Repository navigation
fix(agent-crdt): port non-value widget gate to graph intents - #19737
christian-byrne wants to merge 230 commits into
Conversation
…esign-system (#19098) - Linear: FE-3046 **Merge order:** #19096 (ADR 0034 amendment) merges first. Until it does, ADR 0034 rule 4 still forbids account-ui from using the design system, which B does. ## Why `buttonVariants` lived in `src/components/ui/button/button.variants.ts`, inside the private root package, so a workspace package could not render a button with the app's exact classes. PR B (`@comfyorg/account-ui` checkout surfaces) needs those classes to render the embedded checkout identically in the cloud app and in billing-web. No visual change. ## How - `git mv` of `button.variants.ts` into `packages/design-system/src/`, content unchanged. - `packages/design-system/package.json`: new export `./button.variants`, a `files` entry, and `cva` (`catalog:`) as a dependency. - Every importer in `src/` (Button.vue, Button.stories.ts, CloudSocialAuthButtons.vue, InlinePromptEditor.vue, and the ten components that import the `ButtonVariants` type) now reads from `@comfyorg/design-system/button.variants`. The old path is deleted, no re-export shim. `apps/website` keeps its own copy. ## Review Focus The file content is byte-identical (rename only). The only decision is the entry name `./button.variants`, mirroring `./workspaceAvatar`. ## Stack | PR | branch | base | title | |---|---|---|---| | #19097 | `jaewon/billing-web-remove-plan-grid` | `main` | `FE-3045 refactor(billing-web): drop the billing-web plan grid; hosts own plan selection` | | **A #19098 (this PR)** | **`jaewon/design-system-button-variants`** | **#19097** | **`FE-3046 refactor(design-system): move buttonVariants into @comfyorg/design-system`** | | B #19099 | `jaewon/account-ui-checkout-surfaces` | A | `FE-3046 feat(account-ui): shared checkout confirm, saved-method selector, card form and success` | | C #19100 | `jaewon/workspace-adopt-account-ui-checkout` | B | `FE-3046 refactor(workspace): render the subscription checkout steps from account-ui` | | D #19101 | `jaewon/billing-web-shared-checkout` | C | `FE-3046 feat(billing-web): render the shared embedded checkout` | | E #19104 | `jaewon/billing-web-success-invite` | D | `FE-3046 feat(billing-web): invite your team from the checkout success step` | | F #19105 | `jaewon/checkout-3ds-failure-copy` | E | `FE-3047 fix(account-ui): clear decline toasts once a retry succeeds` | | G #19106 | `jaewon/billing-web-return-close-tab` | F | `FE-3048 fix(billing-web): return to the host by closing the checkout tab` | | H #19115 | `jaewon/checkout-cta-reachable` | G | `FE-3050 fix(account-ui): keep the checkout CTA reachable when notices grow the confirm step` | ```mermaid graph LR P0[#19097: remove plan grid] --> A[A: design-system buttonVariants] A --> B[B: account-ui checkout surfaces] B --> C[C: workspace adopts account-ui] C --> D[D: billing-web shared checkout] D --> E[E: billing-web success invite] E --> F[F: stale decline toasts] F --> G[G: close the billing tab] G --> H[H: reachable checkout CTA] style A fill:#f9d71c,stroke:#333,color:#000 ``` ## AS IS / TO BE Rename only, no rendered change, so no screenshots. | AS IS | TO BE | | --- | --- | | `@/components/ui/button/button.variants` | `@comfyorg/design-system/button.variants` | --------- Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
…18974) Closes **E1** and **E3** of [PM-1473](https://linear.app/comfyorg/issue/PM-1473/e-new-labels-agent-paywall-source-attribution-e1e5). Both were shipped as `Done` by #18390 but do not work in merged code. Unblocks revenue attribution **R4**. It does **not** close **E4**'s `allowance_exhausted`, which is blocked on a separate decision — see below. ## E1 — the agent CTA never reached `app:subscribe_now_button_clicked` The only emitter of that event is `SubscribeButton.vue`, and the agent paywall's subscribe CTA **bypasses that component entirely**: it goes `AgentPanelRoot.onPaywallAction` → `useAccountPreconditionDialog` → `showSubscriptionRequiredDialog`. The subscribe leg of the funnel therefore carried no agent attribution at all — not a wrong value, no event. `onPaywallAction`'s subscribe branch now emits the shared event with `reason: 'agent_paywall'`, mirroring the `trackAddApiCreditButtonClicked` call in the add-credits branch directly above it. `SubscriptionMetadata.reason` already existed and is already typed `PaymentIntentSource`; nothing new was needed on the type. ### How this shipped as done Worth recording, because #18390 disclosed E3's and E4's limits at length and claimed E1 shipped. E1 was judged complete on the strength of `VALID_PAYMENT_INTENT_SOURCES ... satisfies Record<PaymentIntentSource, true>` — "adding a union member breaks the build until the key is added." That guard is real, and it is still here, but it proves the value is **registered**, not that a call site **emits** it. Type-safety can show a value is able to flow; it cannot show anyone made the call. The asymmetry is visible in one function. E2 needed an explicit call site and got one. E1's symmetric call was never written, and no test asserted the subscribe leg, so both the compiler and the suite were satisfied by a funnel with a missing event. The guard against a repeat is behavioural, not structural: `AgentPanelRoot.test.ts` now asserts the agent CTA reports `agent_paywall` on that event. Removing the emit fails it. A helper shared with `SubscribeButton` would not have caught the original bug, which was an absent call site rather than a malformed one. ### Scope — this does not reach every paywall→subscribe path `app:subscribe_now_button_clicked` has exactly two emitters after this change: `SubscribeButton` and the agent CTA. Other surfaces that open subscription checkout — `subscribe_to_run` (`useCoreCommands`), `upload_model_upgrade`, `settings_billing_panel`, and the two `app.ts` precondition callers — emit **no** `subscribe_clicked` today and still don't. That is pre-existing and deliberately untouched: the shared choke point is `useSubscriptionDialog.showPricingTable`, and emitting there would fire the event on every path that currently only fires `modal_opened`, changing existing funnel volumes well outside this ticket. `SubscribeButton` now also names its own surface (`reason: 'subscribe_now_button'`, the same value it already passes to the dialog it opens). Strictly additive — the field was previously absent on that event — and it makes `reason` total across both emitters, so the funnel can segment the subscribe leg rather than reading "not agent" as an absent field. Flagging it explicitly since it changes an existing event's payload for non-agent clicks. ## E3 — `billing.topup.succeeded` could not carry attribution `TopupBillingEvent` declared neither attribution field, so this was structural rather than a call-site omission. The field is added as **`payment_intent_source`**, matching `SubscriptionCheckoutBillingEvent`, `BillingOperationBillingEvent` and `ResubscribeBillingEvent`, rather than the spec's `entry_source`. The existing divergence is deliberate and kept: `entry_source` is the journey's own smaller enum (`CheckoutEntrySource`), a different value space. The value is fed from the operation record the poller **already holds**. `billingOperationStore` reads `operation.paymentIntentSource` for its `operation` and `subscription_checkout` events on the same object; the `topup` branch simply never included it. Plumbing: `dialogService` → component `source` prop → `adoptPendingOperation` → `startOperation` → `operation.paymentIntentSource` → terminal `billing.topup.*`. The poller is the common path for a real payment — a synchronous `completed` is the exception — so the `adoptPendingOperation` hop is the one that matters for R4, and it has its own test. #18390 deferred this on the grounds that feeding the field meant editing `handleBuy` (cyclomatic 20), `handleSuccess` (16) and `handleFailure` (14), turning a telemetry change into a refactor. That cost did not materialise: adding a property to an object literal introduces no decision point, so the complexity delta is 0 and no extraction was needed. Diff-scoped Fallow is clean (see Testing). **Every emit site has its own falsifier.** All four store stages (`succeeded`, `failed`, `reconciliation_needed`, `timeout`) and all five dialog sites (`started`, both synchronous failures, `succeeded`, `reportPurchaseError`) were mutated one at a time; each now fails its suite. The dialog cases assert the whole projected (stage, source) collection, so a source dropped from any single site cannot hide behind the others. **Correction.** An earlier revision of this description said `timeout` was not directly asserted because driving one "costs the full top-up poll budget in fake time". That was wrong — the multi-hour budget is the `awaiting_invoice_payment` parked case; a plain pending top-up times out at `TIMEOUT_MS` (two minutes), so the row is cheap and is now included. Two paths deliberately out of scope, both pre-existing: - **The legacy top-up rail** (`TopUpCreditsDialogContentLegacy.vue`) receives only `isInsufficientCredits`; `dialogService` does not forward `source` to it. It emits no `succeeded` event at all, so it is not on R4's path. - **The SDK rail** (`billingSdkStore`) emits `billing.operation.*` only, never `billing.topup.*` — `toBillingTelemetryEvent` has no topup case, and `billingSdkStore.ts:225` additionally drops non-resumed topup lifecycle telemetry. Its `operation` events have carried `payment_intent_source` since #13363. **Consequence, raised by CodeRabbit and confirmed pre-existing:** a pending top-up on the SDK rail can finish without any `billing.topup.succeeded`, so it has no terminal top-up attribution to carry. That was already true on `main` before this change and is unaffected by it — `adoptPendingOperation`'s early return on that rail is deliberate, since the SDK lifecycle adopted the operation when the command was issued and a second registration would mean two pollers on one operation. Closing it means adding new `billing.topup.*` emission to the SDK rail plus guarding against both rails emitting for one operation; that belongs in its own PR. ## The rehydration trap `agent_paywall` has to survive hosted checkout: with embedded checkout off the user leaves for Stripe and the journey resumes from storage through `normalizeRecord`, which is the path that produces the `.succeeded` events R4 reads. The `satisfies Record<T, true>` hardening on `ENTRY_SOURCES` / `ENTRY_FLOWS` / `UI_MODES` is **unchanged**, and verified still load-bearing: adding an unregistered `CheckoutEntrySource` member fails `vue-tsc` at `checkoutJourney.ts:44` with `Property 'unregistered_probe' is missing in type ... but required in type 'Record<CheckoutEntrySource, true>'`. The existing table-driven rehydration test is **extended** rather than paralleled. It already pinned every entry source from a persisted hosted journey into the resumed telemetry context; it now continues through `bindOperationToCheckoutJourney`. That is the hop R4 joins on — binding rewrites the persisted record and every later phase re-reads it, so an entry source lost there would strand `operation_linked`, and with it the `billing_op_id` join, on `'unknown'` while `entered` still read true. ## E4 `allowance_exhausted` — still blocked, unchanged Not addressed here, and not addressable in the frontend. The agent gratis allowance is a Metronome scoped credit at a burn priority; once spent with no paid credits the frontend sees `hasFunds: false` and a zero balance, indistinguishable from a plain out-of-credits user. @huntcsg — this still needs the answer #18390 asked for: will billing-status expose `funding_class` or allowance-remaining client-side, in what shape, and when? Please align it with PM-1467 (A11/A12) and PM-1484 (M3) so one field serves all three. ## Testing All figures measured locally on `135c30a336`. `pnpm typecheck` exit 0 across all projects. `pnpm lint` **0 errors**, 198 warnings (all pre-existing `import-x/no-named-as-default-member` in untouched files). `pnpm format` clean. Pre-commit hooks (oxlint type-aware, eslint, `typecheck:app`) passed on commit. **45 files / 1097 tests passed**, 0 failed, across the affected suites: ``` Test Files 45 passed (45) Tests 1097 passed (1097) Duration 142.09s ``` covering `AgentPanelRoot.test.ts` (308), `billingOperationStore.test.ts` (127), `checkoutJourney.test.ts` (51), `TopUpCreditsDialogContentWorkspace.test.ts` (43), `SubscribeButton.test.ts` (2, new), `useTopupOperation.test.ts`, and all of `src/platform/telemetry/`. Every new test was mutation-verified non-vacuous: | Mutation | Result | |---|---| | Remove the agent `trackSubscription` emit | both E1 cases fail (`Number of calls: 0`) | | Remove `reason` from `SubscribeButton` | its case fails; the dialog-source case stays green | | Remove `payment_intent_source` from the store's topup `succeeded` | that case fails; the two failure-stage cases stay green | | Drop `paymentIntentSource` from the `adoptPendingOperation` hand-off | the poller-hop case fails | | Lose `entry_source` at the bind hop only | 6 of 7 rehydration rows fail — the 7th is `unknown`, which the mutation cannot distinguish | The `checkoutJourney.test.ts > does not resurrect a journey whose removal from storage failed` case that #18390 reported failing locally on Node 26 **passed** here; all 51 in that file are green. `AgentPanelRoot.test.ts` still needs `--localstorage-file` to run locally on Node 26. **Not run locally: E2E / Playwright.** Nothing in this diff was exercised by a browser test here — those only run in CI, and no claim is made about them. ## Backport Production serves a SHA pinned on `cloud/1.54`, so this is not landed until it is backported. Labelled `needs-backport` + `cloud/1.54`. **Do not approve or merge this PR on my behalf** — `backport-auto-merge.yaml` squash-merges on approval, so approval is the release action.
…tor, card form and success (#19099) - Linear: FE-3046 **Merge order:** #19096 (ADR 0034 amendment) merges first. Until it does, ADR 0034 rule 4 still forbids account-ui from using the design system, which B does. ## Why billing-web is the cloud app's embedded checkout moved to a new tab, with no UX change (ADR BILLING-WEB-0038, #19096). Today billing-web renders a checkout of its own (`EmbeddedCheckout`, `CheckoutPayment`, `CheckoutSubmit`) that has drifted from the app's: different price layout, no saved-card selection, different copy. For both to render the same screens, the app's confirm, price breakdown, saved-card selector, card form, terms notice and success step have to live where both can import them. This PR adds them to `@comfyorg/account-ui`. Nothing renders them yet, so no visual change. ## How - New entry `@comfyorg/account-ui/billing/checkout`, ported line for line from the app components. Same markup, same classes, same conditions: - `CheckoutSubscribeConfirm` ← `SubscriptionAddPaymentPreviewWorkspace.vue` (plan header, credits, total due, discounts, promo code, saved methods, card form, 3DS prompt, parked-checkout recovery, terms). - `CheckoutTransitionConfirm` ← `SubscriptionTransitionPreviewWorkspace.vue` (immediate vs scheduled change, reactivation banner and threshold checkbox). - `CheckoutSavedMethods` + `CheckoutSavedMethodSelect` ← the saved-method row / `SingleSelect` picker (Reka select with the app's select classes). - `CheckoutPaymentForm` ← `UnifiedStripePaymentSelector.vue` minus the app bindings (key, theme and telemetry become props and a `phase` event). - `CheckoutSuccess` ← `SubscriptionSuccessWorkspace.vue`, with `details`/`actions` slots for the app's team invite. - `CheckoutTermsNote`, `CheckoutPromotionCode`, `CheckoutPaymentNotices`, `CheckoutVerificationPrompt`: blocks both confirms repeated. `CheckoutSubscribeSummary`/`CheckoutSubscribeActions` and `CheckoutReactivationBanner`/`CheckoutTransitionSummary` split the two confirm templates under the fallow complexity gate; the rendered markup is the same. - `checkoutQuote.ts`: the app's `subscriptionQuoteFormatting.ts` + `planDuration.ts`. - `@comfyorg/account-ui/billing/catalog` (TS-only entry): the tiers' prices and monthly credit grants (the numbers in the app's `TIER_PRICING`). The app's confirm reads the grant from this table (`TIER_PRICING`), not from the quote's `new_plan.credits_cents`, so billing-web reads the same table to show the same number; C points the app at it. A separate entry so non-Vue consumers (the Playwright runner imports `tierPricing`) never load the components. - The package has no i18n runtime. Hosts pass a typed `CheckoutCopy`; interpolated strings are functions, and sentences that wrap an element (terms links, reactivation amount) keep `{token}`s that the component splits. - Hosts pass a resolved `CheckoutPlan` (name, monthly/yearly price, credits, `pricedByQuote`) instead of the package reading the tier catalog. - `selectVariants` moves to `@comfyorg/design-system/select.variants` (same move as `buttonVariants` in #19098), so the picker renders the app's select classes. - Packaging: `design-system` variants and `tailwind-utils`'s `cn` are bundled into the published build (dev dependencies). `tailwind-utils` is not on npm and `design-system` ships TS source, so neither can be a runtime dependency of a published package. `reka-ui` is a peer. The rendered classes need the design-system theme and primeicons at runtime, so `@comfyorg/design-system` and `primeicons` are peers too, and the package description says to load their CSS and add the package's dist to the consumer's Tailwind sources. The smoke test imports the new entry in the typed consumer and treats it as bundler-only, like `billing/stripe`. ### Reactivation renewal date (live testcloud recording) The reactivation notice and the summary could name different renewal days for the same quote. On `m2-qa-0925-3ds` the notice said "renew … on Oct 27" and the summary "Renews … on Oct 25, 2026". - **Summary:** reads `renewal_at ?? new_plan.period_end` (`resolveRenewalDate`). - **Notice:** read `new_plan.period_end` only. When that was absent, it computed `effective_at` + one period on the client (Sep 27 + 1 month = Oct 27). - **Not a server inconsistency:** the server sent one renewal date, `renewal_at`, which the contract defines as "the next recurring charge and target-plan period end". - **Fix:** the notice now reads the same server date as the summary (`resolveRenewalDate`: `renewal_at`, else the plan's server `period_end`). The client-side computation is gone. With no server date, the notice drops its renewal clause through new `reactivation.withoutRenewalDate` copy, the way the summary drops its date. - **Contract:** `renewal_at` is optional in `PreviewSubscribeResponse` ("Optional in the schema for older clients"; "current servers always return this"). The server should always send it; the client no longer computes one when it doesn't. - **Tests:** - A new test fails on the old code (Oct 27 vs Oct 25). - The tests that pinned the client computation (account-ui and the app's `SubscriptionTransitionPreviewWorkspaceReactivation.test.ts`, in C) are replaced by a no-renewal-date test. - C adds the app's `withoutRenewalDate` strings. ## Review Focus - The ports are meant to be verbatim. C (the next PR) swaps the app onto these components and the app's existing component tests (about 2,100 lines across the two confirms and the success step) pass with only stub names changed. That is the parity check. - `CheckoutPlan.pricedByQuote` reproduces the app's price rules exactly, including the odd ones: a tier plan shows the quote's price once one arrives and `$0` on the success step without one; a team credit stop always shows its own price. ## Stack | PR | branch | base | title | |---|---|---|---| | #19097 | `jaewon/billing-web-remove-plan-grid` | `main` | `FE-3045 refactor(billing-web): drop the billing-web plan grid; hosts own plan selection` | | A #19098 | `jaewon/design-system-button-variants` | #19097 | `FE-3046 refactor(design-system): move buttonVariants into @comfyorg/design-system` | | **B (this PR)** | **`jaewon/account-ui-checkout-surfaces`** | **A** | **`FE-3046 feat(account-ui): shared checkout confirm, saved-method selector, card form and success`** | | C #19100 | `jaewon/workspace-adopt-account-ui-checkout` | B | `FE-3046 refactor(workspace): render the subscription checkout steps from account-ui` | | D #19101 | `jaewon/billing-web-shared-checkout` | C | `FE-3046 feat(billing-web): render the shared embedded checkout` | | E #19104 | `jaewon/billing-web-success-invite` | D | `FE-3046 feat(billing-web): invite your team from the checkout success step` | | F #19105 | `jaewon/checkout-3ds-failure-copy` | E | `FE-3047 fix(account-ui): clear decline toasts once a retry succeeds` | | G #19106 | `jaewon/billing-web-return-close-tab` | F | `FE-3048 fix(billing-web): return to the host by closing the checkout tab` | | H #19115 | `jaewon/checkout-cta-reachable` | G | `FE-3050 fix(account-ui): keep the checkout CTA reachable when notices grow the confirm step` | ```mermaid graph LR P0[#19097: remove plan grid] --> A[A: design-system buttonVariants] A --> B[B: account-ui checkout surfaces] B --> C[C: workspace adopts account-ui] C --> D[D: billing-web shared checkout] D --> E[E: billing-web success invite] E --> F[F: stale decline toasts] F --> G[G: close the billing tab] G --> H[H: reachable checkout CTA] style B fill:#f9d71c,stroke:#333,color:#000 ``` ## AS IS / TO BE No rendered change: nothing renders these components in this PR. The parity screenshots are on C (#19100, app before/after, pixel-identical) and D (#19101, billing-web before/after next to the app). --------- Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
