finished, and the artist is not dragging or typing (at most 30 s), they are registered, `init` and `registerCustomNodes` run on them, node definitions are reloaded, their `setup` runs, and the open workflow is rebuilt once. The rebuild keeps the saved baseline and the unsaved mark and adds no undo entry. Missing-node warnings wait while extensions are held; the rebuild shows the ones still missing. - Flag off, a page booting on Comfy Cloud, or every script within 2 s: startup is exactly today's. ## Decisions - Hold state lives in `extensionStore` as one discriminated union (`immediate` / `holding` with the held list). The orchestration lives in unclassified `src/platform/lateCustomNodes/`, so `extension-runtime` imports no new zone. `lateNodeTypes.ts` exists only because `workflow-management` may not import `extension-runtime`. The new files show as "not claimed by any domain" in `architecture:check`, like other unclassified code. - Late extensions' startup steps run after every early extension's `setup` and after the first graph load; the rebuild runs the graph-load hooks again. Written up in ADR-EXTENSION-LOADING-0048 and `docs/extensions/core.md`. - The late work belongs to `GraphCanvas`: it aborts on unmount and the held extensions are dropped. - Failures: a late script that failed joins the usual load-failure report. An extension that cannot register, or whose startup step throws, is reported with `reportError` and the rest of the batch goes on. - The last batch rebuilds even when no late extension arrived, so missing nodes still show. - The late node-definition fetch uses a new `app.fetchNodeDefs()` that returns `{ ok, defs }` or `{ ok: false, status }` (`api.getNodeDefs()` now wraps `api.fetchNodeDefs()`), so an HTTP error answer counts as a failure; startup still reads it as `{}`, as today. When that fetch fails, late extensions still start, and the definitions stay owed: the next successful `reloadNodeDefs` runs every extension's `addCustomNodeDefs` once and registers what they add. While they are owed, `reloadNodeDefs` uses the same status-checked fetch: a refused answer keeps them owed, is reported, makes `reloadNodeDefs` resolve `false` and shows the refresh failure toast instead of "Updated"; with nothing owed it works as today. ## Try it Try it: no preview deploy for this repo; build the frontend with the flag on and open it against a deployment whose custom node pack is slow to serve: the editor opens in about 2 s and the pack's nodes start working when it arrives, with no reload. Recorded locally with ComfyUI and a fake pack whose script takes 20 s to serve. Screenshots: in the review ask's Slack thread: before and after the editor opens, and the pack turning on without a reload (taken locally with ComfyUI and a fake pack that takes 20 s). ## Tests - `src/platform/lateCustomNodes/lateLoading.test.ts` drives `loadExtensions` with real dynamic imports (data URLs that wait, then register): flag off and Comfy Cloud wait as today; fast scripts start as today; late extensions held, then steps, refresh and one rebuild in order; waits for startup, a held pointer, a focused text field (at most 30 s); unsaved mark and baseline kept; failures reported; abort stops the batch. - `extensionService.test.ts` (hold, release, register now, `invokeStartupHook`), `workflowService.test.ts` (missing nodes wait), `GraphCanvas.test.ts` (unmount aborts). - Each behaviour change was reverted once to check a test fails. Not run: Safari, the Playwright suite, a real deployment. Ref BE-23307
…ady (#21935) ## TL;DR While late custom nodes start, a toast says so, counts the seconds and names the packs still loading; when they are on it says "Custom nodes are ready". ## **Why** After #21915 the editor opens before its custom nodes work, often for 20 s or more while the deployment's CPU server starts (as #21915 measured). Without a word, an artist sees nodes that do nothing and no reason why. ## Before / After - **Before:** the editor opens and nothing says custom nodes are still coming. - **After:** "Custom nodes are starting... 12 s" with "Usually about 20 s. Their buttons and menus will show up here on their own. Waiting for: rgthree-comfy", updated every second, then "Custom nodes are ready" for 5 s. If no late extension arrived, the toast just goes away. `toastStore` gains `update(id, changes)` to change a shown toast's text without showing it again. ## Try it Try it: no preview deploy for this repo; build the frontend with `late_custom_nodes_enabled` on and open it against a deployment whose custom node pack is slow to serve: the starting toast counts up and names the pack, then says ready. Screenshots: in the review ask's Slack thread: the starting toast at 3 s and 12 s, and the ready toast (taken locally with ComfyUI and a fake pack that takes 20 s). ## Tests `lateLoading.test.ts` (toast text, seconds, waiting packs, ready, no ready when nothing arrived, nothing after abort) and `toastStore.test.ts` (update in place). Not run: Safari, the Playwright suite, a real deployment. Stacked on #21915 Ref BE-23307
…re, then Reload is offered (#21936) ## TL;DR A late custom node script that could not be fetched is tried once more after 5 s; if that fails too, a toast names the pack and offers Reload. ## **Why** A script whose server is still starting can fail to fetch instead of arriving late. Today that pack silently never turns on. ## Before / After - **Before:** a fetch failure is reported and the pack stays off for the session. - **After:** a fetch failure whose error names that script's URL exactly goes on the waiting list in the starting toast and is imported again with `?retry=1` after 5 s. If it fails again: "Some custom nodes didn't start. Reload the page to turn on: <pack>" with a Reload action. A script that threw, or loaded at all, is reported as today and never retried. A fetch failure within the first 2 s is treated as late too. ## Decisions - A built editor routes dynamic imports through Vite's preload helper, and `App.vue` cancels `vite:preloadError`, so a failed fetch resolves to `undefined` instead of rejecting. `importNoticingFetchFailure` decides from that import's own failure: it is retried only when the error is a fetch failure naming the script's exact URL. Global error events and message substrings no longer decide it, because two scripts failing together (or `main.js` beside `main.js.backup.js`) were confused and a script ran twice. - Chosen behaviour, narrower than the ticket: Safari's fetch error carries no URL, and Firefox names a nested file instead of the script, so in those cases the script is reported and not retried, as on main today. No Reload toast either: a broken plugin's bad relative import looks the same, and the toast would show on every page load. - Chrome keeps a failed nested import for the page's life, so retrying the script cannot recover a failed nested file; that case ends in the Reload toast. - The retry stops when the canvas unmounts, and no toast shows after that. - The rationale JSDoc on `importCustomExtension` is removed under the repo's comment policy since the function changed. ## Try it Try it: no preview deploy for this repo; build the frontend with `late_custom_nodes_enabled` on, open it against a deployment, and block one pack's script in DevTools: after 5 s it is tried again, and the Reload toast names that pack. Screenshots: in the review ask's Slack thread: the Reload toast after the retry also failed (taken locally, the fake pack answering 503). ## Tests `lateLoading.test.ts`: retry after 5 s turns the pack on; a failure the editor's preload helper swallowed is noticed and retried; a fast fetch failure is late; second failure shows Reload and is reported once; a script that threw is not retried; abort stops the retry and the toasts. `importNoticingFetchFailure` cases for each preload-error shape. `extensionService.test.ts` for `takeHeldExtensions`. Not run: Safari, the Playwright suite, a real deployment. Stacked on #21935 Ref BE-23307
## Summary Adds black-box coverage for the user-visible half of [BE-17612](https://linear.app/comfyorg/issue/BE-17612): a server-terminal `agent_ask_resolved{expired}` frame removes the run-consent card and cannot emit an answer request. This accompanies the durable backend arbitration in [cloud#16307](Comfy-Org/cloud#16307). ## Verification - `pnpm typecheck:browser` - `pnpm exec oxfmt browser_tests/fixtures/data/agent/agentAskFrames.ts browser_tests/tests/agent/agentAskConsent.spec.ts` - `pnpm exec oxlint browser_tests/fixtures/data/agent/agentAskFrames.ts browser_tests/tests/agent/agentAskConsent.spec.ts` - Targeted cloud Playwright case: 1 passed. - Full `agentAskConsent.spec.ts` cloud project: 5 passed after the documented cold-Vite first-run retry. - Pre-commit and pre-push hooks passed (type-aware lint, browser typecheck, dependency boundaries, knip, fallow).
…missing (#21937) ## TL;DR While custom node packs are still starting, their nodes show as waiting for their pack instead of missing. ## **Why** A node whose type is not registered yet draws as a missing node: red outline and a cross for a title. For a pack that is still starting that is wrong and alarming. ## Before / After - **Before:** red outline, no title, while the pack is on its way. - **After:** no red outline, the node type as its title, and a "Waiting for its pack" badge in the palette's badge colours. The late batch's rebuild replaces the node with the real one, or shows it as missing if its pack never came. Nothing the placeholder saves changes. ## Decisions - The title comes from a per-node `getTitle` instead of the `title` setter, because the setter is a tracked graph edit. ## Try it Try it: no preview deploy for this repo; build the frontend with `late_custom_nodes_enabled` on and open a workflow that uses a slow pack's node on a deployment: the node shows the waiting badge until the pack arrives. Screenshots: in the review ask's Slack thread: the pack's node shown grey while waiting, then turned on (taken locally with ComfyUI and a fake pack that takes 20 s). ## Tests `waitingNodes.test.ts` (waiting look, registered types untouched, badge once, nothing saved changes) and `workflowService.test.ts` (waiting instead of missing while packs may still arrive). Not run: Safari, the Playwright suite, a real deployment. Stacked on #21936 Ref BE-23307
## Summary Add black-box Agent coverage for applying a second instance of an existing subgraph definition. The proof exercises the current `widgets_values_form` contract, asserts both instances render through the shared graph API, and verifies editing the copied instance does not mutate the original. Carrier: [comfy-cli #1067](Comfy-Org/comfy-cli#1067). Scope: [PM-763](https://linear.app/comfyorg/issue/PM-763/agent-creates-subgraphs-and-edits-widget-promotion) / [BE-10796](https://linear.app/comfyorg/issue/BE-10796/comfy-cli-pack-subgraph-op-instantiate-an-existing-subgraph-definition). ## Verification - `pnpm typecheck:browser` - `PLAYWRIGHT_LOCAL=1 PLAYWRIGHT_TEST_URL=http://127.0.0.1:5173 DISTRIBUTION=cloud pnpm exec playwright test browser_tests/tests/agent/agentSubgraphFollower.spec.ts --project=cloud --workers=1` (5 passed)
… on (#21938) ## TL;DR The editor can ask which step its deployment's CPU server is on (`GET /api/workspaces/{id}/deployment/cpu-server`) and follow it until the server exists. The next PR shows it in the starting toast. ## **Why** The starting toast says "Usually about 20 s", which is wrong on a first open, when the CPU server is still being created or its image built (minutes). ## Before / After - **Before:** the editor cannot tell why custom nodes are slow. - **After:** `workspaceApi.getDeploymentCpuServer(workspaceId)` reads `{ deployment_id?, cpu_server: none | creating | building | created }`. `watchCpuServer` follows the step every 5 s until `created` or `none`, for at most 30 min, or until its owner aborts; an abort or the 30 min end the wait at once, even with a read still out, and clear the step. ## Decisions - The response type and its zod schema come from `@comfyorg/ingest-types`. The first commit adds them with the package's generator: it is the generated difference between cloud's `openapi.yaml` before and after cloud#16170 (cloud@2bc59b43), applied on top of the package as it is. The full sync in #21701 brings in billing changes that break other code, so nothing else comes in here. - `none` (the deployment gets no CPU server) stops asking, and so does a 404 (workspace not found, or the caller is not a member); a 502 with code `cpu_server_unknown` keeps the last step; any other failure, a 502 without that code included, keeps the last step and is reported once per watch. - Every answer is bound to the workspace generation it was asked in; an answer that outlives a workspace change is dropped, even after switching back to the same workspace. ## Try it Nothing shows yet; the next PR uses it. ## Tests `cpuServerStep.test.ts` (no workspace, 404, `cpu_server_unknown` 502, other 502, other failures, A to B to A, every 5 s until created, settles on none, keeps the step on 502, reports once, stops after 30 min, a read still out at 30 min, stops on abort, abort during the first or a later read, an owner aborted before the watch starts, a read that fails after the abort) and `workspaceApi.test.ts` (request, rejecting an unknown step, an Axios 404 or 502 turned into a `WorkspaceApiError`). Workspace state comes from the real store. `lateLoading.test.ts`: a healthy `main.js` beside a `main.js.backup.js` that could not be fetched is imported once, and only the backup is tried again. Not run: Safari, the Playwright suite, a real deployment. Stacked on #21937 Ref BE-23336
…aved card (#21800) ## Summary Billing-web top-up can now pay once with a new card or a saved card the customer picks, through `POST /api/billing/topup/interactive` (BE-21644 / BE-21646). It no longer needs an invoice default or a trip to the billing portal. Part of FE-3562 under BE-21140; resolves the frontend half of BE-5732. ## Why A workspace with an attached card but no live invoice default could not top up from billing-web (BE-5732). The page only charged the method on file, and the only way out was the portal. The backend now accepts one explicit pay-once authority. Billing-web was the missing consumer. ## Root cause `useFullPageTopup` always dispatched `quoted` with `method: 'on_file'` and posted to `/billing/topup`. That charges the invoice default, so there was no card form and no way to pick a saved card. ## How - **account-core** `createTopupCommand` gains `createInteractiveTopup({ amountCents, payment, idempotencyKey })`. `payment` is exactly one of `{ confirmationToken, returnUrl }` or `{ savedPaymentMethodId }`. `save_payment_method` and `make_default` are never sent. The body is checked against `zInteractiveTopupRequest` before sending. A 404 maps to `NOT_AVAILABLE`, with no fallback to the on-file route. `TopupQuote` now carries the quote's `paymentMethodConfigurationId`. - **Gate**: the pay-once rail needs both `unified_billing_settings` on (read once per page from Cloud `/api/features`, sharing the `billing_web_checkout_ui` read, fail-closed to `false`, DEV `ff:` override honoured) and a quote that includes `payment_method_configuration_id`. Flag off, unresolved or failed keeps today's on-file flow exactly as on main (`purchase.submit()`, no card form, no saved-method picker). The full-page top-up itself stays behind `billing_web_checkout_ui = full_page`. - **billing-web** reuses the subscribe checkout's rail as it is (`collect` rail, Saved / Add new payment tabs, `SavedMethodSelect`, `StripePaymentForm`). Saved methods are filtered to `type === 'card'`. Any card can be chosen, including when none is the default. The card form mounts with the quote's configuration, and the new `payOnce` prop gives `mode: 'payment'` with no `setupFutureUsage`. A new method whose token is not a card is refused before anything is sent ("Credits can only be paid for with a card"). - **Idempotency**: the page holds one purchase (its payment and key) until the server answers it for good: success, decline, or a 4xx other than 409/429. While held, Pay resends the same request under the same key. The same saved card, or the card form again (its fresh confirmation stands for the card still typed in), counts as the held purchase. After a decline, the next Pay starts a new purchase (the server answers a same-key replay with `PREVIOUS_OPERATION_FAILED`). - **Unknown outcome (FE-3593 §3, DES-1088 canon)**: a Pay with no answer (no response, 5xx, or its own operation's watch lapsing), a 429 or a 409 keeps the page locked (form inert, Pay busy, back arrow hidden, no card, no Contact support) while `lifecycle.recover` re-reads what the key became. It then lands on exactly one of: Success (paid); the 342-4767 footnote "This payment is already processing and can't be canceled." (processing), then Success or its card once it settles; Payment in progress 372-4951 (bank settling); We couldn't confirm your payment (DES-1212: parked for a human, or the re-read itself failed, with no reference code when no operation id is known); or back on the form with the Stripe element still mounted, so Pay resends under the same key (not charged). A 429 that charged nothing shows the inline "Payment couldn't be processed" card. A 409 follows the blocking operation's waiting state instead of a retry. `purchase.reset()` no longer runs on the pay-once rail. A decline of the operation an unanswered Pay became releases the key. - **Retired**: `EARLIER_PAYMENT_UNRESOLVED`, its locale string, and the changed-card branch (`HELD_FOR_ANOTHER_PAYMENT`). The lock means no other card can be chosen while an outcome is unknown. `isLocked` now also covers a capture re-reading a Pay, which the subscribe checkout shares (its 409 re-read is now inert too). - **3DS / processing / unknown**: the lifecycle runs these through `usePaymentAttempt('topup')`, so the in-page challenge and polling are unchanged. The new-card `return_url` is the top-up page's own link (`/v1/top-up`, same product / return_to / source / amount / canonical `workspace`). A return or a duplicate callback is a fresh mount that runs `lifecycle.recover` and follows the same operation. Mounting never POSTs. - **#21366 (FE-3450, on hold)**: not shipped. Its Stripe hosted Checkout direction is replaced by this in-page card form. Pointing the Cloud Add credits dialog at billing-web stays with that PR / FE-3505. ### Where this sits in the plan | Item | What | State | | --- | --- | --- | | BE-21644 / BE-21646 | Interactive top-up API + workflow, public routes | Done (cloud#14879, v0.485.11) | | BE-22387 | Staging ConfirmationToken reads, isolated payment verification | Done | | **FE-3562 (this PR)** | **Billing-web pay-once card form + saved-card choice** | **Open** | | #21366 / FE-3450 | Cloud Add credits → hosted checkout | On hold, to be reworked to route to billing-web | | FE-3505 | Add credits opens billing-web top-up for everyone | Merged routing; depends on this | | BE-21140 / BE-5732 | Parent acceptance | Open until browser acceptance passes | ```mermaid graph TD BE21140[BE-21140 interactive top-ups] --> BE21644[BE-21644 API + workflow] BE21140 --> BE21646[BE-21646 public routes] BE21140 --> BE22387[BE-22387 staging verification] BE21140 --> FE3562[FE-3562 billing-web card choice<br/>this PR] BE5732[BE-5732 top-up fails after saving a card] --> FE3562 FE3562 --> PR21366[#21366 FE-3450 reconciled:<br/>hosted Checkout dropped] FE3562 --> FE3505[FE-3505 Add credits opens billing-web] style FE3562 fill:#f9d71c,stroke:#333,stroke-width:3px ``` ## Rollout - The new flow is behind `unified_billing_settings` and stays off until that flag turns on. Prod already serves `payment_method_configuration_id`, so without the flag a billing-web deploy would have switched every full-page top-up to the card form. - The billing-web prod deploy is still held until staging acceptance passes. ## AS IS / TO BE Captured from billing-web running locally (`vite` on `origin/main` vs this branch at `a936cba91f`, `VITE_BILLING_ENV=test`) against the real testcloud backend, signed in, `/v1/top-up?...&amount_cents=2500`, with DEV overrides `ff:billing_web_checkout_ui` = `"full_page"` and `ff:unified_billing_settings` = `true`. The real quote carries `payment_method_configuration_id`. Pay was never clicked and no card data was entered. | AS IS (main) | TO BE: Saved tab, real response | | --- | --- | | <img width="480" alt="AS IS on-file Pay" src="https://github.com/user-attachments/assets/52779352-1cc6-43a9-9bc6-bacead6a0a8e" /> | <img width="480" alt="TO BE saved load error" src="https://github.com/user-attachments/assets/e9048cb9-7ae1-47b6-b5d2-5ceb3a559df6" /> | | On-file Pay only, no payment method choice | Saved / Add new payment. testcloud answers `GET /api/billing/payment-methods` with 400 `FEATURE_DISABLED` ("Embedded checkout is disabled"), so the saved list shows its load error | | TO BE: Add new payment tab, real Stripe form | TO BE: saved card picker | | --- | --- | | <img width="480" alt="TO BE new card form" src="https://github.com/user-attachments/assets/1904e279-eee5-4de6-af1a-72edd04c25d5" /> | <img width="480" alt="TO BE saved card picker" src="https://github.com/user-attachments/assets/5ed2656b-ddeb-41e1-8648-96225f8e5f18" /> | | Card form on the quoted configuration | **State forced with a response override**: `payment-methods` returns Visa 4242, Mastercard 4444 (neither default) and a `link` method. Only the two cards are offered | ## Verification - FE-3593 §3: `apps/billing-web` `vitest run src/views/FullPageTopupView.test.ts src/checkout src/config`: 743 passed; `src/views src/components` (subscribe checkout included): 551 passed. New / adapted cases in `FullPageTopupView.test.ts`: - 503, no response, 429 and 409 each keep the page locked with no card and no Contact support while the re-read is out, with no second POST (mutation-checked: dropping the re-read clause from `isLocked` turns all four red) - re-read finds a success: Success; bank settling: Payment in progress; parked: We couldn't confirm (code = op id); re-read fails: We couldn't confirm with no code; none send a second POST - 342-4767: still processing shows the can't-be-canceled footnote with Pay disabled, then Success when it settles - nothing charged: the card form is never remounted, no card shows, and the next Pay resends the identical request and key - 429 with nothing charged: the couldn't-be-processed card, no Contact support, key kept - 409: the blocking operation's waiting state, no retry card - double-charge regressions, adapted: same request and key after a 503, a new key after a decline; an earlier settled operation found by recovery never releases the key; a decline of the operation an unanswered Pay became does (mutation-checked: no held key turns 4 red, sending the old verdict turns 12 red, keeping the key after that decline turns 1 red) - `apps/billing-web` `vitest run src/views/FullPageTopupView.test.ts src/checkout/topupRequest.test.ts src/views/FullPageCheckoutView.test.ts src/views/FullPageCheckoutView.journey.test.ts`: 224 passed (before FE-3593). Earlier cases: - pays with the saved card the customer picks when none is the invoice default, and offers only cards - pays once with a new card on the quoted configuration, with a return URL for the same workspace and amount - refuses a non-card new method without sending anything - one or two return loads follow the pending operation and POST nothing - a quote with no configuration keeps the on-file charge - a quote with a configuration keeps the on-file charge when `unified_billing_settings` is off, its read fails, or it answers 5xx; `createInteractiveTopup` is never called (mutation-checked: dropping the flag turns all three red) - `apps/billing-web` `src/config/checkoutUi.test.ts`: `unified_billing_settings` true/false/non-boolean/absent/5xx/refused/signed-out/dev override, one shared `/api/features` read with the checkout UI variant. - `packages/account-core`: `topup.test.ts`, 39 passed. Request bodies for both authorities, the caller's key kept across an unknown outcome, 404 with no on-file fallback, capability refusal, schema refusal, quote configuration passthrough. - `packages/account-ui`: `StripePaymentForm.test.ts`, 29 passed. Pay-once mounts `mode: 'payment'` with no `setupFutureUsage`. - `pnpm typecheck:billing-web`, `typecheck:account-core`, `typecheck:account-ui` and `typecheck:website` (pre-commit) are clean. oxlint and oxfmt are clean. `fallow:audit` on the changed files reports no issues. - No live charge, no prod payment endpoint, and no provider or preference writes. Browser acceptance with a staging payment fixture is still to do (see below).