…or website and billing-web (#18874) > **Do not merge until [BE-17276](https://linear.app/comfyorg/issue/BE-17276) (C11b) settles the `web_session_probe` contract.** Until the server sends the key, every reader resolves to off. Nothing calls these readers yet, so merging changes no request; the hold is only so the contract can't drift under them. ## Why The website and billing-web learn a session visitor's `unified_web_session` only through C11 (cloud#10689). That means `GET /api/features` with `credentials: 'include'` and `X-Comfy-Client`, sent before the site knows the flag. We agreed to gate that request on a global runtime switch served to anonymous callers, `web_session_probe` (BE-17276, Maanil). Then the request and its CORS preflight don't exist until rollout, and turning them off needs no deploy. This PR adds the readers the website header (FE-2896), website credits (FE-2899) and billing-web (FE-2898) will use. ## How - **New entry `@comfyorg/account-core/webSessionFlag`.** - `readWebSessionProbe`: a plain anonymous `GET /api/features` with no credentials and no custom headers, so there is no preflight. It is on only for `web_session_probe === true`. - `resolveUnifiedWebSession({ probe, … })`: sends the credentialed read (`credentials: 'include'`, `cache: 'no-store'`, `X-Comfy-Client`) only when the probe resolves `true`. It returns on only for `unified_web_session === true`. - The probe never feeds into the result: it decides whether to ask, and the per-user answer decides behaviour, so no server state is assembled on the client. - Any failure (network, non-2xx, bad body, timeout) is off, and nothing throws into the apps. Importing the entry is inert, and there are no browser globals in core. - **billing-web.** It reads `web_session_probe` from the anonymous `/api/features` it already fetches for the Firebase config. - `fetchCloudFeatures` / `resolveWebSessionProbe` in `@comfyorg/account-core/firebase` share that memoized request; the timeout must stay 4000 ms to share it. - With the probe absent or false, billing-web's requests are identical to main. `readBillingWebUnifiedWebSession()` is cached per page load. - **website.** It has no anonymous features read today, only the signed-in Bearer read in `workshop-features.ts`. So its probe is one plain anonymous GET. - This is the one accepted flag-off request difference. It is cached per page load, lazy, and never blocks render. - The request is only sent once FE-2896 calls `readUnifiedWebSessionEnabled()`; until then the website is unchanged. - **`X-Comfy-Client`.** `COMFY_CLIENT` is not on main (it lives in #18716), so the value is defined once in this module with the same shape (`name/version`), and a note says it converges with #18716. - The cloud app is unchanged, and `src/platform/auth/webSessionFlagOff.test.ts` passes. ## AS IS / TO BE No UI yet; the evidence is the request tables pinned with `it.for`. | State | billing-web vs main | website vs main | | --- | --- | --- | | probe absent / false / `"true"` | identical | + one plain anonymous GET | | probe true, flag false | + one credentialed GET, result off | + anonymous GET + one credentialed GET, result off | | probe true, flag true | + one credentialed GET, result on | + anonymous GET + one credentialed GET, result on | These rows describe what happens once a consumer calls a reader; nothing calls one in this PR. The `"true"` string row is covered in the account-core tests. A second call sends nothing new. Mutations (each reverted, full suites rerun): | Mutation | Tests red | | --- | --- | | remove the probe gate | core 3, billing-web 3, website 4 | | add credentials to the probe | core 1, website 4 | | OR the probe into the result | core 1 | | drop the cache | billing-web 2, website 4 | | probe timeout ≠ 4000 on billing-web (breaks the shared fetch) | 5 | | truthy comparison in `configSource` | 1 | Checks: vitest passes for account-core 1101, billing-web 198, website 8924 and the flag-off harness 4. Typecheck, oxlint, eslint, oxfmt, knip, fallow (no issues in 12 files) and `checkPublishable` are all clean. For review: - Does ingest check the `X-Comfy-Client` value or only that the header is present? The value carries the package version. - If an anonymous features document ever carried the probe but no Firebase or Stripe fields, the memo would be evicted and refetched. The eviction rule is unchanged from main. ## Links - TDD: https://app.notion.com/p/3de6d73d365081d7a4c8f808c0a0384b (§14 OQ17, §15 C11, §18) - Action items: https://app.notion.com/p/3e56d73d3650812d962decc97fbd9ea3 - Linear: https://linear.app/comfyorg/issue/FE-2891 ; backend switch https://linear.app/comfyorg/issue/BE-17276 - Contract: Comfy-Org/cloud#10689 (C11), BE-17276 (C11b, not yet built) ## Where this PR sits in the plan | Ticket | Scope | Blocked by | State | | --- | --- | --- | --- | | **FE-2891 (this PR)** | flag (cloud) + website/billing-web readers | C11 cloud#10689 + BE-17276 probe (readers) | #18708 merged; readers #18874, do not merge | | FE-2892 | web session client (+ revoke-all follow-up) | C2a cloud#10589 for revoke-all | #18710 merged; revoke-all #18869, do not merge | | FE-2893 | ADR | none | #18707 merged | | FE-2894 | identity + boot rules | BE-17063 on testcloud | #18715, do not merge | | FE-2895 | `authorize()` | C5 cloud#10555 | #18716, do not merge | | FE-2897 | lifecycle, heartbeat | FE-2894, C1 | #18719, do not merge | | FE-2900 | lazy token mint from the session | FE-2895, C8 cloud#10585 | #18870, do not merge | | FE-2901 | non-web clients stay on tokens | — | #18711 | | FE-2903 | cloud app joins the lifecycle | FE-2897 | #18871, do not merge | | FE-2904 | cloud API client on `authorize()` | FE-2903, FE-2895, C5 | in progress | | FE-2905 | socket + media carry workspace | FE-2904, C6, C7 | next | | FE-2896 / 2899 / 2898 | website header, website credits, billing-web | FE-2891 readers, FE-2894/2895, C5, C11 | next | | FE-2906 | cross-origin E2E harness | — | #18718 (harness only) | | FE-2902 / 2907 / 2908 | platform, rollout, cleanup | other repo / ops / after ramp | not started | ```mermaid flowchart LR subgraph BE["BE · cloud ingest and infra"] C0["BE-17071 C0 preview CORS gate"] C1["BE-17061 C1 sessions table + revocable cookie"] C3["BE-17063 C3 GET /session, trusted origins, flag"] C4["BE-17062 C4 shared workspace resolution"] C5["BE-17066 C5 cookie on API: CSRF, Origin, workspace header"] C2["BE-17064 C2 revoke-all, purge, caps"] C6["BE-17067 C6 WebSocket on session"] C7["BE-17068 C7 scoped media"] C8["BE-17069 C8 token mint from session"] C9["BE-17065 C9 Redis cache + load test"] C10["BE-17072 C10 prod config, cleanup"] C11["C11 flag before sign-in (OQ17)"] B135["BE-17135 serve unified_web_session (server AND)"] I2["BE-17070 I2 edge limit, log + DNS audit"] end subgraph SDK["FE · account-core"] F2891["FE-2891 flag"] F2892["FE-2892 web session client"] F2894["FE-2894 identity + boot rules"] F2897["FE-2897 lifecycle, heartbeat"] F2895["FE-2895 authorize()"] F2900["FE-2900 lazy token mint"] F2901["FE-2901 non-web stay on tokens"] end subgraph APPS["FE · apps"] F2896["FE-2896 website header"] F2903["FE-2903 cloud lifecycle"] F2899["FE-2899 website credits"] F2904["FE-2904 cloud API client"] F2905["FE-2905 socket + media"] F2898["FE-2898 billing-web joins session"] F2902["FE-2902 platform"] end subgraph QA["QA · rollout · docs"] F2906["FE-2906 cross-origin E2E"] F2893["FE-2893 ADR + docs"] F2907["FE-2907 ramp + rollback"] F2908["FE-2908 cleanup"] end C0 --> C1 --> C3 --> C5 C3 --> C11 --> F2891 B135 --> F2891 C4 --> C5 C1 --> C2 C1 --> C9 C5 --> C6 & C7 & C8 C3 --> F2892 C1 --> F2892 F2891 --> F2894 C3 --> F2894 F2892 --> F2894 --> F2896 F2894 --> F2897 --> F2903 C1 --> F2897 F2892 --> F2895 C5 --> F2895 F2895 --> F2899 & F2904 & F2900 & F2901 & F2902 C5 --> F2899 & F2904 F2903 --> F2904 --> F2905 C6 --> F2905 C7 --> F2905 C8 --> F2900 F2894 --> F2898 F2895 --> F2898 C3 --> F2898 C5 --> F2898 F2903 & F2904 & F2905 & F2898 --> F2906 C3 & C2 --> F2906 F2906 --> F2907 C9 & C10 & I2 --> F2907 F2907 --> F2908 C10 --> F2908 classDef thisPr fill:#ffe08a,stroke:#b8860b,stroke-width:4px,color:#000 class F2891 thisPr ``` - Blocked by: BE-17276 (probe switch) for the contract; C11 cloud#10689. - Unblocks: FE-2896 (website header), FE-2899 (website credits), FE-2898 (billing-web). --------- Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
…ified_web_session (1/2) (#18880) > **Do not merge.** Part 1 of 2 for FE-2898 (part 2, the transport, is #18881, stacked on this). Gated on BE-17276 via the readers (#18874), on FE-2894 (#18715), FE-2895 (#18716) and C5 cloud#10555. Base is the readers branch; the diff also shows the FE-2894/FE-2895 merges. This PR's own change: 9 files, 299 non-test lines. ## Why billing-web signs in on its own Firebase login and mints a workspace token. With `unified_web_session` on, a visitor already signed in on Cloud should be recognised from the shared session without starting Firebase or signing in again. This part adds the one decision point and the session-backed identity and sign-in. Part 2 adds the workspace scope and the credentialed billing transport. ## How - **`billingWebAuth.ts`: one decision per page load.** It chooses `session-client` (today's path) or `web-session`. - The router guard starts the decision but never waits for it. Until the decision lands, every phase reads `pending`, which is exactly what main shows while Firebase hasn't answered. - A probe that is absent or false settles it as soon as the `/api/features` document arrives. That is the same document Firebase's config comes from, so Firebase starts at the same moment as on main. - The session path gets **800 ms**. After that, the page load stays on today's path, and a late `true` is ignored. - **`unifiedBillingSession.ts`** boots the FE-2894 identity (`@comfyorg/account-core/webSessionIdentity`). Firebase is only a lazy RememberedLogin proof, consulted after the server says there is no session. - **Sign-in controller port:** `SignInView` gets a port that follows the decision. If a Cloud session is found in time, the page is left the same way a restored login leaves it. - The router, `App` and `SignInView` are rewired to the decision point. Entry-workspace rebinding still mints on today's path only. ### Decisions for review 1. **Flag-on first paint.** A visitor with a live Cloud session briefly sees main's pre-config sign-in screen: disabled buttons and the "unavailable" notice. It lasts until the decision lands, at most 800 ms, and then the page leaves it. This is the price of never delaying the flag-off path. 2. **A different remembered Firebase user is not signed out.** While the session is live, nothing on billing-web asks Firebase for a token or mints from `/auth/token`. Identity, scope and every billing request come only from the session cookie. So a different remembered user has no path to any request. It is consulted only after the server reports no session, and a restore goes through the server's silent-restore rules. ## Flag-off parity `billingWebAuth.flagOff.test.ts` renders `App` with the real router: - **Features read hangs:** the page is on `/sign-in`, the notice is shown, and Firebase is not initialised. Same as main. - **Credentialed flag read hangs:** Firebase is not started at 799 ms and is started at 800 ms. A late `true` sends no `/api/auth/session` and doesn't re-initialise Firebase. - **Probe false:** Firebase is initialised well before the cap. The request tables are identical to main with the probe absent or false, and `src/platform/auth/webSessionFlagOff.test.ts` passes. Timing mutations each go red: no cap, cap at 5000 ms, the guard awaits the decision, a late `true` swaps the mode, a false probe waits for the cap. ## AS IS / TO BE | | AS IS | TO BE | | --- | --- | --- | | Flag off | Firebase sign-in | unchanged, same first render and notice | | Flag on, live Cloud session | Firebase sign-in form | recognised from the session in ≤ 800 ms, no Firebase | Screenshots will be added by editing this description once testcloud serves the probe. ## Links - TDD: https://app.notion.com/p/3de6d73d365081d7a4c8f808c0a0384b - Action items: https://app.notion.com/p/3e56d73d3650812d962decc97fbd9ea3 - Linear: https://linear.app/comfyorg/issue/FE-2898 ## Where this PR sits in the plan | Ticket | Scope | Blocked by | State | | --- | --- | --- | --- | | FE-2891 | flag (cloud) + website/billing-web readers | C11 cloud#10689 + BE-17276 probe (readers) | #18708 merged; readers #18874, do not merge | | FE-2892 | web session client (+ revoke-all follow-up) | C2a cloud#10589 for revoke-all | #18710 merged; revoke-all #18869, do not merge | | FE-2893 | ADR | none | #18707 merged | | FE-2894 | identity + boot rules | BE-17063 on testcloud | #18715, do not merge | | FE-2895 | `authorize()` | C5 cloud#10555 | #18716, do not merge | | FE-2897 | lifecycle, heartbeat | FE-2894, C1 | #18719, do not merge | | FE-2900 | lazy token mint from the session | FE-2895, C8 cloud#10585 | #18870, do not merge | | FE-2901 | non-web clients stay on tokens | — | #18711 | | FE-2903 | cloud app joins the lifecycle | FE-2897 | #18871, do not merge | | FE-2904 | cloud API client on `authorize()` | FE-2903, FE-2895, C5 | #18875, do not merge | | FE-2905 | socket + media carry workspace | FE-2904, C6, C7 | #18878, do not merge | | FE-2896 | website header | FE-2891 readers, FE-2894, C11 | #18877, do not merge | | FE-2899 | website credits | FE-2896, FE-2895, C5 | in progress | | **FE-2898 (this PR)** | billing-web on the session (A identity, B transport) | FE-2891 readers, FE-2894, FE-2895, C5 | A + B, do not merge | | FE-2906 | cross-origin E2E harness | — | #18718 (harness only) | | FE-2902 / 2907 / 2908 | platform, rollout, cleanup | other repo / ops / after ramp | not started | ```mermaid flowchart LR subgraph BE["BE · cloud ingest and infra"] C0["BE-17071 C0 preview CORS gate"] C1["BE-17061 C1 sessions table + revocable cookie"] C3["BE-17063 C3 GET /session, trusted origins, flag"] C4["BE-17062 C4 shared workspace resolution"] C5["BE-17066 C5 cookie on API: CSRF, Origin, workspace header"] C2["BE-17064 C2 revoke-all, purge, caps"] C6["BE-17067 C6 WebSocket on session"] C7["BE-17068 C7 scoped media"] C8["BE-17069 C8 token mint from session"] C9["BE-17065 C9 Redis cache + load test"] C10["BE-17072 C10 prod config, cleanup"] C11["C11 flag before sign-in (OQ17)"] B135["BE-17135 serve unified_web_session (server AND)"] I2["BE-17070 I2 edge limit, log + DNS audit"] end subgraph SDK["FE · account-core"] F2891["FE-2891 flag"] F2892["FE-2892 web session client"] F2894["FE-2894 identity + boot rules"] F2897["FE-2897 lifecycle, heartbeat"] F2895["FE-2895 authorize()"] F2900["FE-2900 lazy token mint"] F2901["FE-2901 non-web stay on tokens"] end subgraph APPS["FE · apps"] F2896["FE-2896 website header"] F2903["FE-2903 cloud lifecycle"] F2899["FE-2899 website credits"] F2904["FE-2904 cloud API client"] F2905["FE-2905 socket + media"] F2898["FE-2898 billing-web joins session"] F2902["FE-2902 platform"] end subgraph QA["QA · rollout · docs"] F2906["FE-2906 cross-origin E2E"] F2893["FE-2893 ADR + docs"] F2907["FE-2907 ramp + rollback"] F2908["FE-2908 cleanup"] end C0 --> C1 --> C3 --> C5 C3 --> C11 --> F2891 B135 --> F2891 C4 --> C5 C1 --> C2 C1 --> C9 C5 --> C6 & C7 & C8 C3 --> F2892 C1 --> F2892 F2891 --> F2894 C3 --> F2894 F2892 --> F2894 --> F2896 F2894 --> F2897 --> F2903 C1 --> F2897 F2892 --> F2895 C5 --> F2895 F2895 --> F2899 & F2904 & F2900 & F2901 & F2902 C5 --> F2899 & F2904 F2903 --> F2904 --> F2905 C6 --> F2905 C7 --> F2905 C8 --> F2900 F2894 --> F2898 F2895 --> F2898 C3 --> F2898 C5 --> F2898 F2903 & F2904 & F2905 & F2898 --> F2906 C3 & C2 --> F2906 F2906 --> F2907 C9 & C10 & I2 --> F2907 F2907 --> F2908 C10 --> F2908 classDef thisPr fill:#ffe08a,stroke:#b8860b,stroke-width:4px,color:#000 class F2898 thisPr ``` - Blocked by: FE-2891 readers #18874, FE-2894 #18715, FE-2895 #18716, C5 cloud#10555 - Unblocks: M2 billing-web checkout on the shared session --------- Co-authored-by: mattmillerai <7741082+mattmillerai@users.noreply.github.com> Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
…ession (#18877) > **Do not merge.** Gated on BE-17276 (`web_session_probe`) through the readers in #18874, and on the FE-2894 identity in #18715. Base is the readers branch; the diff also shows the FE-2894 merge. This PR's own change is 5 files: 154 non-test lines and 390 test lines. ## Why On the website, a visitor who is already signed in to Cloud through the shared web session should see that account in the header, without starting Firebase. Today the header always loads the Firebase account island. FE-2896 lets the header take the account from the web session when `unified_web_session` is on for that visitor, and changes nothing when it is off. ## How - **`resolveWorkshopAccountSource()`** (`apps/website/src/config/workshop-web-session-identity.ts`) settles once per page load to `'session'` or `'firebase'`. - The probe (#18874) is absent or false, or the per-user flag is off: `'firebase'` at once. No session read is sent. - The flag is on: the FE-2894 identity boots. A live session gives `'session'` and a signed-out answer gives `'firebase'`. - **Cap: at most 800 ms.** If the cap fires first, the answer is `'firebase'` for the page load and the identity is disposed. A late answer never swaps a mounted header, and a session read is never sent after the cap. - **`HeaderMain.vue`**: the async loaders for `HeaderAccount` and `BuyCreditsDialog` wait for the source. On `'session'` they mount `HeaderSessionAccount` (an initials chip labelled with the email) and an empty dialog, so no Firebase island starts. - **RememberedLogin port**: the proof comes from Firebase only if Firebase is already loaded. The website never force-loads Firebase to build a proof. - **User id**: the Comfy user id is the Firebase uid today (`token.UID` in cloud `firebase_auth.go`). This holds until Phase 2.5. ### Kept out of this PR - **Session-mode header**: no menu, sign-out, credits or workspace switcher. Sign-out is FE-2897. Credits (FE-2899) and the workspace come next. Asking to buy credits in session mode does nothing until FE-2899. - **Playground and `/login`**: these still start Firebase. A stale Firebase login for a different user is signed out only if Firebase was already loaded when the session answered. - **SDK follow-up**: `readWebSession` sends no `X-Comfy-Client`. If C5 requires it on `/api/auth/session`, the fix belongs in account-core. ### /models/ script budget `e2e/models-performance.spec.ts` caps the JavaScript that `/models/` loads. This PR raises that ceiling from 1,810,000 to 1,815,000 bytes. - main measures 1,808,088 bytes, 1.9 KB under the old ceiling. - The header's once-per-page-load session decision ships even with the flag off: the probe read, the flag read and the 800 ms cap. - Two changes already moved code behind the flag: - the session boot now loads lazily; - `COMFY_CLIENT`, which inlines the package manifest, now loads only for the credentialed read. - With both, this branch measures 1,811,181 bytes in CI (an extra ~3 KB, down from ~8 KB). - The catalogue-only delta check (900,000) is unchanged. ## Flag-off parity - Probe absent or false: the same islands mount, and no `/api/auth/session` request is sent (`HeaderMain.webSession.test.ts`). - The header's only new wait is the probe read. It ends when the probe answers, and it never lasts past 800 ms. - `src/platform/auth/webSessionFlagOff.test.ts` passes. The cloud app is not touched. - Mutation checks: 10 mutations, all red. They include: the cap never decides, the session boots after the cap, the flag-off path waits for the cap, the header is ungated, and the boot ignores the flag. ## AS IS / TO BE | | AS IS | TO BE | | --- | --- | --- | | Flag off | Firebase header | Firebase header, unchanged | | Flag on, live session | Firebase header | Session chip (screenshot to follow once testcloud serves the probe) | Screenshots will be added by editing this description. ## Links - TDD: https://app.notion.com/p/3de6d73d365081d7a4c8f808c0a0384b - Action items: https://app.notion.com/p/3e56d73d3650812d962decc97fbd9ea3 - Linear: https://linear.app/comfyorg/issue/FE-2896 ## Where this PR sits in the plan | Ticket | Scope | Blocked by | State | | --- | --- | --- | --- | | FE-2891 | flag (cloud) + website/billing-web readers | C11 cloud#10689 + BE-17276 probe (readers) | #18708 merged; readers #18874, do not merge | | FE-2892 | web session client (+ revoke-all follow-up) | C2a cloud#10589 for revoke-all | #18710 merged; revoke-all #18869, do not merge | | FE-2893 | ADR | none | #18707 merged | | FE-2894 | identity + boot rules | BE-17063 on testcloud | #18715, do not merge | | FE-2895 | `authorize()` | C5 cloud#10555 | #18716, do not merge | | FE-2897 | lifecycle, heartbeat | FE-2894, C1 | #18719, do not merge | | FE-2900 | lazy token mint from the session | FE-2895, C8 cloud#10585 | #18870, do not merge | | FE-2901 | non-web clients stay on tokens | — | #18711 | | FE-2903 | cloud app joins the lifecycle | FE-2897 | #18871, do not merge | | FE-2904 | cloud API client on `authorize()` | FE-2903, FE-2895, C5 | #18875, do not merge | | FE-2905 | socket + media carry workspace | FE-2904, C6, C7 | next | | **FE-2896 (this PR)** / 2899 / 2898 | website header (this PR), website credits, billing-web | FE-2891 readers, FE-2894/2895, C5, C11 | next | | FE-2906 | cross-origin E2E harness | — | #18718 (harness only) | | FE-2902 / 2907 / 2908 | platform, rollout, cleanup | other repo / ops / after ramp | not started | ```mermaid flowchart LR subgraph BE["BE · cloud ingest and infra"] C0["BE-17071 C0 preview CORS gate"] C1["BE-17061 C1 sessions table + revocable cookie"] C3["BE-17063 C3 GET /session, trusted origins, flag"] C4["BE-17062 C4 shared workspace resolution"] C5["BE-17066 C5 cookie on API: CSRF, Origin, workspace header"] C2["BE-17064 C2 revoke-all, purge, caps"] C6["BE-17067 C6 WebSocket on session"] C7["BE-17068 C7 scoped media"] C8["BE-17069 C8 token mint from session"] C9["BE-17065 C9 Redis cache + load test"] C10["BE-17072 C10 prod config, cleanup"] C11["C11 flag before sign-in (OQ17)"] B135["BE-17135 serve unified_web_session (server AND)"] I2["BE-17070 I2 edge limit, log + DNS audit"] end subgraph SDK["FE · account-core"] F2891["FE-2891 flag"] F2892["FE-2892 web session client"] F2894["FE-2894 identity + boot rules"] F2897["FE-2897 lifecycle, heartbeat"] F2895["FE-2895 authorize()"] F2900["FE-2900 lazy token mint"] F2901["FE-2901 non-web stay on tokens"] end subgraph APPS["FE · apps"] F2896["FE-2896 website header"] F2903["FE-2903 cloud lifecycle"] F2899["FE-2899 website credits"] F2904["FE-2904 cloud API client"] F2905["FE-2905 socket + media"] F2898["FE-2898 billing-web joins session"] F2902["FE-2902 platform"] end subgraph QA["QA · rollout · docs"] F2906["FE-2906 cross-origin E2E"] F2893["FE-2893 ADR + docs"] F2907["FE-2907 ramp + rollback"] F2908["FE-2908 cleanup"] end C0 --> C1 --> C3 --> C5 C3 --> C11 --> F2891 B135 --> F2891 C4 --> C5 C1 --> C2 C1 --> C9 C5 --> C6 & C7 & C8 C3 --> F2892 C1 --> F2892 F2891 --> F2894 C3 --> F2894 F2892 --> F2894 --> F2896 F2894 --> F2897 --> F2903 C1 --> F2897 F2892 --> F2895 C5 --> F2895 F2895 --> F2899 & F2904 & F2900 & F2901 & F2902 C5 --> F2899 & F2904 F2903 --> F2904 --> F2905 C6 --> F2905 C7 --> F2905 C8 --> F2900 F2894 --> F2898 F2895 --> F2898 C3 --> F2898 C5 --> F2898 F2903 & F2904 & F2905 & F2898 --> F2906 C3 & C2 --> F2906 F2906 --> F2907 C9 & C10 & I2 --> F2907 F2907 --> F2908 C10 --> F2908 classDef thisPr fill:#ffe08a,stroke:#b8860b,stroke-width:4px,color:#000 class F2896 thisPr ``` - Blocked by: FE-2891 readers #18874 (BE-17276 probe), FE-2894 #18715, BE-17063 on testcloud - Unblocks: FE-2899 website credits (next, stacked on this branch) --------- Co-authored-by: mattmillerai <7741082+mattmillerai@users.noreply.github.com> Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Closes FE-3060. Follow-up to #19074.