## Summary - The Billing tab shows one support banner when `has_parked_billing_operation` is true. - The button goes to https://comfy.org/contact/. ## Why A parked payment (`paid_ungrantable` or `supersede_uncancelled`) needs an owner-facing support banner. `billing_operation_parked` is only a denial reason and misses a parked first payment. ## How Root-cause: the workspace fact is now on the capabilities response. Render `has_parked_billing_operation` as received. An omitted fact stays hidden. ## AS IS No banner for a parked payment. ## TO BE When the fact is true, the tab shows "Something went wrong with a recent payment", "Contact support so we can fix it.", and a white Contact support link. ## AS IS / TO BE Captured from the running app (dev server on `origin/main` vs this branch at `29813c6449`, cloud distribution, real testcloud backend, `ff:unified_billing_settings` = `true`, SDK read rail on). The test workspace has no parked operation, so `has_parked_billing_operation: true` is **state forced with a response override** on the real `GET /api/billing/capabilities`. | AS IS (main), `has_parked_billing_operation: true` | TO BE | | --- | --- | | <img width="560" alt="AS IS no banner" src="https://github.com/user-attachments/assets/8bb36d2b-9c3d-427a-8434-9773fc1361bc" /> | <img width="560" alt="TO BE parked banner" src="https://github.com/user-attachments/assets/e6e9a014-8f3f-4fa1-96c2-5a62b8b108b3" /> | | No banner | Parked payment banner with a white Contact support link | ## Verification - Billing tab tests for the parked banner passed.
… spec (#21515) ## Why The Billing tab still used the earlier settings spacing: 24px between sections, 8px card radius, and 40px banner buttons. [FE-3530](https://linear.app/comfyorg/issue/FE-3530/billing-settings-visual-qa-comfyui-billing-tab-billing-site) (design source [DES-1272](https://linear.app/comfyorg/issue/DES-1272/billing-settings-visual-qa-comfyui-billing-tab-billing-site)) asks for Medium labels and a 12px card shell. Copy, server facts, and which banner shows stay as they are. The old Plan & credits panel is unchanged. The tab mounts only when `unified_billing_settings` is on. ## How - Section gap is `gap-12` (48px). Plan and handoff are a Medium label above a card (`rounded-xl`, `px-5 py-4`). - Card border uses `border-card`. `--card` and `--secondary` are both charcoal-600 `#262729`. `--border` is charcoal-200 `#494a50`, which is the token the ticket called out as too light. `border-card` is the existing semantic surface token that matches `#262729`. - Labeled buttons use the existing `md` size: 32px height, 12px label, radius 8. Banner buttons drop from `lg`. - Every Billing tab banner uses the same shell: `rounded-xl`, `p-3`, `gap-2` between title and body, `md` buttons. That covers the subscription paused banner and the plan checkout, auto top-up and local credit banners in `src/platform/billing/components/billingTab/`. - Plan names and credit pool amounts stay `font-semibold` (600). Row titles and the Available amount use `font-medium`. - Open billing and auto top-up Manage use `arrow-up-right`. Other outbound actions keep `external-link`. Plan Manage/Subscribe opens the in-app subscription dialog, so it has no outbound icon. - Auto top-up sits in the Credits section, 12px under the card. Loading and load error use the same card shell. - With `unified_billing_settings` on, the settings close button is a ghost (`textonly`) with a non-shrinking 16px lucide `x`, passed from SettingDialog to every Settings tab. Flag off and other modals keep `pi pi-times` on `secondary`. The workspace header ⋯ is `textonly` (foreground) only with the flag on; flag off it stays `muted-textonly`. - The 48px avatar and 24/32 SemiBold workspace name apply only when `unifiedBillingSettingsEnabled` is on, so the old Plan & credits header stays. The flag is read inside `useSettingUI`, which already owns it. - Flag-off rendering is unchanged from main: every shared component (WorkspaceHeaderMenu, BaseModalLayout, SettingDialog, SettingsWorkspaceHeader, WorkspaceProfilePic) keeps its main output when `unified_billing_settings` is off. Skipped, low priority, no fitting token: - Content background `#111112`. The closest palette stop is charcoal-900 `#111111`, and that stop is not a semantic color. `--background` is `#171718`. - Content padding 32. Settings content padding is the shared `px-6` on every workspace category, including the old panel when the flag is off. The handoff card holds the existing action. Figma's extra sentence is not in the locale, and this pass does not change copy. The plan section label reuses the existing "Plan" string. ## AS IS / TO BE Captured from the running app (dev server on `origin/main` vs this branch, cloud distribution, real testcloud backend, `ff:unified_billing_settings` = `true`). Real account state, no response overrides. | AS IS (main) | TO BE (this PR) | | --- | --- | | <img width="640" alt="AS IS billing tab" src="https://github.com/user-attachments/assets/1cda62de-993f-41cc-9f8e-664cadbecd21" /> | <img width="640" alt="TO BE billing tab" src="https://github.com/user-attachments/assets/cad9f667-7369-4b16-9b53-b3baefd22c66" /> | | 24px section gap, plan row without a card, small workspace header, filled close button | Plan label above a 12px card, 48px gaps, 48px avatar and larger workspace name, ghost close button, `arrow-up-right` on Open billing | ## Plan | Item | Ticket | PR | Status | | --- | --- | --- | --- | | Design source | DES-1272 | | Done | | **ComfyUI Billing tab** | **FE-3530** | **this PR** | **In review** | | Billing site | FE-3530 | platform, separate | Not in this PR | ```mermaid flowchart LR DES1272["DES-1272"] --> FE3530["FE-3530"] FE3530 --> ThisPR["this PR: ComfyUI Billing tab"] FE3530 --> Platform["platform billing site"] style ThisPR stroke-width:4px ``` ## Test plan - [ ] With `unified_billing_settings` on, open Settings → Billing and check plan, credits, auto top-up, handoff, loading, and load error. - [ ] With the flag off, the Plan & credits panel and the settings close button are unchanged. - [ ] With the flag on, the close button is a ghost with a 16px x on every Settings tab. - [ ] Open billing and auto top-up Manage use the arrow-up-right icon. View usage, Turn on, and Update payment keep external-link. - [ ] Paused, plan checkout, auto top-up and local credit banners share the 12px shell and 32px buttons.
## TL;DR The starting toast says which step the CPU server is on, and on a first open the editor imports custom node scripts only once the server exists. ## **Why** On a first open the CPU server is still being created or built, which takes minutes, not "about 20 s". And while it does not exist, every route its packs serve fails; Chrome keeps a failed import for the page's life, so a script imported then could only be turned on by a reload. ## Before / After - **Before:** "Usually about 20 s" whatever the server is doing; on a first open the scripts fail and end in the Reload toast. - **After:** the toast's second line follows the step: - Creating: "Creating the server for your custom nodes." / "First-time setup." - Building: "Building the server for your custom nodes." / "First-time setup for this version, takes a few minutes." - Created: "Loading custom nodes." / "This usually takes under half a minute." - No answer: as before. - The first read starts before the core extensions load and is waited for at most 1 s, within the 2 s. If it says creating or building, the editor opens at once and the scripts are imported when the server exists. A retry of a script that could not be fetched waits for the server only while it is creating or building; when the step is unknown it goes ahead after 5 s as before. ## Decisions - Asking stops when the late scripts are on, when the canvas unmounts, or after 30 min; nothing is imported after an unmount, including one during the first 1 s wait. ## Try it Build the frontend with `late_custom_nodes_enabled` on and open a deployment whose CPU server is asleep: the toast says Creating (or Building on RunPod), then Loading custom nodes, then ready, with no reload. ## Tests `lateLoading.test.ts`: each step's text, no ask with the flag off, scripts held until the server exists, the 1 s wait counted within the 2 s, retry waits for a server being set up but not for an unknown step, the toast moving on from Creating, asking stops when done, nothing imported after unmount (also during the first wait), a gone editor's late run leaving the next editor's extension hold alone (the hold itself is #21937's). Not run: Safari, the Playwright suite, a real deployment. Stacked on #21938 Ref BE-23336
… edits other fields (#21947) ## Summary When an agent frame changed one field on a node (for example a rename), the follower re-applied every synced field from the document. That put the document's old colour, shape and properties back over a local edit. The change collector now records which synced keys each frame changed, and the applier mirrors only those keys onto the live node. - Linear: https://linear.app/comfyorg/issue/FE-3494 - Foreign changes to colour, shape and properties still propagate. The existing colour-update unit cases and the subgraph follower cases pass unchanged. - No reconcile step, no new op, and the follower still never writes the shared doc. Remote writes still go through the `LGraphNode` setters. A whole-node replace (`upsertNode`) and a lineage `replace` frame still apply every field. - A rejected `set_node_field` now restores only the register it claimed (`flags.collapsed` restores `flags`). - Not in this PR: local `properties` edits are still not written to the document during a turn. That needs a wider `set_node_field` across comfy-multi-player, comfy-cli, cloud and this repo. Colour and shape stay local, which matches the open product-decision default. ## Verification - Black-box: `browser_tests/tests/agent/agentLocalNodeAppearanceSurvivesAgentEdit.spec.ts`, with both `test.fail()` markers removed (the FE-3494 pins from #21711). On the parent commit both fail at the field assertion: colour expected `rgb(51, 34, 34)`, got `rgb(34, 51, 34)`; radius expected `0px`, got `12px`. With the fix both pass (`--project=cloud`, local backend, `DISTRIBUTION=cloud`, `VITE_USE_LEGACY_DEFAULT_GRAPH=true`). - Neighbouring agent specs pass, 17 of 17: rename write-back, tab-switch catch-up, ghost flag, human add plus tab switch, and draft matches canvas. - Unit: a new `liveGraphApplier.test.ts` case for a title-only frame keeping local colour, shape and a property. It fails on the parent and passes with the fix. `src/workbench/extensions/agent/crdt/`: 1270 passed, 1 expected fail. - `typecheck`, `typecheck:browser`, oxfmt, oxlint, eslint, and the pre-push `fallow:audit` are clean.
## Summary - Fixes [PM-2221](https://linear.app/comfyorg/issue/PM-2221/cloud-agent-tutorial-covers-the-sign-in-confirm-modal-on-first-desktop): the Agent onboarding coach no longer covers the managed desktop sign-in confirmation. - Adds `dialog_open` as a coach deferral reason, derived reactively from the dialog store. - Reads the dialog store directly without an intermediate composable; dialog mutation and rendering stay with the dialog host. ## Test plan - [x] `pnpm test:unit src/workbench/extensions/agent/AgentPanelRoot.test.ts --retry=0` (551/551) - [x] `pnpm lint:unstaged:fix` - [x] `pnpm architecture:check` - [x] `pnpm typecheck:app` - [x] pre-push dependency, knip, and Fallow gates ### Black-box regression - [x] `browser_tests/tests/agent/agentConsent.spec.ts` at buggy parent `78492d6a93` with the new case applied: failed at the visible overlap assertion (`Expected: 1`, `Received: 2` dialogs). - [x] The same focused `@cloud` case at `97f0971cd4`, zero retries: 1 passed. - [x] Full consent spec, zero retries: the new case and 24 others passed; one pre-existing reload case hit the cold-server 15-second timeout and passed immediately when rerun alone. - [x] Browser typecheck, type-aware Oxlint, formatting, dependency cruise, knip, and Fallow gates pass. --------- Co-authored-by: Alexander Brown <drjkl@comfy.org>
## Summary Adds current-architecture black-box coverage for [BE-9740](https://linear.app/comfy/issue/BE-9740): initial workflow sync succeeds, a host update is withheld to simulate a silent subscription stall, and the follower automatically catches up through its retained state vector after the stale probe. The test also proves the existing selected canvas node keeps the same DOM identity, so recovery does not replace or wipe the live document. The recovery implementation is already present on `main`; this is a test-only carrier that closes the missing user-visible proof. ## Verification - `pnpm typecheck:browser` - `pnpm exec oxfmt --check browser_tests/tests/agent/agentFollowerSilentStallRecovery.spec.ts` - `pnpm exec oxlint browser_tests/tests/agent/agentFollowerSilentStallRecovery.spec.ts` - `PLAYWRIGHT_LOCAL=1 PLAYWRIGHT_TEST_URL=http://127.0.0.1:6200 PLAYWRIGHT_SETUP_API_URL=http://127.0.0.1:8188 DISTRIBUTION=cloud pnpm exec playwright test browser_tests/tests/agent/agentFollowerSilentStallRecovery.spec.ts --project=cloud --workers=1 --retries=0 --repeat-each=3` (3/3 passed) - Mutation check: suppressing only the stale-probe resubscribe call fails at the missing catch-up node; restoring it returns the test to green. - Guarded push hooks passed dependency-cruiser, knip, and changed-file fallow audit.