…icient credits (#19249) ## Summary Cloud refuses a workflow run from a workspace with no credits with **HTTP 429** and `{"error":{"type":"PAYMENT_REQUIRED","message":"Insufficient credits to queue workflows"}}`. The workflow client mapped errors by status alone, so the page told the user they were rate limited, and the failure was counted as a service failure (`rateLimit`) rather than an account condition (`noCredits`). ## Changes - **What**: On a failed Cloud response, read the bounded error body and classify `PAYMENT_REQUIRED` as `insufficient_credits` whatever the status. Unreadable or other bodies fall back to the existing status mapping, so real 429 rate limits still report `rate_limited`. ## Review Focus Found while running the workflow pages live against production: `POST /api/prompt` returned exactly this 429 for a key whose workspace had `has_funds: false`. The new table test covers the 429 credit refusal, a genuine 429, an unreadable 429 body and a plain 402; the credit-refusal case fails on the previous code. Browser regression (`e2e/workshop-workflow-runs.spec.ts`): `/api/prompt` answers 429 `PAYMENT_REQUIRED`; the page shows "Not enough credits. Add credits to continue.", never the rate-limit message, and submits once. It fails on the previous client and passes with this change. Validation: website typecheck, lint, and 4,182 website tests across config, workshop components, composables and scripts.
) ## Summary **Stacked on #18988 — base is `claude/ppformula-metrics`, not `main`.** That PR corrects PP Formula's vertical metrics; this one removes the remaining offset, which is not the font's. Together they put a button label exactly on its button's centre. Retarget this to `main` if #18988 lands first. `BrandButton` wraps its slot in a label box. When a call site passes bare text that box is exactly one line tall, but when it passes an element — which every button carrying an icon does — the box establishes an inline formatting context and its strut reserves descender space the label never occupies. The box comes out 24px for a 20px line, the button grows with it, and the label sits above the button's centre. ## Changes - **What**: `BrandButton.vue`'s label box becomes a flex container (`inline-block` → `flex items-center`), which has no strut to reserve that space. `e2e/button-height.spec.ts` asserts the label box is never taller than its own line. - **Breaking**: none. - **Dependencies**: none. The wrapper element itself stays. `pages/pixal3d-trellis2.astro` writes its run-button labels through it (`el.querySelector('span') ?? el`), and it is the box the button's `items-center` acts on. Nothing else in the repository selects it, and no `BrandButton` call site passes more than one top-level node, so making it a flex container changes no call site's layout. ## Verification Measured on `/download` at 1440px, `origin/main` and both branches built from the same checkout. "Should be" is the button's own padding + border + line height: | build | button height | should be | label vs button centre | | --- | --- | --- | --- | | `origin/main` | 57px / 59px | 52px / 54px | 3.5px high | | #18988 alone | 56px / 58px | 52px / 54px | 2px high | | **this branch (both)** | **52px / 54px** | 52px / 54px | **0** | The label box goes from 24px to 20px against a 20px line, on 4 of the 9 buttons the page has — the four whose label is wrapped in an element. The other five pass bare text and were already correct, which is why this never looked like a button bug. **The test is proven both ways** against built pages: on the base branch it fails with `DOWNLOAD DESKTOP / Expected: 20 / Received: 24`, and on this branch it passes. **Visual baselines did not move.** All four `website-e2e` shards are green, the `visual` project among them, so nothing this change touches is under a screenshot baseline — the four buttons it affects are on `/download`, which has none. An earlier revision of this description predicted the opposite; CI says otherwise and CI is right. `pnpm exec eslint apps/website`, `pnpm format:astro:check` and the pre-commit gate are clean. The `desktop` Playwright project was also run locally in full: 491 passed, and its 27 failures are all `workshop-*` and `models-*` specs that need `WORKSHOP_IN_BUILD=1` and its companions per `e2e/README.md` — that local build has no `/workshop` routes at all, and the same specs fail identically on the base branch. CI, which builds them, is green on all of them. ## Screenshots The change is 5px of height and a 2px shift, which a still frame carries badly — the numbers above are the evidence, and the preview is the place to see it. On `/download`, the three buttons under the hero are the ones to look at: their labels and icons sit on the button's centre line rather than slightly above it, and the buttons are correspondingly shorter. --------- Co-authored-by: Claude <noreply@anthropic.com>
…#19247) ## Summary Workshop apps become catalogue data, like models and workflows: each app is an `APP` line in `workshop-workflows.jsonl` plus a `workshop-display.json` entry, listed in the catalogue's Apps tab and served at `/models/apps/<slug>/` (`/models/apps/cinematic-studio/`, `/models/apps/reshoot/`), each under its own name. Part of FE-2737. ## Changes - **What**: - Catalog: `{"id":"apps/<slug>","type":"APP","app":"studio"|"reshoot"}` lines, parsed by `parseAppCatalog` (workflow parsing unchanged; IDs stay unique across both). Display schema accepts `type: "APP"`, `apps/` slugs, and root-relative media paths for thumbnails this site serves. - `AppWorkshopModel` (`type: 'APP'`, `appId`) in the catalogue data and `catalogue.json`; kept out of `[...slug]` pages and related-model lists. - `/models/apps/[app]` renders the app page from the same data (title and OG image from the entry); the app switcher moves between the two paths. `/cinematic-studio` becomes a noindex redirect that keeps `?ux=`/`?model=` and sends `?app=reshoot` to Re-shoot. - Catalogue: the Apps tab lists the app entries with the standard card (`data-kind="app"`, "Comfy app" provider line) when `workshop-apps-enabled` is on; tabs show only what a visitor can open (Workflows only with workflows, Apps only with the apps flag). The "coming soon" placeholder is gone. - Studio hub cards and the back link point at the new paths / `/models/?type=apps`. ## Review Focus - Places that treated "no `routerId`" as "workflow" now narrow on `type` (`searchText`, `WorkshopModelCard`, `WorkshopSearchPanel`, `ModelsPage`'s flag filter). - Adding an app = one jsonl line + one display entry + the page component the `app` key names (documented in `src/content/README.md`). Validation: full website unit suite (474 files), `typecheck:website`, eslint, knip, fallow; production build emits both app pages with their own titles and the redirect; `workshop-apps`, `workshop-workflows` and `workshop-visibility` e2e specs pass against the build.
… overrides (#18988) ## Summary Pangram Pangram redrew the weights we license with the caps centred in the face's own content box. This carries out #18609: the corrected files go in and the whole stand-in mechanism comes out — the `@font-face` metric overrides, the Safari fallback, and the marker classes that fallback acted on. ## Changes - **What**: - The six `apps/website/public/fonts/PPFormula-*.woff2` replaced. Measured from the file, the gap above the caps against the descent goes from 906/−306 against an ink cap height of 826 (22.6% of the em apart) to 1051/−231 (0.6% apart); the narrow cut goes from 888/−288 against 808 (20.8% apart) to 1031/−219 (0.4% apart). - **`ascent-override` / `descent-override` removed** from all seven declarations, with the comment above them. The faces now speak for themselves. - The Safari fallback is gone: the probe in `BaseLayout.astro` that set `ppformula-metric-fallback` on `<html>`, the two rules behind it in `global.css`, and the `ppformula-text-center`, `ppformula-text-center-sm` and `ppformula-narrow-text-center` markers across 36 components. - The three `src/platform/cloud/onboarding/assets/fonts/PPFormula-*.woff2` replaced. `onboarding/assets/css/fonts.css` declares them with no overrides at all, so onboarding carried the defect undiluted and corrected files fix it there with no CSS. - **`BaseLayout.astro` gains preloads for Light and Semibold**, taking the preloaded set from three faces to five (~134 KB to ~220 KB of woff2 at high priority, on every page the layout serves). This is a site-wide critical-path change; see Review Focus. - **`e2e/font-preload.spec.ts` is new** — one case per page across five routes, asserting that every PP Formula face rendered above the fold is preloaded with `crossorigin`. - `e2e/font-metrics.spec.ts` measures each declared face from its file and asserts no declaration carries an override — ascent, descent or line-gap. - The `visual-responsive.spec.ts-snapshots` baselines regenerated by CI. - **Breaking**: none. - **Dependencies**: none. ## Review Focus **The preload work is arguably its own change.** @benjcooley's review is right that the two added preloads and the new spec were not declared here, and that the preloads are unconditional in the layout while the need is per-page: `/pricing` is the page whose headline was rendering in the Light fallback, but `BaseLayout` serves 82 pages and all of them now pay for it. The Changes list above now declares it. Whether it *stays* here or moves to its own PR — together with the two gaps below and a `preloadFonts` prop so a page declares what it renders — is open; see the discussion on `font-preload.spec.ts`. **Two preload gaps this PR does not close**, both found in review and both confirmed: - `AuthLayout.astro` imports `global.css` and sets `font-formula`, but declares zero preloads. Under `font-display: optional`, `login`, `signup` and `forgot-password` render in the fallback on a cold load. - `PPFormula-Medium.woff2` (weight 500) is preloaded nowhere, while `font-medium` appears in 77 files. Both predate this PR. The new spec did not catch either, because its `PAGES` list is five hand-written routes; driving it from the layouts instead is the right fix and is part of the open question above. **The files are the foundry's own woff2.** An earlier revision of this branch carried repackages built from the OTF sources, because the foundry's first woff2 build was TrueType-flavoured where everything we ship is CFF — it dropped ~2,100 outline points per weight and added hinting that reshaped glyphs at screen sizes. The full corrected set has since arrived and its woff2 are CFF with **outlines and advance widths byte-for-byte identical to the OTFs**, verified per glyph across all nine styles, and independently re-verified in review. One metric moved with that swap, and not in our favour: NarrowSemibold was 1050/−243 in the repackages and is 1031/−219 in the delivered file, against an unchanged 808 cap height. By the measure above that is 0.1% apart becoming 0.4% apart. Through the browser's own canvas metrics — which is what renders and what the e2e case asserts — it reads 0.05% off centre either way, so this is below anything observable and not worth another round with the foundry. **Why the overrides had to go rather than stay.** Keeping them looked like the zero-risk option and is not: they pin the content box at 121.2% and discard the metrics the foundry now ships, and Safari through 26.6 ignores them — so with corrected files in place, keeping them leaves Safari and every other browser disagreeing about inline box height by 5.8% (9.9% narrow), a gap that does not exist today. #18609 called this out in advance. The reason not to keep them for layout safety does not hold either. Tailwind's preflight sets `line-height: 1.5` on `html` and 1,397 text-size utilities set their own, so across five pages **zero of 1,020 PP Formula text elements resolve to `line-height: normal`** — the content-box growth the overrides were holding back applies to nothing on this site. ## Verification Built `origin/main` and this branch from the same checkout and compared five pages at 1440px with animations and video suppressed. A control run (main against itself) is exactly zero on every row, so everything below is signal: | page | elements | moved >0.5px | worst top shift | worst height change | full page height | | --- | --- | --- | --- | --- | --- | | `/` | 495 | 15 | 1px | 2px | 8511 → 8511 | | `/pricing` | 479 | 41 | 1px | 1px | 7955 → 7955 | | `/about` | 230 | 7 | 1px | 2px | 6209 → 6209 | | `/download` | 270 | 220 | 2px | 2px | 6709 → 6707 | | `/customers` | 297 | 6 | 1px | 2px | 7570 → 7570 | Nothing moves more than 2px and no page changes height by more than 2px out of ~7,000. **Cap centring, measured element by element across the whole site.** Both builds, seven pages, every element that carries text in a PP Formula face — 1,093 of them — measured for how far its capitals sit from the centre of their line box: | | before | after | | --- | --- | --- | | exactly centred | 186 of 1,093 | **874 of 1,093** | | median offset | 3.57% of text size | **0%** | | largest offset anywhere | ½ pixel | ½ pixel | 729 elements improve, 323 are unchanged, and 41 move the other way — all 41 by that same half pixel, which is the smallest step the browser can draw. Nothing anywhere ends up further out than that. Measured in Chromium; WebKit was not available in that environment, so Safari rests on CI's `mobile-safari` project rather than on this measurement. **Browser-measured result.** Through the canvas metrics the e2e case uses, every standard face now sits 0.45% of the em off centre and the narrow cut 0.05%, against 11.3% and 10.4% before this branch. **Visual baselines.** A 1–2px shift on text clears the configured `maxDiffPixels: 100`, so the `visual` project's baselines had to be regenerated, as #18609 predicted. They could not be produced in the environment this was written in — it has Chromium 1194 while the visual project asks for the headless shell build 1228, and baselines rendered by a different browser build would be wrong for CI — so they were regenerated by the repository's own `/update-website-screenshots` workflow. Website E2E, the `visual` project included, is green on this branch's current head. **Test guard.** Measured from the bare file at 1000px with a tolerance of 10 units, the files on `main` come out at 213–228 and the files here at 4–6. The previous version of this spec measured each face *through* its overrides, which is why files 22% apart passed it for as long as they did. `pnpm exec eslint apps/website`, `pnpm typecheck:website`, `pnpm format:astro:check` and the pre-commit gate are clean locally. ## Screenshots Full-page comparisons of `/pricing` and `/download`, main above and this branch below, are visually indistinguishable at 1:1 — see the measurements above for what actually changed. The preview is the place to confirm the type on Safari, which is the browser this PR most changes. Closes #18609 --------- Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: github-actions <github-actions@github.com>
## Summary
Eric reported five Workshop workflows failing on comfy.org (staff flag).
Cloud's recorded job errors traced them to three causes: two conversion
errors in our published graphs, and partner nodes never receiving a
credential. With this PR, four of the five complete on production Cloud.
## Changes
- **What**:
- **replace-character**: both SCAIL segment-offset math nodes
(`213:184`, `262:241`) compute `(b - c) * (a - 1)` but were published
without `b` and `c`. Cloud error: `'b' is not defined`. In the upstream
template these come from the subgraph's `length` (81) and
`previous_frame_count` (5) inputs. The graph already carries those
values for the inputs' other consumers.
- **animate-reference-sheet, restore-archival-footage**: the API
conversion skipped the `FLOAT,INT` `frame_rate` widget, which shifted
the frame rate's 25 into `LTXVEmptyLatentAudio.batch_size` (upstream
widgets are `[97, 25, 1]`). The sampler then failed concatenating a
batch of 25 audio latents with 1 video latent (`Sizes of tensors must
match … Expected size 1 but got size 25`).
- **Partner nodes** (8 of 30 workflows: Gemini, Seedream, MiniMax, Bria,
HitPaw, Topaz, ByteDance): they failed with `Unauthorized: Please login
first to use this node.` Ingest only forwards a partner credential the
submission carries in `extra_data` (`partnernode_token.go`), which is
what the Cloud app sends (`src/scripts/api.ts`). Changes:
- Browser runs now send `auth_token_comfy_org`.
- CLI (`workflow_render`) runs and the API snippet send
`api_key_comfy_org`.
- The body is built per send, so a 401 retry carries the renewed token.
- Regression guards over all published graphs:
- every math expression variable is supplied;
- LTX audio latents use the same batch size as their video latents.
Both fail on exactly the four broken nodes without the data fix.
## Review Focus
- **Verification**: real production Cloud runs with the example inputs,
through `workflow_render` and an API key.
| Workflow | Before | After |
|---|---|---|
| character-turnaround | failed in 12s | succeeded in 40s, 3 outputs |
| animate-reference-sheet | failed in 52s | succeeded in 50s |
| restore-archival-footage | failed in 280s | succeeded in 152s |
| replace-character | failed | succeeded in 195s |
- **Not fixed: product-photo-to-video.** The auth error is gone, but the
Cloud worker dies (`ServiceError: RIP to the server your workflow was
running on`) around 200s, on two runs. The graph decodes three 1080p
MiniMax videos into frame tensors and concatenates them for the combined
video, 20+ GB at peak, which is likely out-of-memory. The fix is a
content decision (stitch at 768p, downscale before stitching, or drop
the combined video), left out of this PR.
- Static checks of all 30 graphs found no other dropped or shifted
values. Checked against Cloud's `object_info.json` and
`supported_models.json`, and against the upstream templates.
- Sending the session JWT in `extra_data` matches the Cloud app, and
ingest re-mints it at submit.
## Summary
A failed Cloud job reports its failing node and exception in
`execution_error`, and the site dropped it. So "The run failed" reports
(Eric's five workflows) needed live reruns to diagnose. `run_finished`
for a failed workflow run now records where the job broke.
## Changes
- **What**:
- `cloudWorkflowResult` reads `execution_error` from failed jobs only.
It exposes `failure: { nodeId, nodeType, exceptionType }`, and keeps a
value only if it's identifier-shaped (`^[\w.:+-]{1,120}$`). Anything
else is dropped.
- `run_finished` (`status: 'failed'`, Cloud workflow runs) gains
`failed_node_id`, `failed_node_type` and `cloud_exception_type`.
- The exception message, traceback, and current inputs and outputs are
never kept or sent. They can contain prompt text or raw provider errors,
which FE-2736 keeps out of routine logs.
## Review Focus
- **Privacy boundary:** only identifier-shaped node IDs, ComfyUI class
names and exception class names leave Cloud. The tests assert that
nothing from the message or traceback reaches the parsed run or the
analytics payload.
- **Split with Jim:** his scheduled workflow sweep reads the same fields
server-side from `GET /api/jobs/{id}`, so the two will corroborate.
- **No e2e test:** the change is analytics-only with no visible UI. Unit
tests cover the parser, including the identifier filter, and the
`run_finished` payload, and both are mutation-checked. The existing
browser tests exercise the same run flow.