## Why A sales-managed plan still showed Manage on the Billing tab. Contact sales should take that slot, and the button should be white only when Cloud reports the subscription as ended. ## Root cause The plan action was chosen only from `can_open_plans` and the row kind. `plan_kind` is on the capabilities snapshot and was not passed through, so a `sales_managed` plan kept the self-serve Manage button. ## How - `useBillingCapabilities` exposes `plan_kind` as received (`receiveServerFact`). Its type is the generated `NonNullable<BillingCapabilitiesResponse['plan_kind']>` from `@comfyorg/ingest-types`. - `planRowAction` (`src/platform/billing/components/billingTab/planRow.ts`) returns `contactSales` when `plan_kind` is `sales_managed`, ahead of the `can_open_plans` gate. - `BillingPlanRow.vue` renders Contact sales as an external link to `https://comfy.org/contact/` (new tab, `noopener`). The other actions keep main's `openPlan` button. - The Contact sales variant is `inverted` when the served `subscription_status` is `ended` and `secondary` otherwise. It is not combined with `isCloud` or any other client fact. - `isSalesManagedTier` is not used for this button. ## AS IS / TO BE Captured from the running app (dev server on `origin/main` vs this branch, cloud distribution, real testcloud backend, `ff:unified_billing_settings` = `true`). The test account is self-serve, so `plan_kind: sales_managed` (and `subscription_status: ended` in the third image) is **state forced with a response override** on the real responses. These three use the legacy read rail (`ff:billing_sdk_topup_enabled` and `ff:billing_sdk_subscription_enabled` = `false`); see the fourth image for the SDK rail. | AS IS (main), `sales_managed` | TO BE, `sales_managed` | TO BE, `sales_managed` + `ended` | | --- | --- | --- | | <img width="420" alt="AS IS Manage" src="https://github.com/user-attachments/assets/8dc26be3-7994-4037-93ba-8be10f583e3b" /> | <img width="420" alt="TO BE Contact sales" src="https://github.com/user-attachments/assets/d7d90145-c457-4f07-8979-8db1c9c62b9e" /> | <img width="420" alt="TO BE Contact sales ended" src="https://github.com/user-attachments/assets/a95f5075-0247-4085-8465-8345872376d7" /> | | Manage still shows | Secondary Contact sales link | White Contact sales link | On the SDK read rail (testcloud default), the same forced `plan_kind: sales_managed` still renders Manage: `projectBillingCapabilities` (`src/platform/billing/sdk/billingCapabilitiesView.ts`) does not carry `plan_kind` through, so `planKind` is `null`. <img width="420" alt="TO BE on SDK rail still Manage" src="https://github.com/user-attachments/assets/46cdfacd-f4bc-402a-b0cc-e9839722600d" /> ## Plan position | Item | Ticket | PR | Status | | --- | --- | --- | --- | | Activating plan row | FE-3526 | #21511 | merged | | Contact sales replaces Manage | FE-3524 | this PR | **this PR** | ```mermaid graph LR activating["PR 21511 Activating plan row (merged)"] --> sales["FE-3524 Contact sales"] style sales fill:#f9f,stroke:#333,stroke-width:2px ``` ## Verification - `pnpm exec vitest run src/platform/billing/components/billingTab/ src/platform/billing/composables/useBillingCapabilities.test.ts src/platform/billing/components/SubscriptionPanelContentWorkspace.test.ts`: 11 files, 790 tests pass - BillingTab table covers `subscription_status` `ended` (inverted), `canceled` and `active` (secondary); the sales-managed case asserts the contact href and that Manage is gone
…#21536) ## Why A top-up can be paid on another device and still be waiting on the bank. The credits flyout and the Add credits dialog never showed this, so the same person could start a second payment. ## How Both surfaces read the first page of `GET /api/billing/topup/checkouts`, then read each session, and render `state` as served (generated `HostedTopupCheckout` / `HostedTopupCheckoutList` from `@comfyorg/ingest-types`). The flyout line and the dialog notice appear only when a session's `state` is `payment_processing`. The dialog title uses that session's `payment_amount_cents` only when exactly one session is processing and every listed session loaded (`A $25 payment is still processing`). Otherwise it leaves the amount out (`A payment is still processing`). Presets and Add credits stay enabled. The client derives no banner priority and shows no unfinished-checkout line. - The read repeats when the tab returns (`registerRefreshOnReturn`), so a checkout that settled elsewhere clears the notice without a remount. - After the surface unmounts, no new read starts, including a queued trailing refresh, and late results and failures are ignored. - The dialog notice sits in the footer, directly above Add credits (DES-1267 1d). Gated by `unifiedBillingSettingsEnabled`. Flag off keeps today's flyout and dialog, and the checkout list is never read. Files: `src/platform/billing/processingTopupPayment.ts`, `src/platform/billing/composables/useProcessingTopupPayment.ts`, `src/platform/billing/components/TopUpCreditsDialogContentWorkspace.vue`, `src/platform/workspace/components/CurrentUserPopoverWorkspace.vue`, `src/platform/workspace/api/workspaceApi.ts`. ## AS IS / TO BE Captured from the running app (dev server on `origin/main` vs this branch at `7566f8c386`, cloud distribution, real testcloud backend, `ff:unified_billing_settings` = `true`). The account has no hosted top-up checkouts, so `GET /api/billing/topup/checkouts` and the session read are **state forced with a response override** (one session, `state: payment_processing`, `payment_amount_cents: 2500`). Everything else is the real response. The Add credits dialog was opened from the flyout; nothing in it was clicked. | | AS IS (main) | TO BE | | --- | --- | --- | | Credits flyout | <img width="300" alt="AS IS flyout" src="https://github.com/user-attachments/assets/a1583b5f-e839-42cf-9476-fee3c6bb8c9d" /> | <img width="300" alt="TO BE flyout" src="https://github.com/user-attachments/assets/5e0ddcfa-f7af-4ea3-9f86-8c9ec57c179a" /> | | | No processing line | "Payment processing. You don't need to pay again." under Add credits | | Add credits dialog | <img width="400" alt="AS IS dialog" src="https://github.com/user-attachments/assets/63e1fdb3-299a-4f37-ada2-3603fcc506a8" /> | <img width="400" alt="TO BE dialog" src="https://github.com/user-attachments/assets/2ba58981-bd84-4259-ae6c-eaecaafa77ef" /> | | | No notice | "A $25 payment is still processing" / "No need to pay again." above Add credits | ## Plan position | Item | Ticket | PR | Status | | --- | --- | --- | --- | | Billing tab top-up banners | FE-3467 | PR 21468 | In review | | Plan checkout banners | FE-3467 | PR 21471 | In review | | Credits flyout and Add credits notice | FE-3467 | this PR | **this PR** | ```mermaid graph LR be["BE-21183 checkout state"] tab["PR 21468 billing tab"] plan["PR 21471 plan banners"] flyout["PR this credits flyout"] be --> tab be --> plan be --> flyout style flyout fill:#dbeafe ``` ## Verification - `pnpm exec vitest run src/platform/billing/processingTopupPayment.test.ts src/platform/billing/composables/useProcessingTopupPayment.test.ts src/platform/workspace/components/CurrentUserPopoverWorkspace.test.ts src/platform/billing/components/TopUpCreditsDialogContentWorkspace.test.ts src/platform/workspace/api/workspaceApi.test.ts`: 240 passed. - Flag off: the new line is absent and the checkout list is not read. - Flag on, `payment_processing`: the flyout line and the dialog notice appear, including the title without an amount. The notice is ordered after the amount presets and before Add credits. Add credits and the $50 preset stay enabled. - Unmount while a refresh is pending, with a second return queued: no extra list read. Unmount while a detail read is pending, then that read rejects: nothing is reported. Removing the disposal guard fails exactly these two tests.