## Summary Closing the Google or GitHub sign-in popup now shows the "window closed" warning and gives the buttons back right away, on comfy.org and the cloud login pages, instead of after Firebase's 8 to 10 second wait. ## Changes - Firebase waits a hardcoded 8 seconds after the popup closes because a closed window alone does not say whether the result was already handed over. The shared account package now watches that handover, so it knows without a timer: closed with no result means the visitor gave up; result received means the sign-in is finishing, however slow. - When the visitor gave up, sign-in fails at once with the same `auth/popup-closed-by-user` error Firebase raises later, so each app's existing toast, telemetry and buttons work unchanged, just sooner. - A result that still arrives late either finishes through the app's normal sign-in, or is dropped before Firebase uses it when another sign-in started, someone else signed in, or the page moved on. These rules live once in the package; the website and the cloud app only turn the watch on. - Breaking: none. Apps that do not opt in behave as before (billing-web). No new strings, no flag. ## Review focus and testing - The file to read is `packages/account-core/src/firebase/popupWatch.ts`: it subclasses Firebase's public popup resolver and reads two of its internals. If they change, nothing is watched and sign-in keeps Firebase's own timing; `popupWatchSdkContract.test.ts` fails on an SDK upgrade that changes them. - Each sign-in is keyed by its provider object, so a double click never ties the live sign-in to Firebase's stray popup. - [x] Staging, real Google popup, Chromium and Firefox: warning in 27 to 202ms on both apps (main: 8.4 to 10.1s), shown once. Proof table in the QA comment. - [x] Real Firebase SDK against the auth emulator: close, success, close then email, late result kept or dropped, retry, double click. - [x] Unit: account-core 1250, website 9257, app 24659 (the 18 failures fail the same on main); every new guard caught by a mutation (40/40). - [x] Website `auth-*` e2e 29/29; typecheck, lint, format, knip clean; fallow adds nothing. - [x] Complete a real Google and GitHub sign-in on the preview in Chrome, Firefox and Safari. ## Risks and open questions - This depends on Firebase internals behind a public class. Is that acceptable given the fallback? The public-API alternatives are redirect sign-in (needs a `/__/auth` proxy on comfy.org) or Google's own sign-in library (Google only). - Google hosts the popup page. If it ever closed before handing over the result, the worst case is the warning followed by a normal sign-in, never a wrong account. - A late result whose token exchange outlasts Firebase's own 8 second timer is still rejected by Firebase, as on main. - The cloud social buttons are not disabled while a sign-in runs, so a double click can let Firebase's stray popup sign the user in without the app's flow. Plain Firebase does the same today; worth a follow-up. - The website's watched identity skips `getAuth`'s emulator default and `authTokenSyncURL`; neither is used in this repo. - Not included: billing-web, and Platform, which can opt in once `account-core` is published past `1.0.0-alpha.0`. Fixes FE-3002. Replaces #18775.
…ix (#18516) ## Summary Pre-commit ESLint no longer runs `better-tailwindcss/enforce-canonical-classes`. The rule is autofixable, and the `CI: Lint Format` workflow already runs `pnpm lint:fix` and commits the result back to the PR branch, so a non-canonical class committed locally is canonicalized by CI's `[automated] Apply ESLint and Oxfmt fixes` commit rather than failing. ## Changes - **What**: the lint-staged ESLint command adds `--rule 'better-tailwindcss/enforce-canonical-classes: off' --report-unused-disable-directives-severity off`. The rule's per-process Tailwind warm-up (~4.5 s, inside `tailwindcss`'s `canonicalizeCandidates`) was the single largest fixed cost of a commit touching any `.vue` file. - **Breaking**: none. Non-canonical classes are fixed by CI's auto-fix commit (same-repo PRs) or reported by `pnpm lint` (fork PRs, merge queue) instead of at commit time. ## Review Focus - The severity flag is not optional: with the rule turned off, `eslint --fix` treats every `eslint-disable … enforce-canonical-classes` line comment as unused and deletes it (reproduced on a `.ts` probe; `<!-- eslint-disable -->` in Vue templates is handled by `vue/comment-directive` and was unaffected). `--report-unused-disable-directives-severity off` keeps them; the full run still reports genuinely unused ones. `scripts/lint-staged-config.test.ts` pins the two flags together. - An earlier revision added a pre-push hook that re-linted pushed files with the full config. It was dropped: the saving is small next to the `typecheck:app` step the same hook runs (measured here at 123 s and 4.4 GB peak RSS), and CI's auto-fix already covers the rule. Tracks FE-2769. <details> <summary><b>Measurements</b></summary> All numbers from one orb (16 threads), `--no-cache`, `/usr/bin/time -f "%e s %M KB"`, best of two runs unless noted. <details> <summary>Where the time goes</summary> Earlier ablation over `src/**/*.vue`, single thread: 48 s with the full config, 31 s without `enforce-canonical-classes`, 24 s without any `better-tailwindcss` rule. Of the ~17 s the canonical rule costs on a big run, ~4.5 s is a one-time cost in the first `canonicalizeCandidates` call inside `tailwindcss` 4.3.2 (not the plugin; Tailwind context creation itself is ~0.9 s). The rule is lazy, so a commit without class literals never pays it: | Files | full config | canonical off | |---|---|---| | 1 `.vue` | 7.5 s | 3.0 s | | 12 `.vue` | 8.7 s | 3.4 s | | 12 `.ts` stores | 2.27 s | 2.22 s | </details> <details> <summary>Pre-commit ESLint command, 8 sidebar <code>.vue</code> files</summary> | Command | Wall | Max RSS | |---|---|---| | `eslint --concurrency auto --no-warn-ignored <8 files>` | 8.45 s / 8.66 s | 839 MB / 847 MB | | same + `--rule '…enforce-canonical-classes: off' --report-unused-disable-directives-severity off` | 3.71 s / 3.47 s | 490 MB | </details> <details> <summary>Directive-deletion probe</summary> ```ts import { cn } from '@comfyorg/tailwind-utils' // eslint-disable-next-line better-tailwindcss/enforce-canonical-classes export const probe = cn('mt-2 mb-2 flex') ``` | Command | Directive after run | |---|---| | `eslint --fix --rule '…: off' probe.ts` | deleted | | `eslint --fix --rule '…: off' --report-unused-disable-directives-severity off probe.ts` | kept | </details> </details> Co-authored-by: Amp <amp@ampcode.com>
## Summary Cut the per-entry cost of the merge queue: skip suites a candidate cannot affect, shard the cloud Playwright project, build the website once, and stop running non-required checks on `merge_group`. ## Changes - **What**: - `changes-filter`: with `evaluate-candidate: 'true'`, every output (not just `should-run*`) is evaluated against the merge-group candidate's parent. Tests Unit, Tests E2E and Website E2E opt in; their gates (`test`, `e2e-status`, `website-e2e`) already emit whatever the verdict, so a docs-only or website-only entry no longer runs 73 jobs. - Tests E2E: the single `cloud` job (461 tests, ~20 min, the critical path of every entry) becomes a 4-way `playwright-tests-cloud-sharded` matrix. `merge-reports` merges chromium and cloud as a matrix and still publishes `playwright-report-cloud`. - Website E2E: one `website-build` job builds the Astro site and uploads `apps/website/dist/`; the four shards download it instead of rebuilding. - Removed `merge_group` from Ecosystem Matrix, Fallow, Binary Size Validation, Validate Action SHA Pins, Website Unit, Billing Web Unit and Billing Web E2E. None is in the ProtectMain ruleset (`test`, `lint-and-format`, `e2e-status`, `website-e2e`, `cla-assistant`), so their queue runs could never block a candidate and only contended for runners. They keep their `pull_request` and (where present) `push: main` triggers. ## Review Focus - The queue's biggest cost driver, bypass merges to `main` rebuilding all in-flight entries, is a process/ruleset issue and is not addressed here; see the Linear issue. - Flaky specs listed in the Linear issue are also out of scope for this workflow-only PR. - `changes-filter` now has one `mode` step that every output and both paths-filter steps key on. Callers without `evaluate-candidate` are unchanged (hard-true outside `pull_request`). - If `ecosystem-matrix` is later made a required check, its `merge_group` trigger must come back (noted in the workflow header). - Verified locally with yamllint (`scripts/cicd/check-yaml.sh`), actionlint 1.7.7, oxfmt and `check-comfyui-container-version.sh`. Linear: [FE-3061](https://linear.app/comfyorg/issue/FE-3061/merge-queue-is-slow-bypass-merges-reset-in-flight-groups-runner) --------- Co-authored-by: Amp <amp@ampcode.com>
…16) (#19073) *PR Created by the Glary-Bot Agent* --- Item 1 of the ranked options on [PM-1783](https://linear.app/comfyorg/issue/PM-1783/fix-widget-values-set-on-the-canvas-change-on-prompt-submit-while-the). Standalone by request, ahead of the larger catalog work. - Fixes PM-1716 ## Problem A widget edit the doc host refuses arrives on the wire and nothing in the UI reacts. The canvas keeps the value the human typed, the shared document keeps the old one, and a later run uses the document's. The divergence is silent — which is the actual complaint behind PM-1783 and PM-1785. `docFrameClient.ts`'s `DocOpFailure.code` was never matched against a vocabulary, so nothing fired. ## Change `rejectedOpNotice.ts` raises an error toast from the batch the sender already tracks, matching on the host's `code` — never its `message`, per the `comfy-multi-player` contract. Two details that are not obvious: - The applier raises `uncatalogued_widget_write` and `unknown_widget` for `add_node` too, so the widget-specific wording additionally requires the op the host *named* to be a `set_widget`. Everything else gets generic copy. `widget_out_of_range` is included as a fourth widget-write code. - An uncatalogued node refuses **every** edit it receives, so a slider drag would have stacked one sticky toast per rejection. Toasts are throttled per rendered message and expire; each distinct code reports to telemetry once per follower. ## Tests The sibling divergence test asserted no toast and no alert — it encoded the *absence* of this fix, so those two assertions are removed and its title loses "silently". What it actually pins (a rejected write is not rolled back; the document keeps the old value) is unchanged. The file header's claim that the code is "never matched against a fixed vocabulary" is updated for the same reason. `useAgentCrdtFollower.rejectedHumanAdd.test.ts` was an `it.fails` pin asserting exactly the `reportError` call this adds. It stayed red for a mechanical reason — the body read `send.mock.calls[1]` synchronously while `opCoalescer` defers the flush to a microtask. Left alone it was a CI landmine (`it.fails` passes only while the body throws) and it was the *only* test touching the new telemetry path. Now `async` with a `waitFor`, green, `.fails` dropped. 16 Vitest cases drive the real follower end to end. ## Verification | Gate | Result | |---|---| | `pnpm typecheck` | exit 0 | | `pnpm lint` | exit 0 (198 pre-existing warnings, 0 errors) | | `pnpm format:check` | exit 0 | | `vitest run src/workbench/extensions/agent/crdt/` | 964 passed, 1 failed | The single failure is `agentSubgraphDefinitions.test.ts` — `RangeError` in lib0's recursive `deepFreeze`. **Pre-existing**: it fails identically on a stashed clean tree and imports nothing this PR touches. ### Browser verification The cloud Playwright project now directly asserts the rendered error toast with a visible DOM locator in the existing divergence scenario. Local execution against the cloud distribution passed at this head with retries disabled: 2/2 tests in `agentDuplicateInsertOpaqueWidgetWrite.spec.ts`. Mutation-checking the expected rendered message made the scenario fail at `toBeVisible` because no matching toast element existed; restoring the expected rejection copy returned it to green. ## Review Three `deep-reviewer` rounds. The first caught that the applier raises the same codes for `add_node` (fixed by the op-kind gate). The last returned FAIL on the red pin above and on batch-level refusals: those arrive as `{ok:false, code}` with **no** `failed` object, and the adapter dropped the top-level `code`, so every one collapsed to `'unspecified'` — and since `reportedCodes` keys on that, one transient `overloaded` permanently suppressed the report for a later `catalog_mismatch`. Both fixed. One recommendation deliberately **not** taken: reducing the predicate to the op-type check alone. It would give widget wording to a `set_widget` refused for an unrelated reason, which two existing tests deliberately pin as generic, and the review rated it an optional cleanup. ## Companions - [cloud#10824](Comfy-Org/cloud#10824) — item 2, class-local catalog mismatch - [cloud#10823](Comfy-Org/cloud#10823) — item 4 prototype (its title is wrong by my mistake; see the correction comment on it) - [FE-3036](https://linear.app/comfyorg/issue/FE-3036/implement-adopt-widgets-values-named-end-to-end-and-demote-the-widget) — item 5, filed not implemented --------- Co-authored-by: Glary-Bot <glary-bot@users.noreply.github.com>
…te PRs (#19255) ## Summary Mar's Hub V2, staged as one reviewable phase. This combines the 23 PRs from her [#proj-comfy-hub post](https://comfy-organization.slack.com/archives/C0AEPRS8N74/p1790528696016959), including the latest head of #19004, while preserving her presentation and commit history. Review fixes remain isolated in #19256 above this PR. Design reference (not for review): #17804 · [V2 Hub preview](https://comfy-website-preview-pr-17804.vercel.app/hub/) ## Included **Independent:** #19031 Hub not Models · #18916 search with filter and sort · #18915 shorter workflows banner · #18979 filter chips · #19012 one "browse all" · #19016 shelf stops promising more · #18987 home hub workflows tab · #19025 Apps tab · #19090 filter named after what it filters · #19017 phone video controls · #19011 API tab endpoint · #19014 workflow header **Catalogue cards:** #18921 → #19015 **Workflow page V2:** #19009 → #19004 → #19008 → #19010 → #19029 **Run states:** #19005 → #19006 → #19007 → #19022 ## Integration with current main - Rebases the phase onto current `main`, including the canonical `/models/apps/<slug>/` routes and the separate Apps feature flag. - Keeps Mar's Apps shelf and card presentation while sourcing its entries and links from the canonical app catalogue. - Keeps Apps gated by `workshop-apps-enabled`; Workflows remain gated independently by `workshop-workflows-enabled`. - Keeps Mar's latest #19004 Details cleanup and single template-author attribution while retaining the later interactive graph work in #19008/#19010. - Resolves the combined `/models/` JavaScript budget to 1,820,000 bytes. ## Not included - #19241 (API tab rail): conflicts with #19011 and still needs a design call. - #19227 (draft), #17865 and #18306: not part of the Hub V2 set. ## Verification - Full website unit suite: 9,494 passed, 4 skipped. - Focused post-rebase app/catalogue suite: 98 passed. - Website typecheck: 0 errors. - Commit and push checks: formatting, lint, app/website typechecks, knip and Fallow passed. - `workshop-workflows-enabled` gates the home tab, catalogue tab, direct `?type=` URLs and detail pages without flashing gated content. --------- Co-authored-by: Mariana Martinho <mariana@comfy.org> Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: github-actions <github-actions@github.com>
…on canvas (#18969) ## Summary Re-authors the regression guard from #18702 — closed without merging — as a black-box `browser_tests/` case, so the live bug it covered (ask the agent to clear the canvas, the turn reports the tab is empty, the nodes stay on screen — PM-1500, PM-1504) has a test again. ## Changes - **What**: new `browser_tests/tests/agent/agentClearedWorkflowStaysCleared.spec.ts` (two cases) plus one `browser_tests/types/globals.d.ts` declaration. Test-only; no production code touched. ## Why re-authored rather than reopened #18702's base was `fix/agent-delete-all-scope-drop-retry` (#18404), which is not landing, and it asserted against `AgentCrdtProjection` and `EcsFollowerAdapter` — both deleted by #18700. Rebasing it would have restored a test onto a base that cannot merge, stated in code that is being removed. This states the same property where the user reads it: the rendered canvas. Every API it touches was checked against #18700 and survives — `agentConversationTest` / `conversationCase`, `addedNodeIds`, `runTurns`, `subscribeCount`, `VueNodeHelpers.nodes` / `getNodeLocator`, `Topbar`, and the `agent-rec-clear-workflow` fixture. It asserts nothing in `agentCrdtProjection`, `ecsFollowerAdapter`, `graphMutations`, `agentNodeMaterializer` or `agentSubgraphHostSlots`, and `globals.d.ts` is untouched by #18700, so there is no conflict there either. ## Review Focus **It closes a vacuous-pass hole, which is most of the point.** The recorded turn adds one node and then clears the workflow, so asserting only that the canvas ends empty cannot distinguish *the clear landed* from *neither operation did*. That is the failure mode flagged as PM-1755 — a recorded replay can skip a clear and still report a pass — and it is real here, measured, not hypothetical: | fault injected into the replay | `agentConversationReplay` (recorded case) | `agentDeleteAndClearWorkflow` on #18700 | this spec | | --- | --- | --- | --- | | none | pass | pass | pass | | clear frame withheld from the client (the live bug) | **fail** | — | **fail** | | whole turn's graph ops withheld | pass | pass | **fail** | The distinguishing assertion records every `data-node-id` the DOM mounts, so it reads only what the user sees and does not care which graph API removed the node. **This lands green — it is a pin, not a live repro, and the PR does not claim a fix is needed.** The mechanism #18702 drove the bug through was a mutation batch rejected because the app briefly could not resolve which workflow tab the edit was bound to. #18700 removes that mechanism rather than fixing it: `AgentPanelRoot.vue` there has no `getScope` and no `createGraphMutations`, and `graphMutations.ts` / `ecsFollowerAdapter.ts` are deleted. So there is no surviving surface to inject the race from, and #18404's `dropWorkflowScope()` — which reached into Pinia to delete `tab.activeState.id` — would be asserting against the very thing being deleted. What is worth holding across that change is the user-visible outcome, on both sides of it. **Second case, tab round trip.** Returning to the tab re-subscribes the follower, which replays the bound document over the tab's own snapshot; the case waits for that subscribe before asserting, so it is about the catch-up rather than about a canvas nothing has touched yet. `agentConversationReplay` never switches tabs, so this was unpinned. ## Verification Local, against a `DISTRIBUTION=cloud` dev server over a containerised ComfyUI backend with devtools: ``` PLAYWRIGHT_LOCAL=1 DISTRIBUTION=cloud pnpm exec playwright test \ browser_tests/tests/agent/agentClearedWorkflowStaysCleared.spec.ts --project=cloud ``` - `main` (9cb7f5b): 2 passed - #18700 (540a2b8), spec copied in unchanged: 2 passed - red/green table above reproduced by temporary, uncommitted edits to `replayResponse` in `agentConversationFixture.ts` - `oxfmt`, `eslint`, `oxlint --type-aware`, `pnpm typecheck:browser`: clean (pre-commit hook ran all four) ## Note for #18700 #18700 adds its own `browser_tests/tests/agent/agentDeleteAndClearWorkflow.spec.ts`. This is a **different file**, deliberately, so there is no add/add conflict on rebase. Its `clear_canvas` case is the one that passes vacuously in the table above; the two can coexist, or that case can be dropped in favour of this one. --------- Co-authored-by: Amp <amp@ampcode.com>
## Summary Harden the workflow reload failures in [the reported job](https://github.com/Comfy-Org/ComfyUI_frontend/actions/runs/36384870201/job/108808501844) and recurring flakes found across eight recent E2E runs, including [retry-only failures on a green run](https://github.com/Comfy-Org/ComfyUI_frontend/actions/runs/36384921740). ## Changes - Wait for the requested workflow to finish saving. Use a 30-second Chromium budget after independent reload tests exhausted 15 seconds across three CI runs. - Register canvas event listeners before clicking; wait for agent turn acceptance before injecting stream events. - Clear ProseMirror through keyboard input and disable transitions during exact screenshot comparisons. - Measure the current widget after layout, requesting a redraw if an asynchronous widget replacement has no coordinates yet. - Process modifier-key state synchronously so key release cannot erase redo before the queued history shortcut runs. - Preserve loaded model folders when deferred startup replaces the folder structure. Reuse the existing refresh logic rather than waiting out the race in tests. - Wait for rejected agent submissions to settle before checking duplicates. The existing expected-failure annotation remains; no test is newly skipped or marked as an expected failure. - Keep the mocked running job available across queue refreshes and wait for queue-backed cancel readiness before starting execution. - Hold a widget drag until the target row has crossed its reorder threshold, then verify the persisted order. - Await template-save persistence so the deletion test cannot mistake a late save response for its delete response. - Wait for the topbar menu's entrance animation before hovering, and select Vue node headers through locators rather than canvas coordinates. ## Review Focus The budget change affects the whole Chromium project, including `test.slow()` multipliers. Assertion timeouts and retries are unchanged. Browser assertions remain intact, and the save helper still permits overwrite confirmation. No sleeps or baseline updates. - Five deterministic unit regressions failed locally and in [test-only CI](https://github.com/Comfy-Org/ComfyUI_frontend/actions/runs/36400840370). After the fixes, `pnpm test:unit src/scripts/changeTracker.test.ts src/stores/modelStore.test.ts src/components/sidebar/tabs/ModelLibrarySidebarTab.test.ts` passed all 146 tests. - Local `--retries=0` checks passed: 60 history/model-library repetitions, 32 related browser regressions, 60 reload repetitions, 20 composer repetitions, 20 promoted-widget repetitions, 110 agent-panel/chip-removal cases, and 14 widget regressions. Earlier broader runs passed 71 core and 12 cloud cases with CI's default-workflow setting. - The duplicate-submit regression produced its intended failure in all 20 repetitions, with no unexpected passes or setup failures. - Cancellation and promoted-input reorder passed 40 repetitions with retries disabled, followed by all 14 cases in their two spec files. - All eight template cases passed 20 repetitions each with retries disabled: 160 passes across both renderers and import/export. - Theme switching and subgraph boundary-link conversion passed 40 repetitions with retries disabled, followed by all 11 related menu/subgraph cases. - App/browser typechecks, ESLint, oxlint, formatting, Knip, and the changed-file Fallow gate passed. - [Full E2E CI](https://github.com/Comfy-Org/ComfyUI_frontend/actions/runs/36413761653) passed. Inspected all seven reports: 2,647 expected outcomes, 10 existing skips, zero flaky tests, and zero unexpected failures. The known duplicate-submit expected failure reached its intended assertion without retrying. - [Full unit CI](https://github.com/Comfy-Org/ComfyUI_frontend/actions/runs/36413761147) passed on the same commit. - The subsequent selection-assertion follow-up uses `expect.poll` and passed 20 repetitions with retries disabled. The full CI results above precede this assertion-only change. - The model-reload review follow-up also preserves folders opened while replacement contents load. Its regression failed in [test-only CI](https://github.com/Comfy-Org/ComfyUI_frontend/actions/runs/36467500560), including both retries. The unchanged test passes after the fix, and the three targeted suites now pass all 147 tests locally. [Full unit CI](https://github.com/Comfy-Org/ComfyUI_frontend/actions/runs/36468888856) passed on this fix. The earlier E2E result predates this follow-up. The history fix addresses shortcut ordering only; it does not change how subgraph boundary measurements participate in graph equality. --------- Co-authored-by: Amp <amp@ampcode.com>
## Summary Reduce Vitest startup, module reloads, timer waits and CI elapsed time. Follow-up to #18931, with separate commits for each change. ## Changes - Split `frontend` and `tooling` projects. Tooling defaults to Node; two browser-backed fixtures retain happy-dom. All 1,712 files remain covered; the ecosystem matrix retains its own discovery and real timers. - Disable retries for two expected reconnect failures without shortening their two-second recovery checks. - Preserve #17008's real-timer default in AgentPanelRoot; advance tab-creation delays with local fake clocks. - Reuse agent-extension and workspace-UI imports with state and effect-scope cleanup. - Test dev-agent proxy policy directly, retaining one real Vite-config wiring test instead of eight subprocesses. - Run eight root shards and a separate package job concurrently. Package tests run two processes with three workers each; the required check rejects package failures, cancellations and unexpected skips. Merge root shard reports before checking thresholds; upload root and package coverage to Codecov. ## Review Focus Shared setup, network isolation, coverage thresholds, proxy credential forwarding and loopback-only binding remain unchanged. Observed unit-workflow timings, including setup/reporting, with one CI run per configuration: | Configuration | Elapsed | Runner-minutes | | --- | ---: | ---: | | [Four shards, sequential packages](https://github.com/Comfy-Org/ComfyUI_frontend/actions/runs/36343555639) | 9m46s | 28.0 | | [Six shards, parallel packages](https://github.com/Comfy-Org/ComfyUI_frontend/actions/runs/36346702973) | 6m49s | 30.4 | | [Eight shards, parallel packages](https://github.com/Comfy-Org/ComfyUI_frontend/actions/runs/36347238009) | 6m19s | 34.1 | Eight shards favor latency; six save 3.7 runner-minutes at an observed 30-second cost. These are individual CI observations, not guaranteed speedups. Package test steps fell from 24s to 18s and no longer delay the root tests' reporting job. Latest unit CI: 1,712 root files, 24,475 passing tests, 30 expected failures and 7 existing skips; package suites: 98 files and 1,586 passing tests. Coverage thresholds and all seven Codecov input paths passed verification. The full local suite passed with four workers and retries disabled before the CI-only follow-ups. Also checked shuffled AgentPanelRoot/module-reuse suites, the 87-file tooling project, session tests with retries enabled, matrix discovery/real timers, blob merging, app/script typechecks, lint, formatting and Knip. Inverting the proxy credential host restriction fails three tests, including the real-config wiring test. Worker checks on four allowed CPUs: two package processes averaged 16.5s/13.9s/14.0s with 2/3/4 workers each (two measured runs per setting). A 107-file root sample with CI shard/reporting settings passed at 4/6/8 workers in 101.8s/99.8s/101.4s, using 2.6/3.3/3.7 GiB. Root workers retain the CPU-based default; the small timing differences do not justify the memory increase. An earlier root screen omitted the shard environment variable and failed full-suite coverage thresholds; it is excluded from these CI-style timings. Rejected global within-file concurrency because it breaks shared Pinia state; the pure-Node cohort gained little. Rejected the AgentPanelRoot split prototype: it saved 28% locally but required roughly 775 duplicated fixture/helper lines or a broad mutable fixture API. --------- Co-authored-by: Amp <amp@ampcode.com>
…heckout util (#19250) ## Why `performTeamSubscriptionCheckout` POSTs `/api/billing/subscribe` with no confirmation. Its only caller was removed in #14998. Deleting the dead util prevents the FE-1656 charge-on-landing path from being reintroduced. ## Root cause n/a (cleanup); the original FE-1656 root cause was fixed in #14998. ## How - Deleted `src/platform/cloud/subscription/utils/teamSubscriptionCheckoutUtil.ts` and its own test file. - Dropped the eslint-suppressions.json entry scoped to that file. - Removed the now-dead `performTeamSubscriptionCheckoutUtil` mock and its two `not.toHaveBeenCalled()` assertions from `CloudSubscriptionRedirectView.test.ts`. The tests still prove the team deep link opens the pricing table with the team `initialCheckout` via the existing `showPricingTable` assertions. That view test has no fetch/billing-API mock to assert against directly, so "no subscribe request" for the team deep link is covered instead by `browser_tests/tests/dialogs/pricingTableDeepLink.spec.ts:770`, which asserts 0 subscribe POSTs before confirmation. ## Verification - `pnpm typecheck` — pass - `pnpm exec vitest run src/platform/cloud/onboarding/CloudSubscriptionRedirectView.test.ts` — 12/12 pass - `pnpm exec eslint` / `pnpm exec oxlint` / `pnpm exec oxfmt --check` on changed files — clean - `pnpm knip` — no new findings ## AS IS / TO BE n/a: no UI or behavior change; the deleted function has no production callers. This PR changes no visible behavior, so screenshots are not applicable.