…ling tab (#21468) ## Summary A hosted top-up paid with a slow method (bank debit, bank approval) shows nothing in ComfyUI until the credits land, so an owner who returns, reloads or switches device sees no sign the payment went through and may pay again. The Billing tab now reads the served hosted top-up facts (BE-21226) and shows the one DES-1267 banner for the state the server reports. - Linear: [FE-3467](https://linear.app/comfyorg/issue/FE-3467/show-cross-device-top-up-payment-progress) · Design: [DES-1267](https://linear.app/comfyorg/issue/DES-1267) (final) · Priority: [DES-1234](https://linear.app/comfyorg/issue/DES-1234) · Backend: [BE-21183](https://linear.app/comfyorg/issue/BE-21183) / [BE-21226](https://linear.app/comfyorg/issue/BE-21226) (Comfy-Org/cloud#14023, #14080, in staging v0.481.2) ### Why Returning from Stripe, closing the tab or Stripe `complete` is not settlement. Only `GET /api/billing/topup/checkouts/{id}` knows whether a payment is still processing, paid and applying credits, fulfilled or failed. Nothing in the app read it, so the in-between states were invisible across reloads and devices. ### Root cause The served `state` of each hosted top-up checkout was never read on the client, so the only signal of a slow payment was credits eventually landing. ### How | Served `state` (detail read) | Banner | Actions | Amount | | --- | --- | --- | --- | | `payment_processing` | Payment processing · "…so you don't need to pay again." | Dismiss | none (per Figma) | | `credits_applying` | Payment received · "Your credits are being added…" | Dismiss | none | | `failed` (newest checkout only) | Your {amount} top-up didn't go through | Dismiss, Add credits (only when `can_top_up`) | `payment_amount_cents`; dropped when null | | `open`, `expired`, `fulfilled` | nothing | | | - `workspaceApi.listTopupCheckouts` / `getTopupCheckout` return the generated `HostedTopupCheckoutList` / `HostedTopupCheckout` from `@comfyorg/ingest-types`. The tab lists the first page (ten newest) and reads each session independently; one unreadable session is reported once and does not hide the others. - `src/platform/billing/components/billingTab/topupCheckoutBanner.ts` is a pure ranker: failed > credits applying > processing. A failed checkout only speaks while it is the newest one started. Dismiss is per session and state. - `src/platform/billing/composables/useTopupCheckoutBanner.ts` reads only while `can_manage_billing` is served (the endpoints are owner-only). It re-reads when the tab becomes visible, polls every 5 s while any checkout is `credits_applying`, never overlaps reads, and queues one trailing read when a refresh is asked for during an in-flight read. - `BillingTopupCheckoutBanner.vue` is presentational and mounted in `BillingTab.vue` next to main's `BillingPlanCheckoutBanner` (FE-3527). Like the plan banner it suppresses `BillingLocalCreditBanner`, whose auto top-up and credit kinds rank below it. - Banner Add credits reports `settings_billing_panel`; the credits card keeps `credits_panel`. Both Add credits buttons turn `inverted` only when Available is 0 and the shown banner is a failed top-up. - `amount_cents` is never shown: it is the credit size before promotions, not the payment. No server fact is derived on the client and none is missing for these three banners. ### Scope and follow-ups (split) - This PR: top-up banners on the Billing tab. - Follow-up: the credits flyout line and the Add credits dialog notice from DES-1267. - Plan checkout banners were split to #21471, which was closed as superseded by #21708 (FE-3527, merged). This PR sits next to that banner. ### Plan position | Item | Ticket | PR | Status | | --- | --- | --- | --- | | Hosted top-up storage + supersede runtime | BE-21218 / BE-21225 | Comfy-Org/cloud#14069, #14080 | Merged | | Hosted top-up read API | BE-21226 | Comfy-Org/cloud#14023 | Merged (staging) | | Banner design | DES-1267 | | Final | | **Top-up banners on the Billing tab** | **FE-3467** | **this PR** | **Open** | | Plan checkout banners | FE-3527 | #21708 (supersedes #21471) | Merged | | Credits flyout line + Add credits dialog notice | DES-1267 follow-up | | Next | ```mermaid graph LR BE["BE-21226 read API"] --> TOP["FE-3467 top-up banners (this PR)"] DES["DES-1267 design"] --> TOP PLAN["FE-3527 plan checkout banners #21708"] --> TOP TOP --> FLY["Flyout + dialog notice"] style TOP fill:#f9d71c,stroke:#333,color:#000 ``` ## Verification - Rebased onto main: files now live in `src/platform/billing/components/billingTab/` and `src/platform/billing/composables/`. - `pnpm exec vitest run src/platform/billing/components/billingTab/`: 10 files, 208 tests pass. New: the visibility-during-in-flight-read regression fails without the queued refresh (1 failed / 56 passed). - `oxfmt` and `oxlint --type-aware` on touched files; pre-commit hook. - `fallow audit --base origin/main`: one introduced boundary finding, `BillingTopupCheckoutBanner.vue` → `Button.vue` (billing/presentation → design-system/presentation). Every billing component on main makes the same import and no public entry point exists, so it needs a billing domain record decision, not a code change. The composable's `workspaceApi` / `reportError` and the ranker's type import no longer cross a boundary. ### AS IS / TO BE Captured from the running app (dev server on `origin/main` vs this branch at `ea9314acb9`, cloud distribution, real testcloud backend signed in as a workspace owner, `ff:unified_billing_settings` = `true`). The account has no hosted top-up checkouts (`GET /api/billing/topup/checkouts` returns `[]`), so the list and detail reads are **state forced with a response override** (one `cs_test_forced1` checkout, `payment_amount_cents: 2500`). Every other read is the real response. Nothing was clicked. | AS IS (main), a payment is processing | TO BE: `payment_processing` | | --- | --- | | <img width="560" alt="AS IS no banner" src="https://github.com/user-attachments/assets/b06bab47-a219-42a7-9cb7-4aa6226a9ade" /> | <img width="560" alt="TO BE processing" src="https://github.com/user-attachments/assets/98dc9468-822a-411e-90a0-361e3e87edb1" /> | | No banner | Payment processing banner with Dismiss | | TO BE: `credits_applying` | TO BE: `failed`, `payment_amount_cents: 2500` | | --- | --- | | <img width="560" alt="TO BE credits applying" src="https://github.com/user-attachments/assets/28ff32c4-0053-4ba7-a294-62a5cf1073b6" /> | <img width="560" alt="TO BE failed" src="https://github.com/user-attachments/assets/4395364e-0864-41d5-ac55-ffc66c82a661" /> | | Payment received banner | "Your $25 top-up didn't go through" with Dismiss and Add credits |
## Summary Publishes the settled, behaviorally complete FE-1968 subset as a syntax-aware guard over the 51 files that form the Agent graph-API follower and transport seams. The required repository check now rejects three deterministic violations: - constructing outbound `update_b64` fields in follower code - citing the widget catalog or vocabulary through a moving branch rather than a commit SHA - hardcoding production Agent endpoints instead of using the shared event transport (deliberately broad: any absolute URL string literal inside these seams counts) Inbound destructuring, including nested object, array, array-rest, and `for...of` initializer patterns, is not treated as outbound construction; a default value that builds an `update_b64` object still is. The review profile explicitly keeps `op_id` retry identity, distribution-branch ownership, installed op-layer purity, and single-applier semantics reviewer-owned. Their proposed check designs are not approximated here: package-runtime coverage, lexical ownership, and suppression policy remain separate decisions. This replan removes the disputed package scanner, custom lexical scope index, and suppression/schema surface from the carrier while preserving the required CI wiring. ## Verification - `NODE_OPTIONS=--no-experimental-webstorage pnpm test:unit scripts/check-follower-invariants.test.ts --retry=0` — 54 passed - `pnpm follower-invariants:check` — 51 current seam files passed - `pnpm exec oxlint --type-aware scripts/check-follower-invariants.ts scripts/check-follower-invariants.test.ts` — passed - `pnpm typecheck:scripts` — passed - `pnpm exec oxfmt --check ...` — passed - mutation proof: deleting each selected rule from the aggregate checker independently made the combined three-rule test fail 3→2; restored head passes - each added inbound destructuring row fails on the checker before its fix (`8d9fb35f9b`, `e5ced3011e`) and passes at this head; dropping the `for...of` initializer-identity check fails the iterable-construction row; exempting every enclosing pattern (defaults included) fails the constructing-default row - pre-commit and pre-push hooks — passed No user-visible behavior changes; this is a mechanical architecture guard, so no Agent harness conversation is added. --------- Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
## Summary Wait for the consent-backed Agent entry state before asserting that desktop approval is the only open dialog. This makes the coach deferral test observe consumed consent state instead of merely observing that its request started. Stacked on #21928; test-only and nonblocking for the root review. ## Verification - `pnpm typecheck:browser` - `pnpm exec oxfmt --check browser_tests/tests/agent/agentConsent.spec.ts` - Focused Playwright case, cloud project, `--repeat-each=5 --workers=1`: 5 passed
…(FE-2247) (#21962) ## Summary Adds the missing black-box remount phase for [FE-2247](https://linear.app/comfyorg/issue/FE-2247). Closing the Agent panel unmounts `AgentPanelRoot`, its CRDT follower and the agent socket; reopening mounts a fresh follower. This pins that host updates resume on the canvas the user already has, with no follower writes. Test-only. Not duplicate coverage: - The `FE-2158` unit case from #17531 already covers the bridge side FE-2247 originally asked for (stale bridge cannot apply, one fresh subscription, next frame applies once). It mocks the projection, so it cannot see the canvas or the wire. - #21942 (BE-9740) covers a stalled subscription inside one live follower, not an unmount/remount. <details><summary>Full context for agent readers</summary> Phases: 1. First follower subscribes; the host broadcasts node A. A is selected. 2. Panel closes. The host records node B with nothing listening. 3. Panel reopens. Exactly one new `doc_subscribe` arrives; the catch-up paints B, and a later live broadcast paints node C. 4. A is still selected (a remove + re-add comes back unselected), the canvas holds exactly A, B and C, no `doc_ops` or `doc_reseed` frame was sent, and the subscribe total is 2. Observed on `main` at 11f5068: both subscribes carry the empty state vector `AA==`. A remount starts a fresh replica, and the full catch-up is applied idempotently onto the live graph. No `doc_unsubscribe` is sent on close, because the panel's socket closes with it. A DOM-attribute identity marker, the technique #21942 uses, was tried first. It stayed green when the injected wipe removed and re-added the node in the same tick, because Vue reuses the keyed element. Selection catches that wipe, so this spec asserts selection instead. </details> ## Verification - `pnpm exec playwright test browser_tests/tests/agent/agentFollowerPanelRemount.spec.ts --project=cloud --repeat-each=5`: 5/5 passed locally (cloud-distribution dev server, Node 26.10.0). - Mutation checks, each red, then reverted: - wiping the graph before a catch-up applies fails on selection - the bridge sending `doc_reseed` after subscribe fails on the write assertion - a remounted bridge that never resubscribes fails with `the remounted follower never subscribed` - `oxfmt`, `oxlint --type-aware`, `typecheck:browser` clean; push hooks passed with no bypass.