…out_ui flag (#19248) ## Why Milestone 3 replaces the checkout on billing.comfy.org with the full-page checkout from the DES-1209..1213 guidebooks (FE-3018..3022). That work must ship dark and roll out per user without touching the Milestone 2 checkout customers use today. This PR adds the gate: billing-web reads the per-user `billing_web_checkout_ui` flag (`embedded` | `full_page`) from the authenticated Cloud `/api/features` and renders one of the two checkouts on the same `/v1/checkout` URL. Root cause: n/a (feature). ## How - `apps/billing-web/src/config/checkoutUi.ts` resolves the variant once per user per tab. It sends the session's Bearer token, pinned to the workspace the session already holds, with `cache: 'no-store'` and a 4 s budget. Anything but an explicit `full_page` is `embedded`: absent key, unknown value, anonymous visitor, error, timeout. A DEV-only `ff:billing_web_checkout_ui` override works like the host's (JSON-parsed). - `apps/billing-web/src/views/CheckoutRouteView.vue` shows the existing loading frame until the variant settles, then mounts exactly one of `CheckoutView` (unchanged) or the new `FullPageCheckoutView` shell. `settleCheckoutUi` makes `settled` terminal, so a customer never sees one checkout swap into the other. - `router.ts` maps the `checkout` intent to `CheckoutRouteView`. No guard or signature change. - The e2e mock Cloud answers the flag on `/features` only to an authenticated read. ## Tradeoffs - The switch lives in the view, not an async router guard. A guard would hold the sign-in page up to 4 s on every cold load and change `createBillingRouter` at 8 call sites. - The answer is memoized per user for the tab, failures included. A flag flip reaches an open tab only after reload. In exchange, a checkout never changes under the customer. - A bare `fetch`, not the billing transport. `/api/features` never returns 401 (a bad token gets 200 anonymous defaults), so the transport's remint and billing error decoding do not apply. ## Blast Radius billing-web `/v1/checkout` only. The backend does not register `billing_web_checkout_ui` yet (cloud `common/featuregates/flags.go`), so every real answer is `embedded` and customers see the Milestone 2 checkout after a brief loading frame. PostHog: created in staging, embedded 100%. ## Verification - `pnpm --filter @comfyorg/billing-web test:unit`: 233 passed. - `pnpm exec playwright test` in `apps/billing-web`: 31 passed, including `e2e/checkout-ui.spec.ts`. The spec covers flag absent (embedded), `full_page` (full-page shell, the flag read carried the session JWT, `/billing/preview-subscribe` never called), and a 500 on the authenticated read whose body still says `full_page` (embedded). - typecheck, oxlint, eslint, oxfmt, knip: clean. ## AS IS / TO BE AS IS:  TO BE (`embedded`):  TO BE (`full_page`):  TO BE (`full_page`, loading):  ## Where this sits in the plan | Phase | Item | Ticket | Status | | --- | --- | --- | --- | | Setup | Gate decision (D0), flag in PostHog staging | - | done | | Flag plumbing | BE registers `billing_web_checkout_ui` on `/api/features` (M3-4) | not filed | todo | | **Flag plumbing** | **billing-web flag reader + `/v1/checkout` switch + e2e (M3-5..7)** | **FE-3067** | **this PR** | | Guidebooks | Flow and entry (root) | FE-3018 | todo | | Guidebooks | Summary layout | FE-3019 | todo | | Guidebooks | Promo codes | FE-3020 | todo | | Guidebooks | Ending screens | FE-3021 | todo | | Guidebooks | 3DS and redirect methods | FE-3022 | todo | | Rollout | QA plan, staged rollout, delete M2 path and flag | - | todo | ```mermaid flowchart LR D0["Gate decision + PostHog flag"] --> THIS["FE-3067 flag reader + checkout switch"] BE["BE: register flag on /api/features"] -.enables full_page.-> THIS THIS --> F18["FE-3018 Flow and entry"] F18 --> F19["FE-3019 Summary"] --> F20["FE-3020 Promo"] F18 --> F21["FE-3021 Ending screens"] --> F22["FE-3022 3DS / redirect"] F18 --> F22 F20 --> QA["QA + rollout + delete M2 path"] F21 --> QA F22 --> QA style THIS stroke-width:3px ``` Plan: https://app.notion.com/p/3e96d73d3650813da0fdfe75e71b21d6
## Summary Behaviour and maintainability fixes from the completed review of Mar's Hub V2 (#19255) and PP Formula (#18988, #19023). #19255 is merged; this PR is rebased directly onto main. This does not redesign Mar's CSS or page layout. The styling edits only remove vertical nudges that compensated for the old font metrics, and the final review explicitly preserves Mar's collapsed provider-mark treatment. ## Changes **Hub V2** - Details graph controls work with mouse/touch, zoom stays centred, drag follows the pointer, and touch screens retain vertical page scrolling. - Graph JSON loads only after Details first opens; previews follow a changed workflow model and fetch its replacement graph. - Graph rendering uses subgraph names rather than UUIDs, omits unlinked widget-input sockets, and matches links by saved socket index. - Filter-chip Clear keeps Browse all open and chip removal preserves keyboard focus. - Preparation failures no longer claim a run may still finish or be billed. - App cards use the catalogue as their single source of routes and artwork, preserving Mar's Re-shoot image. - The unchanged workflow use-case pill now returns to the workflow's real filtered catalogue shelf. - Final regression coverage checks rendered graph nodes, delayed/single graph fetches, model replacement, the Apps-only gate, and functional shelf navigation. - Three unused translation keys are removed. **PP Formula follow-ups** - Removes obsolete sign-in-label and Download/GitHub icon nudges that misalign the corrected PP Formula faces. - Adds a typed per-page preloadFonts prop; /models/ preloads Medium, with browser coverage. ## Review decisions - Mar's provider mark remains collapsed by default and expands on hover/focus. Its accessible text remains available; the automated suggestion to alter the mobile card composition was deliberately declined. - Re-shoot remains the existing product decision from main and is controlled by workshop-apps-enabled. ## Verification - BILL's final re-review: READY at 0550e19; the rebased patch matches the reviewed patch line for line. - 196 focused regression tests pass after rebasing onto merged main. - Website typecheck passes with zero errors. - Commit hooks and Fallow report no introduced findings. - Earlier full website run on the same reviewed patch: 9,494 passed, 4 skipped.
…8871) > **Stacked on #18719 (FE-2897) → #18715 (FE-2894). Do not merge** until those merge and C1/C3 are on testcloud. ## Why With `unified_web_session` on, the cloud app must join the shared session. It creates the session only on a real sign-in, deletes it on sign-out, boots and heartbeats from it, and resets a tab whose account changed. It must also stop re-creating the old cookie on every token refresh, which would undo a sign-out made on another site (TDD §12, §18, F2). With the flag off, nothing may change. ## How - New `useCloudWebSessionStore` (`src/platform/auth/session/cloudWebSessionStore.ts`) wraps account-core's `createWebSessionIdentity`. Its host adapters: - `RememberedLogin`, backed by the Firebase identity; - `createWebVisibilityPort()` for the heartbeat; - `createWebCrossTabRefreshPort<WebSessionResult>()` for the cross-tab leader. - **`start()` is the only place that reads the flag.** It reads it once per page load, on the first call, and returns whether the session is on. - The first call comes from the workspace gate or the extension's `onAuthUserResolved`, both of which run after the signed-in `/api/features` read. So the flag is never read in its anonymous default. - A later flag change applies on the next page load. - **Flag on**, the existing `useSessionCookie` entry points switch over: - `createSession` (called on user-resolved and on token refresh) starts the lifecycle and sends nothing, so the re-create-on-refresh behaviour is gone. - `ensureSessionCookie` waits for the lifecycle to be ready. - `deleteSession` does nothing, because sign-out is owned by `authStore.logout` → `signOut()` → DELETE. - `createSessionOrThrow` (OAuth resume, consent redirect) waits only if the lifecycle is already active. - `login`, `register`, `loginWithGoogle` and `loginWithGithub` call `signedInInteractively`, the only thing that creates a session. - **Account change.** The store clears the workspace selection, resets the team workspace store and the socket, and, only when the account became a different user, shows the `auth.accountChanged` notice (a notice, not a confirm, per open question 11). - Until FE-2904 moves the API client onto the session, the app still needs a Firebase login. So a different user ends in a signed-out page, not an in-place switch, and the copy does not promise a switch. - **Flag off.** `start()` returns false, every entry point falls through to today's code, and `webSessionFlagOff.test.ts` is unchanged and passes (4/4). - **For review: user id equality.** `currentUserId` returns the Firebase uid, which equals `comfy_user_id` today (cloud's Phase 2 invariant in `identity_resolution.go`). Cloud names a Phase 2.5 where they can diverge; the boot rule would then sign out a valid login, so the adapter must change before that phase. ## AS IS / TO BE No UI change with the flag off. With the flag on, the evidence is `webSessionFlagOn.test.ts`: | Scenario | Result | | --- | --- | | interactive sign-in | one session POST | | token refresh | no session request | | sign-out | one DELETE | | boot, 200 same user | GET only | | boot, 200 different user | local Firebase sign-out | | boot, 401 `session_revoked` | signed out, no restore | | boot, 401 `no_session` | one restore | | heartbeat sees a different user | socket and workspace reset, plus the notice | Nine mutations, each reverted with the full suite rerun, all go red, including "ignore the flag", which turns 4 flag-off rows red. Checks: - vitest over auth, stores, extensions, cloud and workspace: 5172 tests pass. - account-core: 1114 tests pass. - typecheck, knip, lint and fallow are clean. ## Links - TDD: https://app.notion.com/p/3de6d73d365081d7a4c8f808c0a0384b (sections 12, 15 F2, 18) - Action items: https://app.notion.com/p/3e56d73d3650812d962decc97fbd9ea3 - Linear: https://linear.app/comfyorg/issue/FE-2903 ## Where this PR sits in the plan | Ticket | Scope | Blocked by | State | | --- | --- | --- | --- | | FE-2891 | flag (cloud) + website/billing-web readers | C11 cloud#10689 + probe switch (readers) | #18708 merged; readers in progress | | FE-2892 | web session client (+ revoke-all follow-up) | C2a cloud#10589 for revoke-all | #18710 merged; revoke-all #18869, do not merge | | FE-2893 | ADR | none | #18707 merged | | FE-2894 | identity + boot rules | BE-17063 on testcloud | #18715, do not merge | | FE-2895 | `authorize()` | C5 cloud#10555 | #18716, do not merge | | FE-2897 | lifecycle, heartbeat | FE-2894, C1 | #18719, do not merge | | FE-2900 | lazy token mint from the session | FE-2895, C8 cloud#10585 | #18870, do not merge | | FE-2901 | non-web clients stay on tokens | — | #18711 | | **FE-2903 (this PR)** | cloud app joins the lifecycle | FE-2897 | #18871, do not merge | | FE-2904 | cloud API client on `authorize()` | FE-2903, FE-2895, C5 | in progress | | FE-2905 | socket + media carry workspace | FE-2904, C6, C7 | next | | FE-2896 / 2899 / 2898 | website header, website credits, billing-web | FE-2891 readers, FE-2894/2895, C5, C11 | next | | FE-2906 | cross-origin E2E harness | — | #18718 (harness only) | | FE-2902 / 2907 / 2908 | platform, rollout, cleanup | other repo / ops / after ramp | not started | ```mermaid flowchart LR subgraph BE["BE · cloud ingest and infra"] C0["BE-17071 C0 preview CORS gate"] C1["BE-17061 C1 sessions table + revocable cookie"] C3["BE-17063 C3 GET /session, trusted origins, flag"] C4["BE-17062 C4 shared workspace resolution"] C5["BE-17066 C5 cookie on API: CSRF, Origin, workspace header"] C2["BE-17064 C2 revoke-all, purge, caps"] C6["BE-17067 C6 WebSocket on session"] C7["BE-17068 C7 scoped media"] C8["BE-17069 C8 token mint from session"] C9["BE-17065 C9 Redis cache + load test"] C10["BE-17072 C10 prod config, cleanup"] C11["C11 flag before sign-in (OQ17)"] B135["BE-17135 serve unified_web_session (server AND)"] I2["BE-17070 I2 edge limit, log + DNS audit"] end subgraph SDK["FE · account-core"] F2891["FE-2891 flag"] F2892["FE-2892 web session client"] F2894["FE-2894 identity + boot rules"] F2897["FE-2897 lifecycle, heartbeat"] F2895["FE-2895 authorize()"] F2900["FE-2900 lazy token mint"] F2901["FE-2901 non-web stay on tokens"] end subgraph APPS["FE · apps"] F2896["FE-2896 website header"] F2903["FE-2903 cloud lifecycle"] F2899["FE-2899 website credits"] F2904["FE-2904 cloud API client"] F2905["FE-2905 socket + media"] F2898["FE-2898 billing-web joins session"] F2902["FE-2902 platform"] end subgraph QA["QA · rollout · docs"] F2906["FE-2906 cross-origin E2E"] F2893["FE-2893 ADR + docs"] F2907["FE-2907 ramp + rollback"] F2908["FE-2908 cleanup"] end C0 --> C1 --> C3 --> C5 C3 --> C11 --> F2891 B135 --> F2891 C4 --> C5 C1 --> C2 C1 --> C9 C5 --> C6 & C7 & C8 C3 --> F2892 C1 --> F2892 F2891 --> F2894 C3 --> F2894 F2892 --> F2894 --> F2896 F2894 --> F2897 --> F2903 C1 --> F2897 F2892 --> F2895 C5 --> F2895 F2895 --> F2899 & F2904 & F2900 & F2901 & F2902 C5 --> F2899 & F2904 F2903 --> F2904 --> F2905 C6 --> F2905 C7 --> F2905 C8 --> F2900 F2894 --> F2898 F2895 --> F2898 C3 --> F2898 C5 --> F2898 F2903 & F2904 & F2905 & F2898 --> F2906 C3 & C2 --> F2906 F2906 --> F2907 C9 & C10 & I2 --> F2907 F2907 --> F2908 C10 --> F2908 classDef thisPr fill:#ffe08a,stroke:#b8860b,stroke-width:4px,color:#000 class F2903 thisPr ``` - Blocked by: #18719 (FE-2897), #18715 (FE-2894), C1/C3 on testcloud. - Unblocks: FE-2904 (API client on the session), then FE-2905; FE-2906 rows E2E-02/05. --------- Co-authored-by: mattmillerai <7741082+mattmillerai@users.noreply.github.com>
## Summary Makes every published Cloud workflow run. Combines the remaining workflow fixes into one PR: #19284 (product-photo-to-video) and #19272 (remove-object, virtual-try-on). #19288's temporary disables are no longer needed because all three are fixed here. Together with #19251 (already merged), all 30 published workflows pass live on production. ## Changes - **product-photo-to-video** (Bob's commits from #19284): Cloud lost the worker at ~700 s decoding the three MiniMax videos to stitch them. The combined video now uses core `ConcatenateVideo` (joins VIDEO objects without decoding); the downloadable graph and SVG match. - **remove-object**: the example now includes a website-owned 1024×1024 mask over the apple (`public/workflow-inputs/remove-object-apple-mask.png`). - **virtual-try-on**: the example now includes the template's own subject (pinned workflow_templates revision). - **Tests**: every published workflow's example must validate in the shared form; a browser test opens remove-object and virtual-try-on and asserts their inputs are pre-filled. ## Review Focus Live evidence on production (`test:workflow-pages --execute`): - product-photo-to-video at `7c3f7304e4`: 3/3 passed, all 4 outputs incl. the combined video — runs `56635afd…` (191 s), `daff6b57…` (184 s), `98babfa5…` (213 s). - remove-object with this mask: run `1de167fa-7317-4963-b3aa-24e3535a016b`; virtual-try-on with this subject: run `33160fe7-ea1e-46a4-b4cf-e8531cc335f9`. - Every other published workflow passed in the full sweep after #19251. Post-deploy: the mask is served from comfy.org once this ships; run remove-object once through the site. Previously accepted by Bill as #19272 and #19284 (same commits). CI's `/models/` JavaScript-budget failure is fixed by #19256, which merges first. Validation: 7,377 website unit tests, website typecheck, 15 workflow browser tests on the CI-equivalent build. ## Why product-photo-to-video has no new browser test The change is inside the Cloud graph (`ConcatenateVideo` replacing the decode-and-stitch subgraph) and only has an effect when Cloud executes it. The website's Playwright suites stub Cloud's `/api/prompt` and `/api/jobs` (`e2e/workshop-workflow-runs.spec.ts`), so no browser test can observe whether Cloud keeps the worker alive or produces the combined video. The page, its form and its submission path are unchanged and remain covered by the existing workflow browser tests; the graph change is verified where it runs — three live production runs above (all four outputs, combined video included) — and by the scheduled live sweep in #19274.
## Summary Start selective within-file Vitest concurrency for FE-3055. Stacked on #19117; merge that PR first. ## Changes - Separate Node tooling setup from frontend Pinia, timer, and DOM hooks. Preserve the fetch guard and keep three script suites on frontend setup. - Define typed `concurrent-safe` and `shared-state` tags, keeping sequential execution as the default. - Run the binary-size suite's 20 cases with async subprocesses and per-test temporary repositories. Keep `useWorkspaceUI` explicitly sequential. - Dispose the Pinia created by setup even if a test clears active Pinia. Reject concurrent frontend tests and ancestor suites while setup still shares globals. - Check LiteGraph teardown within one test. Remove the agent-panel module reset that broke telemetry assertions under shuffle, preserving the import-side-effect check. ## Review Focus Tags control scheduling, not resource isolation. `shared-state` is not a cross-file lock. File isolation stays enabled. Frontend concurrency requires test-owned fixtures and explicit dependencies through nested stores and async callbacks; that work remains open in [FE-3055](https://linear.app/comfyorg/issue/FE-3055/enable-selective-vitest-concurrency-while-serializing-conflicting). Validation: - `pnpm test:unit --maxWorkers=4 --retry=0` passed 1,713 files and 24,509 tests, retaining seven existing skips. The LiteGraph teardown pair became one test. - The eight-file setup/store/agent cohort passed all 435 tests with one worker, no retries, and shuffle seeds 17, 83, and 701. Seed 17 previously exposed seven agent-panel failures. - The Pinia regression [failed in CI before the fix](https://github.com/Comfy-Org/ComfyUI_frontend/actions/runs/36358389892/job/108730584598); the [fixed setup run passed](https://github.com/Comfy-Org/ComfyUI_frontend/actions/runs/36358812400). [Latest unit CI](https://github.com/Comfy-Org/ComfyUI_frontend/actions/runs/36359348341). - A temporary self-registration mutation made the revised import test fail; restoring production code made it pass. Forced frontend concurrency rejected all 38 selected cases, including nested suites. - Mixed tagged suites passed four shuffled runs at verified concurrency limits 2 and 5 without retries. Filtered runs selected the expected 20 and 37 cases. App/scripts typechecks, changed-file lint/format, Knip, and the Fallow audit gate passed. The subprocess suite took 0.35–0.36s at concurrency 5 versus 0.86–0.95s at concurrency 1. End-to-end command time was 2.48–2.49s versus 3.03–3.15s, with similar peak-process RSS. Measurement used four pinned CPUs, one file worker, two warm-ups, then interleaved 1/5/5/1 runs. Vitest 4.1.11 does not propagate CLI `--maxConcurrency` into projects, so these comparisons and the mixed runs set `test.maxConcurrency` through a temporary config instead. Test-duration sums confirmed actual overlap. The original 87-file / 1,038-test tooling cohort took 20.16–20.36s after these changes versus 22.92–24.71s before on four pinned CPUs. These preliminary measurements include the setup split, not just concurrency. Timing runs used a 30s CLI timeout because the unchanged baseline Astro test exceeded its normal 5s timeout twice; committed timeouts are unchanged. No Playwright regression is included because Playwright does not execute Vitest setup hooks or the per-test Pinia lifecycle. The Vitest setup tests exercise exact-instance disposal and cleanup after active Pinia changes, including the linked failing-before/passing-after CI runs. --------- Co-authored-by: Amp <amp@ampcode.com>
…18945) ## Summary Website components fill translation placeholders with named values passed to `t()` instead of calling `String.replace` on the returned copy, so the storage and runtime swap in #18013 no longer carries a 50-file call-site migration. ## Changes - **API**: `t(key, locale, named)` interpolates `{name}` placeholders; a value the caller does not pass keeps its `{name}` visible, and values are inserted literally (no `$&` expansion). `tAround(key, locale, slot, named)` splits the unresolved message around one slot, then fills each side, so a heading can wrap that slot in markup; it throws unless the message contains that slot exactly once. The Cinematic Studio's `tc()`, the Reshoot `rc()` and the Router page's `routerT()` take the same named values through a shared `src/i18n/interpolate.ts`. - **Migration**: About 60 components, templates, utilities and the events/FDCT specs move off `.replace`, including every Cinematic Studio, Reshoot and Router call site. Label objects handed to child components carry functions instead of placeholder strings. The local interpolators in `auth-schemas.ts` and `modelFaq.ts` go. - **Messages**: The split `models.list.heroTitle.before`/`.after` pair becomes one message, `{name} in {brand}` / `{brand} 中的 {name}`, and the hero template no longer adds spacing of its own around the brand. `gallery.heroTitle.before` likewise becomes `gallery.heroTitle` with a `{brand}` slot, which stops the zh-CN gallery heading from repeating ComfyUI. - **zh-CN copy**: Product CTAs say 功能 instead of 属性, and the hub empty/count labels say 工作流 instead of 模板. - **Tests**: `src/i18n` joins unit-test coverage. `tc()`, `rc()` and `routerT()` each have a literal-output test for a parameterized message. ## Review Focus For current values, rendered output is unchanged apart from the two zh-CN wording fixes and the zh-CN gallery heading, which no longer says ComfyUI twice. A value containing a `String.replace` token such as `$&`, `` $` `` or `$'` now renders literally instead of expanding. The dictionary stays in `translations.ts`; #18013 moves it to JSON catalogs rendered with vue-i18n, which keeps the same `t()` signature and missing-value behavior. Part of FE-2045: one localization system for the monorepo (one JSON catalog format, one translation pipeline and workflow). --------- Co-authored-by: Amp <amp@ampcode.com>
…#18920) ## The gap `check-ai-co-authors.sh` reads `%(trailers:key=Co-authored-by)`, so it only sees what is written in the commit **message**. GitHub regenerates `Co-authored-by` trailers from the **authors** of the squashed commits at merge time. A branch with clean messages but AI-authored commits therefore passes the check and still lands the trailers on the squash commit. It is a guardrail that cannot see the artifact that lands. ## Evidence I hit this directly. Four backports failed the check, I rewrote their commit messages to strip the trailers (tree hashes identical, message-only), the check went green, and the squash commits came out with the trailers restored: | squash | trailers after merge | commit author | |---|---|---| | `7a91c98bf` | 1 | `claude[bot]` | | `6b182a122` | 1 | `claude[bot]` | | `a86aef141` | 2 | `Claude <noreply@anthropic.com>` | | `be9f7c239` | 2 | `Claude <noreply@anthropic.com>` | The control is the fifth PR in the same batch, #18728. Same message rewrite, but its commits are authored by humans, and its squash commit has **zero** trailers. That is the mechanism confirmed both ways. ## The change Adds an identity check over `%an <%ae>` and `%cn <%ce>` beside the existing trailer check, and fixes the remediation advice, which currently tells you to do the thing that does not work: ```diff - git rebase -i <base> # edit messages — trailers come back on squash + git rebase <base> --exec 'git commit --amend --no-edit --reset-author' ``` ## Scale, measured on `origin/main` The two paths are independent and both live: | window | AI-authored identities | AI trailers in messages | |---|---|---| | last 50 | 4 | 15 | | last 200 | 32 | 113 | | last 500 | 90 | 251 | So roughly a third as much again slips through the path the check is blind to. ## False positives None found. The last 12 commits on `main` are all human-authored and pass. Validated against the five real branches above: the four AI-authored ones exit 1, the human-authored control exits 0. This PR's own commit passes. ## Not included, deliberately `Check for AI agent co-author trailers` is **not a required check** on `main` or on any release branch. This PR only makes it see correctly; it does not make it block. Given 251 trailer hits in the last 500 commits on main, turning it into a gate is a separate policy decision with real in-flight cost, and it should be someone's explicit call rather than a side effect of a bug fix.
…rom account-ui (#19100) - Linear: FE-3046 ## Why #19099 moved the app's checkout steps into `@comfyorg/account-ui`. Until the app renders those components, the package holds a second copy that nothing keeps in step. This PR makes the app render them, so the app and (in D) billing-web share one implementation. No visual change in the app. ## How - `SubscriptionAddPaymentPreviewWorkspace.vue`, `SubscriptionTransitionPreviewWorkspace.vue` and `SubscriptionSuccessWorkspace.vue` keep their props and events, so their parents are untouched. Each now renders `CheckoutSubscribeConfirm`, `CheckoutTransitionConfirm` or `CheckoutSuccess` and binds what only the app knows: - copy from vue-i18n (`useCheckoutCopy`), including the `CheckoutPlan` built from the tier catalog or the team credit stop, - the Stripe publishable key, the colour-palette theme key and checkout-journey telemetry (previously in `UnifiedStripePaymentSelector.vue`), - subscription status from `useBillingContext` (loaded, cancelled, end date) for the reactivation banner, - the team invite block and Send invites action through `CheckoutSuccess`'s slots. - Deleted, now in the package: `UnifiedStripePaymentSelector.vue`, `SubscriptionTermsNote.vue`, `subscriptionQuoteFormatting.ts` and `planDuration.ts` (plus their tests, which the package's `checkoutQuote.test.ts` and component tests cover). `useSubscriptionCheckout` imports `amountDueTodayChanged` from the package. - `TIER_PRICING` spreads the shared `TIER_CATALOG` (same numbers) and keeps its app-only `videoEstimate`. - `UnifiedStripePaymentSelector.test.ts` becomes `SubscriptionAddPaymentPreviewWorkspace.stripe.test.ts`: the same key, copy, theme and telemetry assertions against the binding. ## Review Focus Parity evidence: the existing component tests for the three steps (`SubscriptionAddPaymentPreviewWorkspace.test.ts`, `SubscriptionTransitionPreviewWorkspace*.test.ts`, `SubscriptionSuccessWorkspace.test.ts`) pass unchanged, apart from renaming the stubbed children (`UnifiedStripePaymentSelector` → `CheckoutPaymentForm`, `SingleSelect` → `CheckoutSavedMethodSelect`, `SubscriptionTermsNote` → `CheckoutTermsNote`). ## Stack | PR | branch | base | title | |---|---|---|---| | #19097 | `jaewon/billing-web-remove-plan-grid` | `main` | `FE-3045 refactor(billing-web): drop the billing-web plan grid; hosts own plan selection` | | A #19098 | `jaewon/design-system-button-variants` | #19097 | `FE-3046 refactor(design-system): move buttonVariants into @comfyorg/design-system` | | B #19099 | `jaewon/account-ui-checkout-surfaces` | A | `FE-3046 feat(account-ui): shared checkout confirm, saved-method selector, card form and success` | | **C (this PR)** | **`jaewon/workspace-adopt-account-ui-checkout`** | **B** | **`FE-3046 refactor(workspace): render the subscription checkout steps from account-ui`** | | D #19101 | `jaewon/billing-web-shared-checkout` | C | `FE-3046 feat(billing-web): render the shared embedded checkout` | | E #19104 | `jaewon/billing-web-success-invite` | D | `FE-3046 feat(billing-web): invite your team from the checkout success step` | | F #19105 | `jaewon/checkout-3ds-failure-copy` | E | `FE-3047 fix(account-ui): clear decline toasts once a retry succeeds` | | G #19106 | `jaewon/billing-web-return-close-tab` | F | `FE-3048 fix(billing-web): return to the host by closing the checkout tab` | | H #19115 | `jaewon/checkout-cta-reachable` | G | `FE-3050 fix(account-ui): keep the checkout CTA reachable when notices grow the confirm step` | ```mermaid graph LR P0[#19097: remove plan grid] --> A[A: design-system buttonVariants] A --> B[B: account-ui checkout surfaces] B --> C[C: workspace adopts account-ui] C --> D[D: billing-web shared checkout] D --> E[E: billing-web success invite] E --> F[F: stale decline toasts] F --> G[G: close the billing tab] G --> H[H: reachable checkout CTA] style C fill:#f9d71c,stroke:#333,color:#000 ``` ## AS IS / TO BE Each image: **left = AS IS** (app on #19099, old components), **right = TO BE** (this PR). Rendered in Storybook (a temporary story, not committed) with the app components' real props, the checkout dialog's frame classes, dark theme, 1440x900, and a fake `js.stripe.com` (the same fake billing-web's e2e uses), so the card element area is blank in both. Not against a live backend. AS IS and TO BE are pixel-identical for all four steps (PIL diff: no differing pixels). **Confirm + card entry (new subscription, no saved method)** <img width="1448" alt="card" src="https://github.com/user-attachments/assets/c00c89fb-1eaa-46bd-b335-9bf51e09fff1" /> **Confirm + saved-card selector (two saved cards)** <img width="1448" alt="saved" src="https://github.com/user-attachments/assets/d2db71b9-2e7f-4e99-bd89-45624e771ce5" /> **Plan change (upgrade, no card form)** <img width="1448" alt="change" src="https://github.com/user-attachments/assets/740beba9-fbb0-41ac-a2a6-9bb3815665da" /> **Success** <img width="1448" alt="success" src="https://github.com/user-attachments/assets/42a18550-21fa-41b7-8e7d-c2ebc7349520" /> --------- Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
## Summary Replace PrimeVue form controls with design-system controls for phase 3 of [FE-2728](https://linear.app/comfyorg/issue/FE-2728). ## Changes - Migrate text, password, select, toggle, slider, number, checkbox, and radio controls. - Add reusable checkbox, radio-group, and Reka NumberField components with unit tests and Storybook stories. Number inputs commit on Enter or blur. Some consumers now include increment and decrement buttons. - Use vee-validate Field and Label bindings, InputGroup components, and the MultiSelect value slot. - Remove 44 PrimeVue import allowlist entries relative to the merge base, from 225 to 181. ## Review Focus Keyboard navigation, numeric commits, disabled states, and Vue-node widget parity. ToggleGroupItem defaults to content width. Consumers that need equal-width items pass `flex-1`. Call-site utility overrides, compatibility JavaScript, and remaining Playwright interaction gaps are inventoried in FE-2728 and the [migration tracker](https://linear.app/comfyorg/issue/FE-2727). Generic controls use core semantic theme tokens. The token cleanup in #18632 and WorkflowTabs migration in #18658 are already in the base branch. --------- Co-authored-by: Amp <amp@ampcode.com> Co-authored-by: github-actions <github-actions@github.com>
…yer verdict (#16796) Docs-only follow-up answering the SemVer question raised on #14790. **Verdict: the brand-layer move is breaking, and the first release carrying it must be `2.0.0`.** Key facts: - The change removes brand tokens from `_palette.css` and stops `base.css` from loading Inter, changing the package's public styling surface. - `@comfyorg/design-system@1.1.0` was published from `main` on 2026-09-05 without the brand-layer move. - #14790 remains outside `main`, so its old in-tree `1.1.0` version does not establish a published precedent for the move. - When the brand-layer stack lands, the release carrying that breaking change must be a major bump from the shipped 1.x line. This PR records the policy (exports are part of the public SemVer surface; removing tokens, `@font-face` rules, or exported files is breaking) and the current release precedent in the README's `Releasing` section. It does not change package code or the package version. ## Verification - `npm view @comfyorg/design-system version` reports `1.1.0`. - `main` currently declares `1.1.0` in `packages/design-system/package.json`. - The PR diff is limited to `packages/design-system/README.md`. <!-- authored-by:agent lane-srx-35-1653 --> --------- Co-authored-by: Ryan Di <ryan.weihao.di@gmail.com>
<!-- ccr-slack-attribution --> _Requested by **Jo Zhang, Christian Byrne** · [Slack thread](https://comfy-organization.slack.com/archives/C0BGH348Z0C/p1789707519592879?thread_ts=1789707519.592879&cid=C0BGH348Z0C)_ ## ELI5 A subgraph node shows several "promoted" widgets from the nodes hidden inside it (like width, height, seed). When you rewire the connection behind just one of those widgets, the others could flicker, duplicate, or vanish for a moment, and the wired-in widget itself could disappear. This PR fixes that and adds tests at both levels: unit tests prove the mechanism, and the browser-level repro tests — which this PR's own CI confirmed now pass with the fix — are kept as plain regression tests. - Fixes #14495 ## Motivation This started as a REPRO-ONLY PR for a reported bug: wiring a new node into a subgraph's promoted "prompt" widget makes the subgraph's *other* promoted widgets render duplicated, and a closely related report has the rewired widget itself disappear with redo misplacing the wire. Investigation traced a confirmed mechanism for the disappearing half at the data-model layer (see Summary). Once that mechanism was nailed down precisely enough to fix without guessing, and a second, more contained duplication theory was checked and refuted, it became worth landing the fix and the validating tests in the same PR rather than leaving a fully-diagnosed, small, well-scoped bug unfixed. ## Provenance - **Authored by:** interactive session, Claude Code - **Verified:** `pnpm exec vitest run src/lib/litegraph/src/subgraph/SubgraphWidgetPromotion.test.ts` — 51 passed, 0 failed (was 50 passed / 1 expected fail before the fix; that case now passes for real and was flipped from `it.fails` to `it`). `pnpm exec vitest run src/lib/litegraph/src/subgraph/ src/renderer/extensions/vueNodes/widgets/utils/multilineTextarea.promoted.test.ts src/stores/widgetValueStore.test.ts src/stores/domWidgetStore.test.ts` — 413 passed, 0 failed. `pnpm typecheck`, `pnpm lint`, `pnpm knip` — clean. E2E confirmation came from this PR's own CI (see Summary). Re-verified after the 2026-10-01 rebase onto main: same scoped suites green, typecheck/lint/knip clean. - **Deviations:** none remaining. The previous push deliberately left the e2e `test.fail()` cases unflipped pending CI evidence; that evidence has now arrived (see "E2E confirmation" below) and they are flipped to plain `test()` in this push, exactly per the plan recorded in the earlier Test plan. ## Reviewer context - **Type:** fix + test — a small, contained timing fix plus the tests that prove it, validate/refute an alternate duplication theory, and lock in the browser-level symptom as regression coverage - **Slots into:** a subgraph promoted-widget-promotion bug investigation; no project or epic, ad-hoc follow-up to that investigation - **Not in scope:** `createPromotedMultilineWidget`'s visibility/suppression path — that's PR #18085's territory (a sibling investigation into the same widget family) ## Summary ### The fix (`src/lib/litegraph/src/subgraph/SubgraphNode.ts`) Rewiring the interior link behind a promoted widget (e.g. dragging a new node onto an already-promoted widget's socket) removes the old link before the new one lands. `SubgraphInputNode`'s `'input-disconnected'` handler used to demote the widget **synchronously** — clearing `input.widgetId`/`.widget`/`._widget` immediately — while deferring the matching `widgetValueStore` cleanup to a `queueMicrotask`. For one tick, `hostNode.widgets.length` (synchronous) and the store's tracked id count for that node (deferred) disagreed. The Vue widget grid (`useProcessedWidgets`/`computeProcessedWidgets`) reads its render *order* from the store and its liveness *filter* from `hostNode.widgets` — i.e. from both sides of that exact disagreement. The fix defers the demotion itself by a microtask (mirroring the already-deferred store cleanup), tracked with a per-input cancellation token. The `'input-connected'` listener clears that token on a same-tick reconnect, so a rewire never drops the widget from `hostNode.widgets` at all. A genuine disconnect with no reconnect still fully demotes once the microtask runs — nothing about final-state behavior changes, only the timing of an in-between state that used to be observable. **Deliberate timing-change note** (per this repo's own ecosystem-impact rule for changes touching `node.widgets` access): this changes when a disconnect becomes observable on `hostNode.widgets` — previously synchronous, now one microtask tick later for a genuine disconnect. Two existing tests in `SubgraphWidgetPromotion.test.ts` asserted this synchronously with no `await`; both were updated to `await Promise.resolve()` after the disconnecting call, with a comment explaining why. No public API shape or callback signature changed — only internal timing. ### New tests - `SubgraphWidgetPromotion.test.ts`, "keeps every promoted widget present while rewiring the interior link behind one of them" (previously `it.fails`, now a plain `it`): builds a host with 4 promoted widgets from 4 different interior nodes, rewires the link behind one of them onto a *new* interior node mid-test, and asserts `hostNode.widgets.length` stays at 4 throughout — including immediately after the disconnect step, before the reconnect lands. - `SubgraphWidgetPromotion.test.ts`, "Promoted textarea dual-registration (PM-1328 duplication lead)": tests a second, independently-raised theory for the duplication symptom — that a promoted textarea's DOM-hosted widget (a `DOMWidgetImpl` living in `useDomWidgetStore`, materialized by `createPromotedHostWidget`) and a store-projected row could both end up registered for the same input, which would duplicate a row in the widget grid. Built a minimal `SubgraphNode` subclass wired the same way the app's real subclass is, exercised it across a rewire and a `configure()` reconfigure. **Both pass as-is — the theory is refuted.