|
Important Review skippedAuto reviews are disabled on base/target branches other than the default branch. Please check the settings in the CodeRabbit UI or the
|
📦 Bundle Size
⚡ Performance Report
Scenario timings
Absolute values
Raw data{
"timestamp": "2026-10-26T15:32:51.099Z",
"gitSha": "d638a5e758d9df82d0b716494297a51cfe840970",
"branch": "christian-byrne/oploop-992d-fe21749-conflict",
"measurements": [
{
"name": "canvas-idle",
"durationMs": 2046.9870000000014,
"styleRecalcs": 7,
"styleRecalcDurationMs": 1.9779999999999998,
"layouts": 6,
"layoutDurationMs": 1.6890000000000005,
"taskDurationMs": 626.282,
"heapDeltaBytes": -18298724,
"heapUsedBytes": 51684276,
"domNodes": -161,
"jsHeapTotalBytes": 12320768,
"scriptDurationMs": 12.305,
"eventListeners": -68,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66585365853659,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "canvas-idle",
"durationMs": 2017.2639999999546,
"styleRecalcs": 10,
"styleRecalcDurationMs": 3.097,
"layouts": 9,
"layoutDurationMs": 3.4070000000000005,
"taskDurationMs": 567.458,
"heapDeltaBytes": -847932,
"heapUsedBytes": 60596524,
"domNodes": 0,
"jsHeapTotalBytes": 4980736,
"scriptDurationMs": 10.554,
"eventListeners": 0,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66611570247934,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "canvas-mouse-sweep",
"durationMs": 1847.611000000029,
"styleRecalcs": 75,
"styleRecalcDurationMs": 13.471,
"layouts": 8,
"layoutDurationMs": 2.2819999999999996,
"taskDurationMs": 911.582,
"heapDeltaBytes": 19917668,
"heapUsedBytes": 81219588,
"domNodes": 0,
"jsHeapTotalBytes": 6029312,
"scriptDurationMs": 167.764,
"eventListeners": 6,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "canvas-mouse-sweep",
"durationMs": 1772.8909999999587,
"styleRecalcs": 76,
"styleRecalcDurationMs": 13.709,
"layouts": 10,
"layoutDurationMs": 2.4749999999999996,
"taskDurationMs": 898.4689999999999,
"heapDeltaBytes": -746444,
"heapUsedBytes": 69956636,
"domNodes": 0,
"jsHeapTotalBytes": 12320768,
"scriptDurationMs": 162.163,
"eventListeners": 6,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66603773584906,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "canvas-zoom-sweep",
"durationMs": 1725.1579999999649,
"styleRecalcs": 18,
"styleRecalcDurationMs": 4.609999999999999,
"layouts": 10,
"layoutDurationMs": 2.568,
"taskDurationMs": 409.299,
"heapDeltaBytes": 21413960,
"heapUsedBytes": 82684212,
"domNodes": 56,
"jsHeapTotalBytes": 4194304,
"scriptDurationMs": 11.861,
"eventListeners": 43,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666346153846156,
"p95FrameDurationMs": 16.799999999999272
},
{
"name": "canvas-zoom-sweep",
"durationMs": 1705.4829999999583,
"styleRecalcs": 19,
"styleRecalcDurationMs": 5.159999999999999,
"layouts": 10,
"layoutDurationMs": 3.04,
"taskDurationMs": 412.12799999999993,
"heapDeltaBytes": 1582196,
"heapUsedBytes": 72864880,
"domNodes": 56,
"jsHeapTotalBytes": 11534336,
"scriptDurationMs": 11.659000000000002,
"eventListeners": 43,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66666666666666,
"p95FrameDurationMs": 16.800000000000182
},
{
"name": "dom-widget-clipping",
"durationMs": 716.8819999999982,
"styleRecalcs": 13,
"styleRecalcDurationMs": 2.9630000000000005,
"layouts": 9,
"layoutDurationMs": 2.4520000000000004,
"taskDurationMs": 511.60199999999986,
"heapDeltaBytes": 8568932,
"heapUsedBytes": 78956488,
"domNodes": 0,
"jsHeapTotalBytes": 11534336,
"scriptDurationMs": 174.02200000000002,
"eventListeners": 0,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66428571428573,
"p95FrameDurationMs": 16.700000000000273
},
{
"name": "dom-widget-clipping",
"durationMs": 676.8630000000257,
"styleRecalcs": 14,
"styleRecalcDurationMs": 2.905,
"layouts": 9,
"layoutDurationMs": 2.432,
"taskDurationMs": 497.533,
"heapDeltaBytes": 7655092,
"heapUsedBytes": 78400196,
"domNodes": 0,
"jsHeapTotalBytes": 11272192,
"scriptDurationMs": 169.412,
"eventListeners": 0,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.668292682926833,
"p95FrameDurationMs": 16.800000000000182
},
{
"name": "dom-widget-order-dense-n200-w200-i120",
"durationMs": 1219.177000000002,
"styleRecalcs": 0,
"styleRecalcDurationMs": 0,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 939.7719999999999,
"heapDeltaBytes": -1961984,
"heapUsedBytes": 75082012,
"domNodes": -191,
"jsHeapTotalBytes": 14163968,
"scriptDurationMs": 74.402,
"eventListeners": -106,
"totalBlockingTimeMs": 537,
"frameDurationMs": 32.881081081081064,
"p95FrameDurationMs": 16.800000000000182
},
{
"name": "dom-widget-order-dense-n200-w200-i120",
"durationMs": 1273.7240000000156,
"styleRecalcs": 0,
"styleRecalcDurationMs": 0,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 960.952,
"heapDeltaBytes": -2865420,
"heapUsedBytes": 75195236,
"domNodes": -191,
"jsHeapTotalBytes": 15474688,
"scriptDurationMs": 76.67599999999999,
"eventListeners": -106,
"totalBlockingTimeMs": 580,
"frameDurationMs": 34.23243243243244,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "dom-widget-order-sparse-n1000-w20-i120",
"durationMs": 898.6370000000079,
"styleRecalcs": 0,
"styleRecalcDurationMs": 0,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 571.92,
"heapDeltaBytes": 1624644,
"heapUsedBytes": 83741936,
"domNodes": -191,
"jsHeapTotalBytes": -262144,
"scriptDurationMs": 174.12,
"eventListeners": -106,
"totalBlockingTimeMs": 229,
"frameDurationMs": 25.979411764705862,
"p95FrameDurationMs": 16.80000000000109
},
{
"name": "dom-widget-order-sparse-n1000-w20-i120",
"durationMs": 855.3039999999328,
"styleRecalcs": 0,
"styleRecalcDurationMs": 0,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 500.62500000000034,
"heapDeltaBytes": 22893896,
"heapUsedBytes": 89822568,
"domNodes": 0,
"jsHeapTotalBytes": 0,
"scriptDurationMs": 168.92499999999998,
"eventListeners": 0,
"totalBlockingTimeMs": 224,
"frameDurationMs": 25,
"p95FrameDurationMs": 16.80000000000109
},
{
"name": "extension-clean-combined-draw",
"durationMs": 10945.762000000002,
"styleRecalcs": 0,
"styleRecalcDurationMs": 0,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 10126.418999999998,
"heapDeltaBytes": 1077912,
"heapUsedBytes": 80319104,
"domNodes": -257,
"jsHeapTotalBytes": 17428480,
"scriptDurationMs": 8054.682000000002,
"eventListeners": -68,
"totalBlockingTimeMs": 984,
"frameDurationMs": 31.50662824207493,
"p95FrameDurationMs": 50.100000000000364
},
{
"name": "extension-clean-combined-draw",
"durationMs": 10470.43999999994,
"styleRecalcs": 0,
"styleRecalcDurationMs": 0,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 9659.738000000001,
"heapDeltaBytes": -25058680,
"heapUsedBytes": 64406276,
"domNodes": -257,
"jsHeapTotalBytes": 3272704,
"scriptDurationMs": 7800.005,
"eventListeners": -68,
"totalBlockingTimeMs": 922,
"frameDurationMs": 31.242686567164178,
"p95FrameDurationMs": 50.100000000000364
},
{
"name": "extension-rgthree-prototype-draw",
"durationMs": 11187.295000000006,
"styleRecalcs": 1,
"styleRecalcDurationMs": 0.03699999999999971,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 10409.556,
"heapDeltaBytes": 74744808,
"heapUsedBytes": 168086192,
"domNodes": -257,
"jsHeapTotalBytes": 79818752,
"scriptDurationMs": 8443.602,
"eventListeners": -68,
"totalBlockingTimeMs": 1321,
"frameDurationMs": 31.769602272727276,
"p95FrameDurationMs": 66.69999999999891
},
{
"name": "extension-rgthree-prototype-draw",
"durationMs": 10417.911000000004,
"styleRecalcs": 0,
"styleRecalcDurationMs": 0,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 9610.305,
"heapDeltaBytes": 107824760,
"heapUsedBytes": 169008892,
"domNodes": 0,
"jsHeapTotalBytes": 79953920,
"scriptDurationMs": 7690.99,
"eventListeners": 0,
"totalBlockingTimeMs": 864,
"frameDurationMs": 31.09313432835821,
"p95FrameDurationMs": 50.100000000000364
},
{
"name": "large-combo-open",
"durationMs": 223.2169999999769,
"styleRecalcs": 21,
"styleRecalcDurationMs": 13.734,
"layouts": 8,
"layoutDurationMs": 2.4400000000000004,
"taskDurationMs": 168.28199999999998,
"heapDeltaBytes": 14628244,
"heapUsedBytes": 73616428,
"domNodes": 721,
"jsHeapTotalBytes": 0,
"scriptDurationMs": 9.172,
"eventListeners": 73,
"totalBlockingTimeMs": 3,
"frameDurationMs": 24.066666666666656,
"p95FrameDurationMs": 83.30000000000018
},
{
"name": "large-combo-open",
"durationMs": 313.3629999999812,
"styleRecalcs": 26,
"styleRecalcDurationMs": 16.266,
"layouts": 12,
"layoutDurationMs": 3.9780000000000006,
"taskDurationMs": 258.97700000000003,
"heapDeltaBytes": -25103596,
"heapUsedBytes": 63314952,
"domNodes": 559,
"jsHeapTotalBytes": 4456448,
"scriptDurationMs": 10.995999999999999,
"eventListeners": 6,
"totalBlockingTimeMs": 5,
"frameDurationMs": 26.39166666666669,
"p95FrameDurationMs": 116.69999999999982
},
{
"name": "large-graph-idle",
"durationMs": 2058.3770000000072,
"styleRecalcs": 7,
"styleRecalcDurationMs": 2.371,
"layouts": 7,
"layoutDurationMs": 2.3159999999999994,
"taskDurationMs": 718.692,
"heapDeltaBytes": -15050900,
"heapUsedBytes": 77173320,
"domNodes": -257,
"jsHeapTotalBytes": 5406720,
"scriptDurationMs": 35.815,
"eventListeners": -68,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666129032258066,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "large-graph-idle",
"durationMs": 2070.0010000000475,
"styleRecalcs": 9,
"styleRecalcDurationMs": 2.81,
"layouts": 8,
"layoutDurationMs": 2.5949999999999998,
"taskDurationMs": 727.679,
"heapDeltaBytes": -9102560,
"heapUsedBytes": 61059908,
"domNodes": -257,
"jsHeapTotalBytes": 913408,
"scriptDurationMs": 35.965,
"eventListeners": -68,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666129032258063,
"p95FrameDurationMs": 16.800000000000182
},
{
"name": "large-graph-pan",
"durationMs": 2836.42100000003,
"styleRecalcs": 69,
"styleRecalcDurationMs": 9.173999999999998,
"layouts": 7,
"layoutDurationMs": 2.1660000000000004,
"taskDurationMs": 1863.5949999999998,
"heapDeltaBytes": -19650200,
"heapUsedBytes": 72515008,
"domNodes": -257,
"jsHeapTotalBytes": 5369856,
"scriptDurationMs": 904.308,
"eventListeners": -66,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.665882352941175,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "large-graph-pan",
"durationMs": 2704.819999999927,
"styleRecalcs": 68,
"styleRecalcDurationMs": 9.091999999999999,
"layouts": 7,
"layoutDurationMs": 1.9120000000000001,
"taskDurationMs": 1812.718,
"heapDeltaBytes": -19157156,
"heapUsedBytes": 72793488,
"domNodes": -257,
"jsHeapTotalBytes": 3534848,
"scriptDurationMs": 874.5509999999999,
"eventListeners": -66,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66625766871166,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "large-graph-zoom",
"durationMs": 3672.724999999957,
"styleRecalcs": 71,
"styleRecalcDurationMs": 9.416,
"layouts": 65,
"layoutDurationMs": 10.411,
"taskDurationMs": 1789.9790000000003,
"heapDeltaBytes": -15233676,
"heapUsedBytes": 69461860,
"domNodes": -233,
"jsHeapTotalBytes": 3534848,
"scriptDurationMs": 778.303,
"eventListeners": -68,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666063348416287,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "large-graph-zoom",
"durationMs": 3809.0909999999667,
"styleRecalcs": 71,
"styleRecalcDurationMs": 10.196,
"layouts": 64,
"layoutDurationMs": 10.802,
"taskDurationMs": 1795.24,
"heapDeltaBytes": 5721028,
"heapUsedBytes": 78322820,
"domNodes": 24,
"jsHeapTotalBytes": 2883584,
"scriptDurationMs": 789.8589999999999,
"eventListeners": 0,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.665502183406115,
"p95FrameDurationMs": 16.800000000000182
},
{
"name": "legacy-node-drag",
"durationMs": 3016.689999999983,
"styleRecalcs": 45,
"styleRecalcDurationMs": 5.281,
"layouts": 5,
"layoutDurationMs": 1.4009999999999994,
"taskDurationMs": 1972.496,
"heapDeltaBytes": -12751936,
"heapUsedBytes": 72628568,
"domNodes": 0,
"jsHeapTotalBytes": 6029312,
"scriptDurationMs": 971.6850000000001,
"eventListeners": 182,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.758100558659216,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "legacy-node-drag",
"durationMs": 3181.664000000069,
"styleRecalcs": 45,
"styleRecalcDurationMs": 7.6419999999999995,
"layouts": 5,
"layoutDurationMs": 1.7039999999999997,
"taskDurationMs": 2042.237,
"heapDeltaBytes": -18236840,
"heapUsedBytes": 67555004,
"domNodes": -242,
"jsHeapTotalBytes": 6156288,
"scriptDurationMs": 975.5789999999998,
"eventListeners": 114,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.84308510638298,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "minimap-geometry-cadence",
"durationMs": 3111.1589999999865,
"styleRecalcs": 8,
"styleRecalcDurationMs": 2.142,
"layouts": 8,
"layoutDurationMs": 1.967,
"taskDurationMs": 806.944,
"heapDeltaBytes": -2318932,
"heapUsedBytes": 81032364,
"domNodes": -257,
"jsHeapTotalBytes": 5894144,
"scriptDurationMs": 382.82700000000006,
"eventListeners": -68,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666129032258066,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "minimap-geometry-cadence",
"durationMs": 3057.984000000033,
"styleRecalcs": 7,
"styleRecalcDurationMs": 1.8359999999999999,
"layouts": 7,
"layoutDurationMs": 2.063,
"taskDurationMs": 774.197,
"heapDeltaBytes": -14796952,
"heapUsedBytes": 67725432,
"domNodes": -257,
"jsHeapTotalBytes": 3796992,
"scriptDurationMs": 348.769,
"eventListeners": -68,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.665573770491804,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "minimap-idle",
"durationMs": 2030.7289999999512,
"styleRecalcs": 8,
"styleRecalcDurationMs": 2.3500000000000005,
"layouts": 8,
"layoutDurationMs": 2.142,
"taskDurationMs": 625.4399999999999,
"heapDeltaBytes": -31903076,
"heapUsedBytes": 61407592,
"domNodes": -257,
"jsHeapTotalBytes": 4882432,
"scriptDurationMs": 31.400000000000006,
"eventListeners": -68,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.6655737704918,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "minimap-idle",
"durationMs": 2011.9809999999916,
"styleRecalcs": 3,
"styleRecalcDurationMs": 0.9209999999999999,
"layouts": 3,
"layoutDurationMs": 0.8479999999999998,
"taskDurationMs": 579.246,
"heapDeltaBytes": 9690980,
"heapUsedBytes": 70158048,
"domNodes": 0,
"jsHeapTotalBytes": 0,
"scriptDurationMs": 31.470999999999993,
"eventListeners": 0,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666115702479335,
"p95FrameDurationMs": 16.800000000000182
},
{
"name": "minimap-progress-execution-cadence",
"durationMs": 2784.7960000000285,
"styleRecalcs": 27,
"styleRecalcDurationMs": 12.885999999999997,
"layouts": 8,
"layoutDurationMs": 2.2249999999999996,
"taskDurationMs": 635.0049999999999,
"heapDeltaBytes": -27580776,
"heapUsedBytes": 65262364,
"domNodes": -242,
"jsHeapTotalBytes": 3309568,
"scriptDurationMs": 198.319,
"eventListeners": -68,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.76626506024097,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "minimap-progress-execution-cadence",
"durationMs": 2689.6709999999757,
"styleRecalcs": 22,
"styleRecalcDurationMs": 10.830000000000002,
"layouts": 3,
"layoutDurationMs": 0.6440000000000005,
"taskDurationMs": 520.298,
"heapDeltaBytes": 180620,
"heapUsedBytes": 62395184,
"domNodes": 19,
"jsHeapTotalBytes": 0,
"scriptDurationMs": 177.095,
"eventListeners": 0,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.665838509316767,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "minimap-topology-cadence",
"durationMs": 3340.4009999999857,
"styleRecalcs": 8,
"styleRecalcDurationMs": 1.9989999999999999,
"layouts": 8,
"layoutDurationMs": 1.9530000000000003,
"taskDurationMs": 975.4560000000001,
"heapDeltaBytes": -5602132,
"heapUsedBytes": 77154556,
"domNodes": -257,
"jsHeapTotalBytes": 4845568,
"scriptDurationMs": 397.524,
"eventListeners": -68,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.74924623115578,
"p95FrameDurationMs": 16.799999999999272
},
{
"name": "minimap-topology-cadence",
"durationMs": 3290.8049999999776,
"styleRecalcs": 7,
"styleRecalcDurationMs": 2.1189999999999993,
"layouts": 7,
"layoutDurationMs": 2.867,
"taskDurationMs": 973.5339999999999,
"heapDeltaBytes": -18198356,
"heapUsedBytes": 75151260,
"domNodes": -257,
"jsHeapTotalBytes": 6193152,
"scriptDurationMs": 377.65500000000003,
"eventListeners": -68,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66598984771574,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "subgraph-dom-widget-clipping",
"durationMs": 592.6429999999527,
"styleRecalcs": 45,
"styleRecalcDurationMs": 4.715,
"layouts": 8,
"layoutDurationMs": 2.4029999999999996,
"taskDurationMs": 436.879,
"heapDeltaBytes": -635924,
"heapUsedBytes": 62252248,
"domNodes": 0,
"jsHeapTotalBytes": 8126464,
"scriptDurationMs": 146.481,
"eventListeners": 46,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.700000000000273
},
{
"name": "subgraph-dom-widget-clipping",
"durationMs": 604.7640000000456,
"styleRecalcs": 48,
"styleRecalcDurationMs": 5.537,
"layouts": 10,
"layoutDurationMs": 2.8230000000000004,
"taskDurationMs": 432.40599999999995,
"heapDeltaBytes": 3122860,
"heapUsedBytes": 74700400,
"domNodes": 0,
"jsHeapTotalBytes": 12582912,
"scriptDurationMs": 143.654,
"eventListeners": 46,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.700000000000273
},
{
"name": "subgraph-idle",
"durationMs": 2055.7620000000156,
"styleRecalcs": 10,
"styleRecalcDurationMs": 3.0869999999999993,
"layouts": 9,
"layoutDurationMs": 2.9450000000000003,
"taskDurationMs": 616.244,
"heapDeltaBytes": -12682024,
"heapUsedBytes": 58649428,
"domNodes": -183,
"jsHeapTotalBytes": 11010048,
"scriptDurationMs": 10.669999999999998,
"eventListeners": -115,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "subgraph-idle",
"durationMs": 2022.7160000000595,
"styleRecalcs": 10,
"styleRecalcDurationMs": 3.391000000000001,
"layouts": 9,
"layoutDurationMs": 3.0140000000000002,
"taskDurationMs": 608.6429999999999,
"heapDeltaBytes": 6482504,
"heapUsedBytes": 79923132,
"domNodes": 0,
"jsHeapTotalBytes": 11010048,
"scriptDurationMs": 12.310000000000002,
"eventListeners": 0,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666115702479342,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "subgraph-mouse-sweep",
"durationMs": 1777.0199999999932,
"styleRecalcs": 80,
"styleRecalcDurationMs": 17.328,
"layouts": 10,
"layoutDurationMs": 3.136,
"taskDurationMs": 922.031,
"heapDeltaBytes": -4610192,
"heapUsedBytes": 68312884,
"domNodes": 0,
"jsHeapTotalBytes": 11796480,
"scriptDurationMs": 116.93799999999999,
"eventListeners": 6,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666037735849056,
"p95FrameDurationMs": 16.799999999999272
},
{
"name": "subgraph-mouse-sweep",
"durationMs": 1774.7909999999365,
"styleRecalcs": 79,
"styleRecalcDurationMs": 14.389999999999999,
"layouts": 9,
"layoutDurationMs": 2.373,
"taskDurationMs": 895.276,
"heapDeltaBytes": -13177896,
"heapUsedBytes": 58743396,
"domNodes": -175,
"jsHeapTotalBytes": 12058624,
"scriptDurationMs": 111.448,
"eventListeners": -113,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66603773584906,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "subgraph-transition-enter",
"durationMs": 1433.4680000000617,
"styleRecalcs": 19,
"styleRecalcDurationMs": 25.977,
"layouts": 5,
"layoutDurationMs": 16.097,
"taskDurationMs": 1032.9640000000002,
"heapDeltaBytes": 9593224,
"heapUsedBytes": 101013292,
"domNodes": 16030,
"jsHeapTotalBytes": 17702912,
"scriptDurationMs": 31.826999999999995,
"eventListeners": 3246,
"totalBlockingTimeMs": 165,
"frameDurationMs": 19.59324324324325,
"p95FrameDurationMs": 16.800000000000182
},
{
"name": "viewport-pan-sweep",
"durationMs": 9373.470999999994,
"styleRecalcs": 250,
"styleRecalcDurationMs": 27.958,
"layouts": 8,
"layoutDurationMs": 1.9940000000000002,
"taskDurationMs": 5894.420999999999,
"heapDeltaBytes": -12621164,
"heapUsedBytes": 61278956,
"domNodes": -257,
"jsHeapTotalBytes": 4845568,
"scriptDurationMs": 2670.6409999999996,
"eventListeners": -66,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666014234875444,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "viewport-pan-sweep",
"durationMs": 10240.02699999994,
"styleRecalcs": 250,
"styleRecalcDurationMs": 29.732000000000003,
"layouts": 8,
"layoutDurationMs": 2.142,
"taskDurationMs": 6182.176,
"heapDeltaBytes": -27069676,
"heapUsedBytes": 64277948,
"domNodes": -257,
"jsHeapTotalBytes": 5406720,
"scriptDurationMs": 2850.549,
"eventListeners": -66,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.665960912052117,
"p95FrameDurationMs": 16.799999999999272
},
{
"name": "vue-large-graph-idle",
"durationMs": 22855.30799999998,
"styleRecalcs": 0,
"styleRecalcDurationMs": 0,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 21983.071,
"heapDeltaBytes": -37307400,
"heapUsedBytes": 264126616,
"domNodes": -3343,
"jsHeapTotalBytes": 11259904,
"scriptDurationMs": 301.24600000000004,
"eventListeners": -653,
"totalBlockingTimeMs": 0,
"frameDurationMs": 39.73756521739131,
"p95FrameDurationMs": 83.30000000000109
},
{
"name": "vue-large-graph-idle",
"durationMs": 23128.663999999957,
"styleRecalcs": 0,
"styleRecalcDurationMs": 0,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 22422.595999999998,
"heapDeltaBytes": -8806976,
"heapUsedBytes": 274753168,
"domNodes": -3343,
"jsHeapTotalBytes": 7852032,
"scriptDurationMs": 317.142,
"eventListeners": -653,
"totalBlockingTimeMs": 0,
"frameDurationMs": 40.30052264808362,
"p95FrameDurationMs": 83.30000000000291
},
{
"name": "vue-large-graph-pan",
"durationMs": 27951.983999999982,
"styleRecalcs": 181,
"styleRecalcDurationMs": 18.987000000000005,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 27639.982999999997,
"heapDeltaBytes": -17772828,
"heapUsedBytes": 274192800,
"domNodes": -3343,
"jsHeapTotalBytes": 13619200,
"scriptDurationMs": 858.407,
"eventListeners": -651,
"totalBlockingTimeMs": 40,
"frameDurationMs": 50.9087431693989,
"p95FrameDurationMs": 100.10000000000036
},
{
"name": "vue-large-graph-pan",
"durationMs": 27801.52600000008,