** `_projectPromotedWidget`'s `if (input._widget) return input._widget` guard is unconditional across every call site and there's no observable window where `input.widgetId` is set while `input._widget` isn't. Kept as regression coverage. ### E2E confirmation (CI run [35419314544](https://github.com/Comfy-Org/ComfyUI_frontend/actions/runs/35419314544)) CI on the previous head answered the open question the `test.fail()` repros were posted to ask — the fix **does** resolve the browser-level symptom end-to-end: - Both `subgraphPromotedWidgetRewireDuplication.spec.ts` repro scenarios (interior rewire and external boundary wire) were reported "Expected to fail, but passed" on every run and retry (video job + shard 12, 4/4 retries). Flipped to plain `test()` regression tests, and retitled to describe the behavior they now guarantee. - `subgraphPromotion.spec.ts` "Promoted STRING widget edit survives a rebind of the interior link" — a *pre-existing* `test.fail()` for #14495 — also passed on all 4 shard-12 attempts. Flipped to plain `test()`; hence "Fixes #14495" above. - The external-boundary undo test went through two CI-informed revisions after the rebase: which snapshot undo targets depends on change-tracker capture timing the test cannot control (captures fire on window mouseup, before the click handlers that apply promotions, so the entry pushed by the slot drop — #18117 — can lag the built baseline by one action), and an attempt to pin the snapshot with an extra canvas click timed out in CI. The test now asserts the invariant both reported symptoms violate instead of a fixed widget set: whatever state undo restores, the host's model holds no duplicate widgets and the grid renders exactly one row per model widget, none otherwise. ## Test plan - [x] Watch this PR's `playwright-video-new-tests` CI check: the `test.fail()` cases were reported as unexpectedly passed (run 35419314544) - [x] Flip those cases (and the pre-existing #14495 one) to plain `test()` - [x] Make the external-boundary undo test deterministic (it now asserts render/model consistency after undo instead of assuming which snapshot undo targets) - [ ] CI green on the rebased head (the "Check for AI agent co-author trailers" check was removed from main in #19393; the first commit still carries trailers, and rewording it remains a maintainer call since it rewrites shared history) 🤖 Generated with [Claude Code](https://claude.com/claude-code) https://claude.ai/code/session_01U6aDANuoAYDtxd9WiCk44Q --------- Co-authored-by: Claude <noreply@anthropic.com>
…mfy API page (#19699) <!-- ccr-slack-attribution --> _Requested by **Robert Paige** · [Slack thread](https://comfy-organization.slack.com/archives/C0B61J7QCJW/p1790829458470609)_ ## Summary Two changes to `/platform/comfy-api` (en + zh-CN): adds a dark CTA banner above "Built for teams", and makes the "Ship in minutes" deploy terminal static and color-coded instead of typewriter-animated. ## Changes ### Skip-the-setup banner - **What**: New `ServerlessSkipSetupBanner.vue`, a dark rounded bar (`bg-primary-comfy-ink-light`, same pattern as `RouterMigrationBanner.vue`) with a bold heading ("Or skip the setup."), muted subtext ("Paste this into your coding agent."), and a pill-shaped yellow-outline "COPY AGENT PROMPT" button with a copy icon that flips to a "COPIED" checkmark state on click (same accessible grid-swap pattern as `RouterMigrationBanner.vue`'s copy button). - **What**: The button copies `deployPromptFor(locale)` from the existing `config/deploy-prompt.ts` rather than duplicating the prompt text, so it can never drift from the prompt the deploy terminal already copies. - **What**: Wired the new banner into `src/pages/platform/comfy-api.astro` and its `zh-CN` twin, directly above `<ServerlessScaleSection>` ("Built for teams"). - New translation keys under `platform.serverlessSkipSetup.*` in `src/i18n/translations.ts`, with `zh-CN` values for every new string. ### Static, color-coded deploy terminal - **What**: Stops the typewriter/typing animation on the deploy terminal: it now renders its transcript in full immediately, with no keystroke-by-keystroke reveal and no blinking cursor. - **What**: Adds an opt-in `typewriter` prop to `LiveTerminal.vue` (default `true`, unchanged everywhere else it's used). - **What**: Colors the transcript with three explicit colors instead of one: the `$` prompt in the brand yellow, its command text in the terminal's default off-white, and a whole `✓`/`✔` success line in green. (First pass used Shiki's generic `shell` grammar, but that colored nearly every word in this free-form text the same muted tone — technically "colored" but visually indistinguishable from plain text. This explicit 3-color scheme reads as clearly syntax-highlighted.) - **What**: Updates the terminal's displayed transcript in `ServerlessDeploySection.vue` to the three commands a real run produces: ``` $ comfy build init ✓ Scanned this ComfyUI install - custom nodes, models, pinned deps $ comfy build push --release ``` The "Copy prompt" button beside the terminal is unchanged and still copies the full agent prompt from `config/deploy-prompt.ts` — the visible transcript and the clipboard payload intentionally differ, matching how this section already worked. ## Review Focus - I reused the existing `RouterMigrationBanner.vue` dark-bar + copy-button pattern for the new banner instead of the sitewide dismissible `AnnouncementBanner.vue`, since this is a page-section element (not a sitewide dismissible announcement) and the visual design (dark rounded bar, pill outline button, copy-to-clipboard) matches `RouterMigrationBanner.vue` much more closely. - `LiveTerminal`'s `typewriter` prop defaults to `true`, so every other page using it keeps its existing typing animation; only this section opts out. ## Verification - `pnpm exec eslint` on touched files: clean. - `pnpm exec oxfmt --check` / website `prettier --check` on touched files: clean. - `pnpm typecheck:website`: 0 errors. - `pnpm exec vitest run` on all five touched test files: passing, including tests asserting the terminal's three distinct colors and the absence of the typewriter cursor. - Built the site with `pnpm build` and inspected the static output HTML directly: the `$` renders with `color:var(--color-primary-comfy-yellow)`, the command text has no color override (inherits the terminal's off-white), and the full `✓ ...` line renders with `color:#A7C080` (green) — confirmed this is real, not just present-but-inert markup. - Took a local screenshot (via a system Chromium binary, since this sandbox can't reach `cdn.playwright.dev` to install Playwright's browser) of the rendered terminal and visually confirmed three clearly distinct colors. - Verified against a local `astro dev` build that both changes render correctly and in the right order (deploy terminal → skip-setup banner → "Built for teams") on both `/platform/comfy-api` and `/zh-CN/platform/comfy-api`, and that both copy buttons still work. --------- Co-authored-by: Claude <noreply@anthropic.com>
## Summary Removes the `asset_rename_enabled` feature flag, which has gated nothing since January 2026. Linear: [FE-3095](https://linear.app/comfyorg/issue/FE-3095/removearchive-asset-rename-enabled-frontend) (part of the [Feature Flag Hygiene](https://linear.app/comfyorg/project/feature-flag-hygiene-a0c6d39b60a6) project) ## Changes - **What**: Deletes the `ASSET_RENAME_ENABLED` enum member, the `assetRenameEnabled` getter, the telemetry map entry, the `RemoteConfig.asset_rename_enabled` field, and the two hand-maintained mock enums (Storybook + vitest `__mocks__`). Also drops the orphaned `assetBrowser.rename.failed` i18n key from all 15 locales. - **Breaking**: None. The flag had zero consumers, so no rendered behavior changes. ## Review Focus **Why this is a no-op, not a feature removal.** The rename UI this flag was built for was deleted from `AssetCard` in January 2026 (`f32cdc5ad4`, `a068d47eb6`). Asset rename then re-shipped **ungated** through `ModelInfoPanel`'s `EditableText` → `queueMetadataUpdate({ name })` → `PUT /api/assets/{id}`. So users can rename assets today even though the flag has sat at 0% rollout and `STALE` in PostHog. `rg assetRenameEnabled` found no product consumer — only the flag's own plumbing. Because the feature is already live and ungated, deleting the plumbing was preferred over hard-coding the flag on, so we don't leave a permanent 0% gate in the registry. If asset rename ever needs gating again, the flag can be reintroduced. **One test was re-pointed, not deleted.** `useFeatureFlags.test.ts`'s "resolveFlag falls through to server when no override is set" used this flag only as an arbitrary example; its real subject is the override precedence chain. It now exercises `asset_deletion_enabled` so that coverage survives. **The i18n key was already dead.** `assetBrowser.rename.failed` lost its only caller in the January removal and is referenced nowhere in `src/` or `browser_tests/`. **Backend follow-up is deliberately out of scope.** Seven references remain in `Comfy-Org/cloud` (`common/featuregates/flags.go`: the flag const, the PostHog const, the evaluator, and `FrontendFlags` membership, plus two prose comments that use this key as their worked example). Per the project's FE-first ordering those land after this deploys — removing `FrontendFlags` membership is what actually stops the PostHog evaluations. The PostHog archive comes last and is not done here. ## How has this been tested? - `pnpm typecheck` — pass - `pnpm lint` — pass (0 errors) - `pnpm test:unit` — 24795 passed. 13 failures in `check-binary-size.test.ts` and `UnifiedPricingTable.test.ts` are pre-existing and reproduce identically on unmodified `origin/main` (local node v25.9 vs the repo's required >=26.8.2) - `pnpm locale:check` — pass - `tsx scripts/check-unused-i18n-keys.ts` — pass
…t drops (PM-1658) (#18801) ## Summary A run-approval consent card could be left permanently unanswerable: it stayed on screen with both actions looking enabled, and clicking either sent nothing at all. `answerAsk` gated on `activeTurnId`, which is null for any ask whose turn has detached. Fixes PM-1658. Test plan tracked in PM-1659. ## Changes - **What**: `answerAsk` no longer gates on `activeTurnId`. The answer endpoint is keyed by thread and ask alone — it never looks a turn up — and the server parks a run approval for days, so `threadId` + `askId` is all it needs. The gate is reached by two live triggers: a turn **completing** (`agent_message_done` → `settleActiveTurn` → `clearActive`), and a **newer turn** taking over. #18834 pins the completion trigger independently and is red on `main`. - Dismissing a card is one store operation (`retireAsk`) that reaches every holder of the message — active transport, settled transports, the stashed turn, and the published copies — because editing only one let another republish it. `agent_ask_resolved` routes through it too, since `ingest` can only reach a turn that still exists. - Retirement binds to the thread the answer was **given** on, not whichever thread is on screen when it settles. Without this, an answer completing after a chat switch edited the wrong thread and left the answering thread's stashed card to be handed back enabled by `resumeBackgroundTurn`. - An accepted answer is retired rather than re-offered, and a retired ask is remembered so a refetch can't rebuild the card from the server's still-`pending` row. Only the card is suppressed — never the turn, which answering is what lets resume. - A failed answer is re-driven once with the same selection before the card is retired with an explicit "could not confirm" notice. A rejection arriving after the ask already resolved stays silent. - **Observability**: `agent_ask_answer_unroutable`, `agent_ask_answer_refused`, `agent_ask_answer_failed`, `agent_ask_answer_superseded`, all carrying `surface: 'agent'`. None of this failure mode was visible before. ## Review Focus - **Main's reconnect recovery is preserved, not replaced.** PM-1199 (#18152) means a socket drop no longer detaches a turn — it stays live and reconciles from REST — so the socket drop is no longer a trigger for this gate. This branch takes main's `onStatus`/`reconcileLiveTurns` wholesale and **removes** the `dismissCommittedAsks`/`committedAskIds` machinery it previously carried: that existed to rescue cards from the old abort-on-drop path, and against main's recovery it would retire a card whose resolution frame is still coming. - **Why a failed answer retires the card** rather than re-offering it: `asks.go` CASes the answer onto the row *before* it wakes the turn and reports 500 for the wake alone, so a failure may already have authorized the run — and the server answers any later click by replaying the first selection. On a spend authorization, re-offering is the wrong way to be wrong. ### Known gaps, both needing a backend change first - The two 409s — an ordinary double-click, and an approval reconciled to `cancelled` because its turn was gone — are distinguishable only by server prose. The second is still silent; it wants a stable `code` on the error body. - A contradictory answer from another tab is caught only when the resolution frame arrives. Catching it while disconnected needs the 202 body to carry the committed selection. ## Verification - Agent suite **2482 passed**, 1 expected-fail. The one failure (`agentSubgraphDefinitions › scans deeply nested definition ids`, a `RangeError` in `lib0`'s `deepFreeze`) is in a file byte-identical to `main`. - `typecheck`, `lint`, `format`, `knip` clean. - No browser spec of its own: the socket-drop spec this PR used to carry was deleted because main's recovery removed its premise — it would now pass with or without this fix. E2E coverage for the surviving trigger is #18834. --------- Co-authored-by: Glary-Bot <glary-bot@users.noreply.github.com>
Preserve the replacement graph-intent architecture and avoid reintroducing the deleted mint wiring. This also removes the prototype serialize accessor implicated by review feedback: #19504 (comment) #19504 (comment) #19504 (comment) #19504 (comment)
Addresses review feedback: #19504 (comment)
Addresses review feedback: #19504 (comment)
Addresses review feedback: #19504 (comment)
|
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 ⚙️ Run configurationConfiguration used: Repository: Comfy-Org/ComfyUI_frontend/.coderabbit.yaml Review profile: ASSERTIVE Plan: Team Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
Comment |
|
Review the following changes in direct dependencies. Learn more about Socket for GitHub.
|
🎭 Playwright: ✅ 2185 passed, 0 failed📊 Browser Reports
🌐 Website E2ETip All tests passed.
🎨 Storybook: ✅ Built — View Storybook🌾 Fallow: ❌ FailedFallow audits the diff for duplication, complexity and dead code. It gates on new findings only, so when the audit completes, a red is something this PR introduced or moved — never inherited backlog. A red can also mean the audit itself did not finish; the details below say which happened. Fallow gates on new findings in changed files only, so everything below was introduced or moved by this PR (2193 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 5ee77e67885a09a4f6ab5356bca7f3494e978a6c
Full docs: ADR-DEVEX-LINT-0015 🔗 Website PreviewWebsite Preview: https://comfy-website-preview-pr-19737.vercel.app Models is built and gated by PostHog (workshop-enabled), using staging Cloud. This commit: https://website-frontend-4m8zmnjpc-comfyui.vercel.app Last updated: 2026-10-01T09:11:55Z for 🧩 Custom Nodes Ecosystem Matrix: ✅ PassedExecutes the frontend JS of every registry pack that ships any, against this change, in both renderers. The eight
On a pin-bump PR, a red gate means the ecosystem moved — not that your diff broke something. That is what the bump PR is for. Outside a pin bump or a cold rebuild of the unpinned tail, a red means the diff. Note the corpus size and the executed count are different numbers, and the verdict is against the executed count: roughly 3,100 of ~5,100 packs ship no extension-shaped JS and are 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 📦 Bundle Size
⚡ Performance Report
Absolute values
Raw data{
"timestamp": "2026-10-01T09:22:42.846Z",
"gitSha": "94ad5bf5e25cd3480cec60a88c88836f2d01344c",
"branch": "christian-byrne/19026-post-merge-review-fixes",
"measurements": [
{
"name": "canvas-idle",
"durationMs": 2021.682999999996,
"styleRecalcs": 10,
"styleRecalcDurationMs": 4.138000000000001,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 415.42199999999997,
"heapDeltaBytes": 373604,
"heapUsedBytes": 66942580,
"domNodes": 0,
"jsHeapTotalBytes": 262144,
"scriptDurationMs": 8.361000000000008,
"eventListeners": 0,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666115702479342,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "canvas-idle",
"durationMs": 2025.2420000000484,
"styleRecalcs": 9,
"styleRecalcDurationMs": 3.8960000000000004,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 506.401,
"heapDeltaBytes": 339492,
"heapUsedBytes": 66695752,
"domNodes": 0,
"jsHeapTotalBytes": 0,
"scriptDurationMs": 14.395999999999999,
"eventListeners": 0,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.665289256198346,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "canvas-mouse-sweep",
"durationMs": 1762.5170000000026,
"styleRecalcs": 73,
"styleRecalcDurationMs": 17.931,
"layouts": 12,
"layoutDurationMs": 4.658,
"taskDurationMs": 805.331,
"heapDeltaBytes": -2994284,
"heapUsedBytes": 63560168,
"domNodes": 0,
"jsHeapTotalBytes": 262144,
"scriptDurationMs": 168.29099999999997,
"eventListeners": 0,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66509433962264,
"p95FrameDurationMs": 16.800000000000182
},
{
"name": "canvas-mouse-sweep",
"durationMs": 1806.2400000000025,
"styleRecalcs": 75,
"styleRecalcDurationMs": 17.762,
"layouts": 12,
"layoutDurationMs": 4.475,
"taskDurationMs": 838.638,
"heapDeltaBytes": -3231816,
"heapUsedBytes": 61652620,
"domNodes": 0,
"jsHeapTotalBytes": 0,
"scriptDurationMs": 171.53300000000002,
"eventListeners": 0,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666055045871563,
"p95FrameDurationMs": 16.800000000000182
},
{
"name": "canvas-zoom-sweep",
"durationMs": 1721.9879999999534,
"styleRecalcs": 32,
"styleRecalcDurationMs": 10.829,
"layouts": 6,
"layoutDurationMs": 1.4240000000000002,
"taskDurationMs": 342.24499999999995,
"heapDeltaBytes": 3508832,
"heapUsedBytes": 69873340,
"domNodes": 53,
"jsHeapTotalBytes": 524288,
"scriptDurationMs": 11.901000000000009,
"eventListeners": 15,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66504854368932,
"p95FrameDurationMs": 16.799999999999272
},
{
"name": "canvas-zoom-sweep",
"durationMs": 1742.4929999999677,
"styleRecalcs": 31,
"styleRecalcDurationMs": 11.350000000000001,
"layouts": 6,
"layoutDurationMs": 1.3169999999999997,
"taskDurationMs": 362.79599999999994,
"heapDeltaBytes": 3455272,
"heapUsedBytes": 70039040,
"domNodes": 53,
"jsHeapTotalBytes": 0,
"scriptDurationMs": 13.564000000000007,
"eventListeners": 15,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.664761904761907,
"p95FrameDurationMs": 16.800000000000182
},
{
"name": "dom-widget-clipping",
"durationMs": 727.6239999999916,
"styleRecalcs": 12,
"styleRecalcDurationMs": 3.8339999999999987,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 494.697,
"heapDeltaBytes": -7647092,
"heapUsedBytes": 59258596,
"domNodes": 0,
"jsHeapTotalBytes": 262144,
"scriptDurationMs": 182.82500000000002,
"eventListeners": 0,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.665909090909096,
"p95FrameDurationMs": 16.699999999999818
},
{
"name": "dom-widget-clipping",
"durationMs": 705.2200000000539,
"styleRecalcs": 11,
"styleRecalcDurationMs": 3.6949999999999985,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 473.80400000000003,
"heapDeltaBytes": -8741880,
"heapUsedBytes": 56156020,
"domNodes": 0,
"jsHeapTotalBytes": 524288,
"scriptDurationMs": 169.778,
"eventListeners": 0,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66428571428573,
"p95FrameDurationMs": 16.800000000000182
},
{
"name": "dom-widget-order-dense-n200-w200-i120",
"durationMs": 1145.7269999999653,
"styleRecalcs": 0,
"styleRecalcDurationMs": 0,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 802.316,
"heapDeltaBytes": 11083472,
"heapUsedBytes": 90721032,
"domNodes": 0,
"jsHeapTotalBytes": 4620288,
"scriptDurationMs": 67.95799999999998,
"eventListeners": 0,
"totalBlockingTimeMs": 516,
"frameDurationMs": 33.8235294117647,
"p95FrameDurationMs": 16.800000000000182
},
{
"name": "dom-widget-order-dense-n200-w200-i120",
"durationMs": 1157.839000000081,
"styleRecalcs": 0,
"styleRecalcDurationMs": 0,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 803.12,
"heapDeltaBytes": 13057448,
"heapUsedBytes": 93242380,
"domNodes": 0,
"jsHeapTotalBytes": 6979584,
"scriptDurationMs": 72.08800000000001,
"eventListeners": 0,
"totalBlockingTimeMs": 521,
"frameDurationMs": 33.82058823529413,
"p95FrameDurationMs": 16.800000000000182
},
{
"name": "dom-widget-order-sparse-n1000-w20-i120",
"durationMs": 879.128000000037,
"styleRecalcs": 0,
"styleRecalcDurationMs": 0,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 483.26300000000043,
"heapDeltaBytes": 18762916,
"heapUsedBytes": 104523836,
"domNodes": 0,
"jsHeapTotalBytes": 266240,
"scriptDurationMs": 165.64799999999997,
"eventListeners": 0,
"totalBlockingTimeMs": 222,
"frameDurationMs": 24.53611111111114,
"p95FrameDurationMs": 16.799999999999272
},
{
"name": "dom-widget-order-sparse-n1000-w20-i120",
"durationMs": 870.4179999999724,
"styleRecalcs": 0,
"styleRecalcDurationMs": 0,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 470.6420000000002,
"heapDeltaBytes": 17686888,
"heapUsedBytes": 104215256,
"domNodes": 0,
"jsHeapTotalBytes": 1576960,
"scriptDurationMs": 162.958,
"eventListeners": 0,
"totalBlockingTimeMs": 226,
"frameDurationMs": 25.488235294117658,
"p95FrameDurationMs": 33.29999999999927
},
{
"name": "extension-clean-combined-draw",
"durationMs": 10658.443000000005,
"styleRecalcs": 0,
"styleRecalcDurationMs": 0,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 9670.360999999999,
"heapDeltaBytes": 9803072,
"heapUsedBytes": 86160716,
"domNodes": -142,
"jsHeapTotalBytes": -5902336,
"scriptDurationMs": 7921.310000000001,
"eventListeners": -74,
"totalBlockingTimeMs": 1228,
"frameDurationMs": 32.567584097859324,
"p95FrameDurationMs": 66.69999999999891
},
{
"name": "extension-clean-combined-draw",
"durationMs": 10139.74900000005,
"styleRecalcs": 1,
"styleRecalcDurationMs": 0.42599999999999927,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 9224.687,
"heapDeltaBytes": 4742888,
"heapUsedBytes": 80255568,
"domNodes": -142,
"jsHeapTotalBytes": -7999488,
"scriptDurationMs": 7539.337,
"eventListeners": -74,
"totalBlockingTimeMs": 891,
"frameDurationMs": 31.566666666666666,
"p95FrameDurationMs": 50.100000000000364
},
{
"name": "extension-rgthree-prototype-draw",
"durationMs": 10136.981999999989,
"styleRecalcs": 0,
"styleRecalcDurationMs": 0,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 9263.821,
"heapDeltaBytes": 11355372,
"heapUsedBytes": 92693388,
"domNodes": 0,
"jsHeapTotalBytes": -14680064,
"scriptDurationMs": 7557.519,
"eventListeners": 0,
"totalBlockingTimeMs": 928,
"frameDurationMs": 32.168253968253964,
"p95FrameDurationMs": 66.59999999999854
},
{
"name": "extension-rgthree-prototype-draw",
"durationMs": 10152.230000000032,
"styleRecalcs": 1,
"styleRecalcDurationMs": 0.057999999999999025,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 9232.662,
"heapDeltaBytes": -192664,
"heapUsedBytes": 68472668,
"domNodes": -142,
"jsHeapTotalBytes": -7213056,
"scriptDurationMs": 7527.633,
"eventListeners": -74,
"totalBlockingTimeMs": 883,
"frameDurationMs": 31.520186335403725,
"p95FrameDurationMs": 50.100000000000364
},
{
"name": "large-graph-idle",
"durationMs": 2007.5400000000059,
"styleRecalcs": 9,
"styleRecalcDurationMs": 4.417000000000001,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 560.943,
"heapDeltaBytes": -1645616,
"heapUsedBytes": 84614184,
"domNodes": 0,
"jsHeapTotalBytes": -225280,
"scriptDurationMs": 38.543000000000006,
"eventListeners": 0,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666115702479335,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "large-graph-idle",
"durationMs": 2027.5510000000168,
"styleRecalcs": 7,
"styleRecalcDurationMs": 2.546,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 553.318,
"heapDeltaBytes": 20737036,
"heapUsedBytes": 82565616,
"domNodes": -328,
"jsHeapTotalBytes": 0,
"scriptDurationMs": 30.507999999999992,
"eventListeners": -164,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.665289256198346,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "large-graph-pan",
"durationMs": 2505.354000000011,
"styleRecalcs": 69,
"styleRecalcDurationMs": 11.039,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 1652.0040000000004,
"heapDeltaBytes": 2864884,
"heapUsedBytes": 89283884,
"domNodes": 0,
"jsHeapTotalBytes": 786432,
"scriptDurationMs": 853.961,
"eventListeners": 0,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.778523489932887,
"p95FrameDurationMs": 16.800000000000182
},
{
"name": "large-graph-pan",
"durationMs": 2587.466999999947,
"styleRecalcs": 67,
"styleRecalcDurationMs": 10.421,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 1678.4630000000002,
"heapDeltaBytes": -38416424,
"heapUsedBytes": 85409576,
"domNodes": -348,
"jsHeapTotalBytes": -3018752,
"scriptDurationMs": 803.544,
"eventListeners": -164,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666013071895428,
"p95FrameDurationMs": 16.800000000000182
},
{
"name": "large-graph-zoom",
"durationMs": 3691.032000000007,
"styleRecalcs": 65,
"styleRecalcDurationMs": 12.757000000000001,
"layouts": 60,
"layoutDurationMs": 14.282,
"taskDurationMs": 1761.9579999999999,
"heapDeltaBytes": -11802384,
"heapUsedBytes": 76973528,
"domNodes": 0,
"jsHeapTotalBytes": 823296,
"scriptDurationMs": 811.222,
"eventListeners": 0,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.665765765765766,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "large-graph-zoom",
"durationMs": 3738.9530000000377,
"styleRecalcs": 62,
"styleRecalcDurationMs": 11.174,
"layouts": 60,
"layoutDurationMs": 14.767999999999999,
"taskDurationMs": 1761.784,
"heapDeltaBytes": 8278948,
"heapUsedBytes": 72668460,
"domNodes": -167,
"jsHeapTotalBytes": 0,
"scriptDurationMs": 792.9459999999999,
"eventListeners": 0,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666222222222224,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "legacy-node-drag",
"durationMs": 3103.31199999996,
"styleRecalcs": 46,
"styleRecalcDurationMs": 7.185999999999998,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 1934.315,
"heapDeltaBytes": 26309360,
"heapUsedBytes": 93804580,
"domNodes": 0,
"jsHeapTotalBytes": 1728512,
"scriptDurationMs": 966.9499999999999,
"eventListeners": 180,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.755675675675683,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "legacy-node-drag",
"durationMs": 3142.667999999958,
"styleRecalcs": 45,
"styleRecalcDurationMs": 7.296,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 1974.597,
"heapDeltaBytes": 26939812,
"heapUsedBytes": 93346488,
"domNodes": 0,
"jsHeapTotalBytes": 942080,
"scriptDurationMs": 1005.0760000000001,
"eventListeners": 180,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.845161290322583,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "minimap-idle",
"durationMs": 2013.703000000021,
"styleRecalcs": 7,
"styleRecalcDurationMs": 3.106000000000001,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 531.655,
"heapDeltaBytes": 20735868,
"heapUsedBytes": 83552224,
"domNodes": -328,
"jsHeapTotalBytes": 0,
"scriptDurationMs": 31.89900000000001,
"eventListeners": -164,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666115702479342,
"p95FrameDurationMs": 16.799999999999272
},
{
"name": "minimap-idle",
"durationMs": 2052.695999999969,
"styleRecalcs": 5,
"styleRecalcDurationMs": 1.981,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 583.1189999999999,
"heapDeltaBytes": -41535096,
"heapUsedBytes": 83170632,
"domNodes": -328,
"jsHeapTotalBytes": -1409024,
"scriptDurationMs": 31.244000000000007,
"eventListeners": -164,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66585365853659,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "subgraph-dom-widget-clipping",
"durationMs": 580.4929999999899,
"styleRecalcs": 47,
"styleRecalcDurationMs": 6.203,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 393.125,
"heapDeltaBytes": -12095016,
"heapUsedBytes": 57120544,
"domNodes": 0,
"jsHeapTotalBytes": 262144,
"scriptDurationMs": 141.407,
"eventListeners": 6,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66571428571429,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "subgraph-dom-widget-clipping",
"durationMs": 620.0330000000349,
"styleRecalcs": 46,
"styleRecalcDurationMs": 5.342000000000001,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 406.88900000000007,
"heapDeltaBytes": -11830968,
"heapUsedBytes": 57163976,
"domNodes": 0,
"jsHeapTotalBytes": 0,
"scriptDurationMs": 144.204,
"eventListeners": 6,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.664864864864875,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "subgraph-idle",
"durationMs": 2014.3959999999765,
"styleRecalcs": 9,
"styleRecalcDurationMs": 3.854999999999999,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 421.071,
"heapDeltaBytes": 825780,
"heapUsedBytes": 70100352,
"domNodes": 0,
"jsHeapTotalBytes": 262144,
"scriptDurationMs": 8.161000000000001,
"eventListeners": 0,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.665289256198346,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "subgraph-idle",
"durationMs": 2037.4259999999822,
"styleRecalcs": 10,
"styleRecalcDurationMs": 4.129999999999998,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 342.87200000000007,
"heapDeltaBytes": -28271564,
"heapUsedBytes": 75921536,
"domNodes": -269,
"jsHeapTotalBytes": -2097152,
"scriptDurationMs": 9.245999999999997,
"eventListeners": -206,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666115702479342,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "subgraph-mouse-sweep",
"durationMs": 1704.9739999999929,
"styleRecalcs": 78,
"styleRecalcDurationMs": 17.346,
"layouts": 16,
"layoutDurationMs": 5.308,
"taskDurationMs": 722.5939999999999,
"heapDeltaBytes": 15050136,
"heapUsedBytes": 68915416,
"domNodes": 0,
"jsHeapTotalBytes": 0,
"scriptDurationMs": 110.06900000000002,
"eventListeners": 0,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66568627450981,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "subgraph-mouse-sweep",
"durationMs": 1724.0040000000363,
"styleRecalcs": 77,
"styleRecalcDurationMs": 20.691000000000003,
"layouts": 16,
"layoutDurationMs": 6.419000000000001,
"taskDurationMs": 813.191,
"heapDeltaBytes": -6382460,
"heapUsedBytes": 62269384,
"domNodes": 0,
"jsHeapTotalBytes": 262144,
"scriptDurationMs": 128.287,
"eventListeners": 0,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66601941747573,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "subgraph-transition-enter",
"durationMs": 1419.1800000000399,
"styleRecalcs": 17,
"styleRecalcDurationMs": 30.705999999999996,
"layouts": 5,
"layoutDurationMs": 14.046,
"taskDurationMs": 988.5509999999999,
"heapDeltaBytes": 23955840,
"heapUsedBytes": 109328996,
"domNodes": 13747,
"jsHeapTotalBytes": 3670016,
"scriptDurationMs": 31.645999999999994,
"eventListeners": 2383,
"totalBlockingTimeMs": 143,
"frameDurationMs": 19.632876712328752,
"p95FrameDurationMs": 16.80000000000109
},
{
"name": "viewport-pan-sweep",
"durationMs": 8570.427999999993,
"styleRecalcs": 250,
"styleRecalcDurationMs": 34.184,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 5840.242,
"heapDeltaBytes": -47260560,
"heapUsedBytes": 75076308,
"domNodes": -328,
"jsHeapTotalBytes": -1740800,
"scriptDurationMs": 2691.006,
"eventListeners": -164,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.698440545808968,
"p95FrameDurationMs": 16.799999999999272
},
{
"name": "viewport-pan-sweep",
"durationMs": 9119.978000000061,
"styleRecalcs": 249,
"styleRecalcDurationMs": 34.49099999999999,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 6097.952,
"heapDeltaBytes": -3842368,
"heapUsedBytes": 79002512,
"domNodes": -142,
"jsHeapTotalBytes": -10915840,
"scriptDurationMs": 2819.321,
"eventListeners": -78,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.696703296703298,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "vue-large-graph-idle",
"durationMs": 16983.93299999998,
"styleRecalcs": 0,
"styleRecalcDurationMs": 0,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 16336.695999999996,
"heapDeltaBytes": -40470788,
"heapUsedBytes": 167939420,
"domNodes": -3045,
"jsHeapTotalBytes": 12079104,
"scriptDurationMs": 254.565,
"eventListeners": -14915,
"totalBlockingTimeMs": 0,
"frameDurationMs": 28.658614864864866,
"p95FrameDurationMs": 50.10000000000218
},
{
"name": "vue-large-graph-idle",
"durationMs": 17106.532000000014,
"styleRecalcs": 0,
"styleRecalcDurationMs": 0,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 16543.02,
"heapDeltaBytes": -8926736,
"heapUsedBytes": 199712948,
"domNodes": -3045,
"jsHeapTotalBytes": 13127680,
"scriptDurationMs": 269.79,
"eventListeners": -14915,
"totalBlockingTimeMs": 0,
"frameDurationMs": 29.03106960950764,
"p95FrameDurationMs": 66.59999999999854
},
{
"name": "vue-large-graph-pan",
"durationMs": 20849.901999999987,
"styleRecalcs": 169,
"styleRecalcDurationMs": 18.56000000000002,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 20485.559,
"heapDeltaBytes": -39184928,
"heapUsedBytes": 168978816,
"domNodes": -3045,
"jsHeapTotalBytes": 6311936,
"scriptDurationMs": 855.2070000000001,
"eventListeners": -14915,
"totalBlockingTimeMs": 22,
"frameDurationMs": 37.83865698729583,
"p95FrameDurationMs": 83.29999999999927
},
{
"name": "vue-large-graph-pan",
"durationMs": 21512.191000000028,
"styleRecalcs": 174,
"styleRecalcDurationMs": 21.93299999999998,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 21121.081000000002,
"heapDeltaBytes": -7949820,
"heapUsedBytes": 193409720,
"domNodes": -3045,
"jsHeapTotalBytes": 8863744,
"scriptDurationMs": 897.4409999999999,
"eventListeners": -14915,
"totalBlockingTimeMs": 15,
"frameDurationMs": 41.186015325670496,
"p95FrameDurationMs": 83.39999999999964
},
{
"name": "vue-node-resize-workload",
"durationMs": 6283.636999999999,
"styleRecalcs": 20,
"styleRecalcDurationMs": 80.68199999999997,
"layouts": 20,
"layoutDurationMs": 23.265999999999995,
"taskDurationMs": 6105.322999999999,
"heapDeltaBytes": 1096056,
"heapUsedBytes": 198977972,
"domNodes": 0,
"jsHeapTotalBytes": 524288,
"scriptDurationMs": 110.92599999999997,
"eventListeners": 0,
"totalBlockingTimeMs": 0,
"frameDurationMs": 36.52906976744186,
"p95FrameDurationMs": 66.70000000000073
},
{
"name": "vue-node-resize-workload",
"durationMs": 6330.328000000009,
"styleRecalcs": 20,
"styleRecalcDurationMs": 84.27799999999999,
"layouts": 20,
"layoutDurationMs": 24.816000000000006,
"taskDurationMs": 6221.240000000001,
"heapDeltaBytes": 1359464,
"heapUsedBytes": 200839688,
"domNodes": 0,
"jsHeapTotalBytes": 786432,
"scriptDurationMs": 113.22899999999998,
"eventListeners": 0,
"totalBlockingTimeMs": 0,
"frameDurationMs": 37.6970238095238,
"p95FrameDurationMs": 66.70000000000073
},
{
"name": "vue-zoom-out-idle",
"durationMs": 8881.008999999949,
"styleRecalcs": 0,
"styleRecalcDurationMs": 0,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 8568.213,
"heapDeltaBytes": -18056332,
"heapUsedBytes": 183806664,
"domNodes": -3045,
"jsHeapTotalBytes": 5525504,
"scriptDurationMs": 147.57299999999995,
"eventListeners": -14915,
"totalBlockingTimeMs": 0,
"frameDurationMs": 30.468384879725097,
"p95FrameDurationMs": 66.60000000000218
},
{
"name": "vue-zoom-out-idle",
"durationMs": 8762.565999999992,
"styleRecalcs": 0,
"styleRecalcDurationMs": 0,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 8523.600999999999,
"heapDeltaBytes": -14108612,
"heapUsedBytes": 199765876,
"domNodes": -3045,
"jsHeapTotalBytes": 9961472,
"scriptDurationMs": 137.16299999999998,
"eventListeners": -14915,
"totalBlockingTimeMs": 0,
"frameDurationMs": 30.486759581881536,
"p95FrameDurationMs": 66.70000000000073
},
{
"name": "workflow-execution",
"durationMs": 138.1619999999657,
"styleRecalcs": 9,
"styleRecalcDurationMs": 12.401,
"layouts": 4,
"layoutDurationMs": 1.625,
"taskDurationMs": 100.938,
"heapDeltaBytes": 3924204,
"heapUsedBytes": 57155988,
"domNodes": -114,
"jsHeapTotalBytes": 0,
"scriptDurationMs": 8.202999999999998,
"eventListeners": 17,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66249999999991,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "workflow-execution",
"durationMs": 104.13899999991827,
"styleRecalcs": 7,
"styleRecalcDurationMs": 8.543000000000001,
"layouts": 3,
"layoutDurationMs": 1.1940000000000002,
"taskDurationMs": 75.35499999999998,
"heapDeltaBytes": 3644624,
"heapUsedBytes": 67711396,
"domNodes": 113,
"jsHeapTotalBytes": 0,
"scriptDurationMs": 5.170000000000001,
"eventListeners": 17,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.671428571428546,
"p95FrameDurationMs": 16.800000000000182
},
{
"name": "agent-composer-empty-conversation-typing",
"durationMs": 972.5449999999682,
"styleRecalcs": 78,
"styleRecalcDurationMs": 107.66000000000001,
"layouts": 76,
"layoutDurationMs": 27.594,
"taskDurationMs": 943.3190000000001,
"heapDeltaBytes": 18524400,
"heapUsedBytes": 85322764,
"domNodes": 76,
"jsHeapTotalBytes": 0,
"scriptDurationMs": 389.66200000000003,
"eventListeners": 876,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.665517241379302,
"p95FrameDurationMs": 16.800000000000182
},
{
"name": "agent-composer-empty-conversation-typing",
"durationMs": 1002.3599999999533,
"styleRecalcs": 78,
"styleRecalcDurationMs": 116.28099999999999,
"layouts": 76,
"layoutDurationMs": 28.078,
"taskDurationMs": 959.6139999999999,
"heapDeltaBytes": 18026100,
"heapUsedBytes": 88909600,
"domNodes": 76,
"jsHeapTotalBytes": 262144,
"scriptDurationMs": 389.098,
"eventListeners": 876,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.947457627118638,
"p95FrameDurationMs": 16.800000000000182
},
{
"name": "agent-composer-long-conversation-typing",
"durationMs": 1427.7000000000726,
"styleRecalcs": 78,
"styleRecalcDurationMs": 265.37,
"layouts": 76,
"layoutDurationMs": 29.585,
"taskDurationMs": 1409.1009999999999,
"heapDeltaBytes": -19439612,
"heapUsedBytes": 108349324,
"domNodes": 76,
"jsHeapTotalBytes": 262144,
"scriptDurationMs": 476.63999999999993,
"eventListeners": 876,
"totalBlockingTimeMs": 0,
"frameDurationMs": 18.37564102564104,
"p95FrameDurationMs": 33.30000000000109
},
{
"name": "agent-composer-long-conversation-typing",
"durationMs": 1441.7829999999867,
"styleRecalcs": 78,
"styleRecalcDurationMs": 267.92,
"layouts": 76,
"layoutDurationMs": 30.113999999999997,
"taskDurationMs": 1422.979,
"heapDeltaBytes": -19403636,
"heapUsedBytes": 107342128,
"domNodes": 76,
"jsHeapTotalBytes": 0,
"scriptDurationMs": 478.491,
"eventListeners": 876,
"totalBlockingTimeMs": 0,
"frameDurationMs": 18.354430379746834,
"p95FrameDurationMs": 33.29999999999927
}
]
}🔬 E2E Coverage
Top 10 files by uncovered lines
|
|
Reduce the counted diff or split the PR, then re-apply the |
|
Superseded by #19738, which carries the same graph-intent fix as a four-file diff directly from current |
This comment has been minimized.
This comment has been minimized.
|
Closing as superseded by #19738, which carries the focused fix on current main and merged green with human approval. |
Summary
mainwhile preserving the replacement graph-intent architectureset_widgetintents for buttons andserialize: falsedisplay widgets inWidgetValueStoreserializestate across registration/rebinding and synchronize runtime widget type changesThis is a companion PR into Kishore's branch; it does not push to the teammate-authored branch directly.
Review findings
Addresses all seven post-merge Cursor findings from #19504:
Validation
pnpm exec vitest run src/stores/widgetValueStore.test.ts src/lib/litegraph/src/widgets/BaseWidget.test.ts— 91/91 passedoxfmt --check— passed