"styleRecalcs": 180,
"styleRecalcDurationMs": 19.589000000000024,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 27356.390000000003,
"heapDeltaBytes": 1984872,
"heapUsedBytes": 285341324,
"domNodes": -3343,
"jsHeapTotalBytes": 14598144,
"scriptDurationMs": 865.2259999999999,
"eventListeners": -651,
"totalBlockingTimeMs": 46,
"frameDurationMs": 50.82065813528337,
"p95FrameDurationMs": 100.09999999999854
},
{
"name": "vue-node-resize-workload",
"durationMs": 8160.224000000027,
"styleRecalcs": 20,
"styleRecalcDurationMs": 76.269,
"layouts": 20,
"layoutDurationMs": 32.94699999999999,
"taskDurationMs": 7981.625999999999,
"heapDeltaBytes": -20659864,
"heapUsedBytes": 271663448,
"domNodes": -3343,
"jsHeapTotalBytes": 17813504,
"scriptDurationMs": 149.53499999999997,
"eventListeners": -653,
"totalBlockingTimeMs": 0,
"frameDurationMs": 49.195180722891564,
"p95FrameDurationMs": 100
},
{
"name": "vue-node-resize-workload",
"durationMs": 8293.326999999976,
"styleRecalcs": 20,
"styleRecalcDurationMs": 66.30099999999997,
"layouts": 20,
"layoutDurationMs": 23.091,
"taskDurationMs": 8160.001999999999,
"heapDeltaBytes": -27124992,
"heapUsedBytes": 255198704,
"domNodes": -3343,
"jsHeapTotalBytes": 20697088,
"scriptDurationMs": 121.193,
"eventListeners": -653,
"totalBlockingTimeMs": 9,
"frameDurationMs": 46.36648044692738,
"p95FrameDurationMs": 99.90000000000146
},
{
"name": "vue-zoom-out-idle",
"durationMs": 11378.95199999997,
"styleRecalcs": 0,
"styleRecalcDurationMs": 0,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 11067.564999999999,
"heapDeltaBytes": 8843316,
"heapUsedBytes": 272491196,
"domNodes": 0,
"jsHeapTotalBytes": 0,
"scriptDurationMs": 148.73999999999998,
"eventListeners": 0,
"totalBlockingTimeMs": 0,
"frameDurationMs": 40.102777777777774,
"p95FrameDurationMs": 83.29999999999927
},
{
"name": "vue-zoom-out-idle",
"durationMs": 11672.54700000001,
"styleRecalcs": 0,
"styleRecalcDurationMs": 0,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 11411.422,
"heapDeltaBytes": 25968016,
"heapUsedBytes": 282938264,
"domNodes": 0,
"jsHeapTotalBytes": -13893632,
"scriptDurationMs": 159.13,
"eventListeners": 0,
"totalBlockingTimeMs": 0,
"frameDurationMs": 40.22827586206897,
"p95FrameDurationMs": 83.39999999999782
},
{
"name": "workflow-execution",
"durationMs": 130.47199999994064,
"styleRecalcs": 9,
"styleRecalcDurationMs": 10.711,
"layouts": 8,
"layoutDurationMs": 4.436000000000001,
"taskDurationMs": 100.994,
"heapDeltaBytes": 4920008,
"heapUsedBytes": 69811728,
"domNodes": 132,
"jsHeapTotalBytes": 0,
"scriptDurationMs": 11.891999999999998,
"eventListeners": 5,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.662500000000023,
"p95FrameDurationMs": 16.700000000000273
},
{
"name": "workflow-execution",
"durationMs": 124.16500000006181,
"styleRecalcs": 9,
"styleRecalcDurationMs": 8.359,
"layouts": 9,
"layoutDurationMs": 3.9459999999999997,
"taskDurationMs": 89.026,
"heapDeltaBytes": 4640124,
"heapUsedBytes": 64578352,
"domNodes": 125,
"jsHeapTotalBytes": 0,
"scriptDurationMs": 7.039,
"eventListeners": 5,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.662500000000023,
"p95FrameDurationMs": 16.700000000000273
},
{
"name": "agent-composer-empty-conversation-typing",
"durationMs": 414.8810000000367,
"styleRecalcs": 5,
"styleRecalcDurationMs": 0.625,
"layouts": 76,
"layoutDurationMs": 14.048,
"taskDurationMs": 370.413,
"heapDeltaBytes": -5011140,
"heapUsedBytes": 39861276,
"domNodes": 4,
"jsHeapTotalBytes": 3670016,
"scriptDurationMs": 173.595,
"eventListeners": 296,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.667999999999992,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "agent-composer-empty-conversation-typing",
"durationMs": 480.8269999999766,
"styleRecalcs": 5,
"styleRecalcDurationMs": 0.737,
"layouts": 76,
"layoutDurationMs": 15.699000000000002,
"taskDurationMs": 441.89300000000003,
"heapDeltaBytes": -4521184,
"heapUsedBytes": 38894316,
"domNodes": 4,
"jsHeapTotalBytes": 2621440,
"scriptDurationMs": 213.707,
"eventListeners": 292,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66333333333335,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "agent-composer-long-conversation-typing",
"durationMs": 575.6270000000541,
"styleRecalcs": 5,
"styleRecalcDurationMs": 2.029,
"layouts": 76,
"layoutDurationMs": 15.14,
"taskDurationMs": 531.314,
"heapDeltaBytes": 1153528,
"heapUsedBytes": 90629484,
"domNodes": 4,
"jsHeapTotalBytes": 528384,
"scriptDurationMs": 215.425,
"eventListeners": 292,
"totalBlockingTimeMs": 0,
"frameDurationMs": 17.15588235294118,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "agent-composer-long-conversation-typing",
"durationMs": 537.0570000000043,
"styleRecalcs": 5,
"styleRecalcDurationMs": 2.852,
"layouts": 76,
"layoutDurationMs": 14.45,
"taskDurationMs": 500.31800000000004,
"heapDeltaBytes": 1000968,
"heapUsedBytes": 102414056,
"domNodes": 4,
"jsHeapTotalBytes": 790528,
"scriptDurationMs": 194.39000000000001,
"eventListeners": 300,
"totalBlockingTimeMs": 0,
"frameDurationMs": 17.1875,
"p95FrameDurationMs": 16.800000000000182
}
]
}🔬 E2E CoverageLine coverage: 60.4% Coverage totals
Top 10 files by uncovered lines
🎭 Playwright: ✅ 3105 passed, 0 failed · 7 flaky📊 Browser Reports
🎨 Storybook: ✅ Built — View Storybook🧩 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
Findings and check detailsThe matrix runs on same-repository version-bump PRs with relevant changes, main-branch pushes, and manual dispatches. This candidate does not qualify, 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. Findings and check detailsFallow gates on new findings in changed files only, so everything below was introduced or moved by this PR (242 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 cdefcc65f34f6aa9ad93598b2abd630924110b6a
Full docs: ADR-DEVEX-LINT-0015 🏛️ Architecture: ✅ PassedFindings and check detailsWarnings (204)
Reproduce locallypnpm architecture:checkFull docs: Frontend domain catalog 🌐 Website E2ETip All tests passed. Test results
|
Codecov Report❌ Patch coverage is @@ Coverage Diff @@
## drjkl/fe-3583-observe-every-document-root #21967 +/- ##
=============================================================================
- Coverage 87.49% 86.64% -0.86%
=============================================================================
Files 2767 3942 +1175
Lines 152963 185032 +32069
Branches 41558 50298 +8740
=============================================================================
+ Hits 133840 160313 +26473
- Misses 18448 22256 +3808
- Partials 675 2463 +1788
Flags with carried forward coverage won't be shown. Click here to find out more.
... and 1167 files with indirect coverage changes 🚀 New features to boost your workflow:
|
## Summary Scope persisted Agent workflow drafts by authenticated subject and remember the CRDT lineage that produced each draft. On restore, the follower now sends its known lineage and replaces stale restored canvas state when the host confirms a different reset lineage, preventing one account or reset lineage from restoring another's draft. Fixes #16347 ## Verification - `pnpm vitest run src/platform/workflow/persistence/stores/workflowDraftStoreV2.identity.test.ts src/platform/workflow/persistence/stores/workflowDraftStoreV2.test.ts src/platform/workflow/persistence/composables/useWorkflowPersistenceV2.test.ts src/workbench/extensions/agent/crdt/docFrameClient.test.ts src/workbench/extensions/agent/crdt/followerSubscription.test.ts src/workbench/extensions/agent/crdt/useAgentCrdtFollower.test.ts src/workbench/extensions/agent/crdt/useAgentCrdtFollower.integration.test.ts` — 288 passed - `pnpm typecheck` - `pnpm typecheck:browser` - `PLAYWRIGHT_LOCAL=1 PLAYWRIGHT_TEST_URL=http://127.0.0.1:5199 pnpm exec playwright test browser_tests/tests/agent/agentDraftRestoreLineageIsolation.spec.ts --project=cloud` — 1 passed; visibly verifies a restored stale node disappears after the host reports another lineage - Push gates: dependency-cruiser, Knip, and Fallow; zero issues introduced in 17 changed files ## Compatibility decision Authenticated Cloud drafts stored before subject scoping are intentionally not restored because they cannot be attributed safely to the current account. This is an accepted one-time cost; the existing logout cleanup removes the legacy namespace. Desktop and local draft keys are unchanged.
dante01yoon
left a comment
There was a problem hiding this comment.
I checked the conflict resolutions by recomputing both merges with git merge-tree and diffing the results against your merge commits.
29518e0233(cdefcc65+ main00d9b864): the only conflict is inliveGraphApplier.tsat theresyncNodesloop. The resolution keeps main's field-keyedfor (const [id, keys] of changes.resyncNodes)→syncFields(graph, doc, id, keys)and drops main'sthis.deleteGroups(graph, changes.removedGroups, context). That matches the branch'sapplyGroupsadd/remove projection, which supersedes #21572's delete-only path. After the merge, nothing undersrcorbrowser_testsreferencesremovedGroupsordeleteGroups, so no producer or test is orphaned.resyncNodesis aMapin bothdocChangeCollectorandrejectedOpChanges.e7fba8b555(+ main6d00a3cd): the merge is clean, and the tree is byte-identical to the automatic merge.- #21572's coverage survives the merge.
agentDeleteAndClearWorkflow.spec.tsand its fixture match main, and the integration test keeps the retry case next to the branch's new host-clear case.
No gates apply to a merge-only PR. CI is green and the merge state is CLEAN. I didn't run tests locally.
|
@DrJKL this brings current Merging writes to your branch, so it is yours to land rather than ours. Could you take it when you get a chance? Worth merging as it stands rather than refreshing the branch first, because a new push here drops the approval. |
|
Closing. This can no longer do what it was opened to do, and the repository cannot land it correctly anyway. Approved and green, but neither of those is the blocker. It no longer makes #21749 mergeable. Head And the merge mechanism was never available. The repository allows squash merges only. Squashing this would put 31 commits of other people's Nothing is lost here. |
Summary
Merges current
mainat6d00a3cd274dc6322481ab54e976696ab9036aa0into the FE-3583 branch without pushing to another author’s branch. The original conflict resolution keeps main’s field-keyed node resync and #21749’s superseding add/remove group projection plus shared batch placement; it does not restore the delete-only group path from #21572.Verification
e7fba8b555c096ae2852e91566703887e11db3d0; both currentmainand fix(agent): observe meta.groups in DocChangeCollector and project groups #21749 headcdefcc65f34f6aa9ad93598b2abd630924110b6aare ancestors.NODE_OPTIONS=--no-experimental-webstorage pnpm test:unit src/workbench/extensions/agent/crdt/ --retry=0— 75 files passed, 1,279 passed, 1 expected failure.agentDeleteAndClearWorkflow.spec.tsandagentInsertWorkflowOverlap.spec.ts: 6 passed.