Files
copilotkit__copilotkit/showcase/integrations/langgraph-python/tests/e2e/declarative-gen-ui.spec.ts
Alem Tuzlak ccc2eabfc2 test(showcase/langgraph-python): regression coverage for A2UI middleware-intercept + display_flight loop
Three classes of regression are now pinned:

1. Secondary-LLM tool name doesn't collide with the A2UI middleware's
   default intercept list (`render_a2ui`). New
   `src/agents/test_a2ui_internal_tools.py` parametrises over
   `beautiful_chat._design_a2ui_surface`, `a2ui_dynamic._design_a2ui_surface`,
   and `a2ui_fixed.display_flight` and asserts none match the
   middleware's `a2uiToolNames` default. Catches accidental rename
   reverts that would re-enable the bypass.

2. `generate_a2ui` force-pins the canonical `catalog_id` even when the
   secondary LLM hallucinates a wrong one. The new test stubs
   `ChatOpenAI` with a fake response carrying a bogus catalogId and
   asserts the surface op carries the module's `CUSTOM_CATALOG_ID`.

3. `generate_a2ui` short-circuits with a clean error string when the
   LLM emits a root component without a `component` field — never
   feeds the renderer the partial tree that surfaced as the "Cannot
   create component root without a type" infinite-loop.

7 unit tests, all green locally (`pytest src/agents/test_a2ui_internal_tools.py`).

E2E tests on the same fixes now also assert:
- No `A2UI render error: Catalog not found` banner on the page after
  Beautiful Chat → Sales Dashboard, Declarative Gen UI → BarChart, and
  A2UI Fixed Schema → Find SFO → JFK round-trips.
- No `Cannot create component … without a type` banner on the same
  three pills.
- Exactly ONE flight card on A2UI Fixed Schema (was 6+ on deploy
  pre-fix from the `display_flight` loop) — `Flight Details` count
  pinned to 1, `Book flight` count pinned to 1.
- At most one ResponsiveContainer on the BarChart pill (loops would
  stack multiple).
- At most two ResponsiveContainers on Beautiful Chat → Sales Dashboard
  (one pie + one bar = single dashboard render).
2026-05-08 19:07:52 +02:00

175 lines
7.5 KiB
TypeScript

import { test, expect } from "@playwright/test";
// QA reference: qa/declarative-gen-ui.md
// Demo source: src/app/demos/declarative-gen-ui/{page.tsx, a2ui/*}
//
// Pattern: A2UI dynamic-schema BYOC. The frontend registers a 7-component
// catalog (Card, StatusBadge, Metric, InfoRow, PrimaryButton, PieChart,
// BarChart) via `a2ui={{ catalog: myCatalog }}`. The Python agent
// (`src/agents/a2ui_dynamic.py`) owns the `generate_a2ui` tool and emits an
// `a2ui_operations` container with `catalogId: "declarative-gen-ui-catalog"`.
// The secondary LLM inside `generate_a2ui` produces a JSON component tree
// that the A2UI renderer binds to the registered React catalog.
//
// There is no `data-testid` in the demo source. We rely on verbatim
// suggestion-pill text and the inline-style fingerprints exported by
// `a2ui/renderers.tsx` (donut SVG, recharts markers, lilac/mint brand
// colours, etc.). Because the secondary-LLM render is multi-step, the
// surface can take 30-60s to paint — all render assertions use a 60s budget.
//
// W8-7: on Railway (showcase-langgraph-python-production.up.railway.app),
// the `generate_a2ui` tool path is occasionally slow / flaky — the KPI and
// StatusReport flows can exceed a 60s budget when the secondary LLM stalls.
// Those two scenarios are skipped; the deterministic PieChart and BarChart
// catalog renderers are kept as the primary render-signal tests since they
// have the strongest visual fingerprints. Un-skip when the agent deployment
// stabilises.
test.describe("Declarative Generative UI (A2UI dynamic schema)", () => {
test.setTimeout(120_000);
test.beforeEach(async ({ page }) => {
await page.goto("/demos/declarative-gen-ui");
});
test("page loads with chat input and no surface rendered", async ({
page,
}) => {
await expect(page.getByPlaceholder("Type a message")).toBeVisible();
// No A2UI surface rendered on first paint (no donut SVG, no recharts
// container).
await expect(page.locator(".recharts-responsive-container")).toHaveCount(0);
});
test("all 4 suggestion pills render with verbatim titles", async ({
page,
}) => {
const suggestions = page.locator('[data-testid="copilot-suggestion"]');
const expected = [
"Show a KPI dashboard",
"Pie chart — sales by region",
"Bar chart — quarterly revenue",
"Status report",
];
for (const title of expected) {
await expect(suggestions.filter({ hasText: title }).first()).toBeVisible({
timeout: 15_000,
});
}
});
test("PieChart pill renders a donut SVG with slice circles + legend %", async ({
page,
}) => {
// The custom DonutChart renderer (a2ui/renderers.tsx) builds an inline
// <svg> with one grey background <circle> + one stroked <circle> per
// slice, wrapped in `transform: scaleX(-1)`. The legend rows end in a
// percentage like "45%". This is the strongest visual fingerprint of a
// correctly-bound catalog PieChart node.
const suggestions = page.locator('[data-testid="copilot-suggestion"]');
await suggestions
.filter({ hasText: "Pie chart — sales by region" })
.first()
.click();
// At least background circle + 2 slice circles. 90s budget: on
// cold starts the secondary-LLM `generate_a2ui` pass can eat most
// of a minute before emitting the PieChart node.
const circles = page.locator("svg circle");
await expect
.poll(async () => await circles.count(), { timeout: 90_000 })
.toBeGreaterThanOrEqual(3);
// A legend row with an integer percentage (e.g. "45%").
await expect(page.getByText(/\b\d+%/).first()).toBeVisible({
timeout: 10_000,
});
});
test("BarChart pill renders a recharts bar chart with rectangles", async ({
page,
}) => {
// BarChart renderer uses a recharts ResponsiveContainer (height 280) +
// a custom shape (AnimatedBar with `barSlideIn` keyframe). We only
// assert on stable recharts markers (class names unchanged across
// versions) — the keyframe-specific CSS is a visual detail not worth
// asserting via DOM.
const suggestions = page.locator('[data-testid="copilot-suggestion"]');
await suggestions
.filter({ hasText: "Bar chart — quarterly revenue" })
.first()
.click();
// 90s budget for the same cold-start reason as PieChart above.
const barChartRoot = page.locator(".recharts-responsive-container").first();
await expect(barChartRoot).toBeVisible({ timeout: 90_000 });
// At least 2 bar rectangles should render. The custom shape renders a
// recharts <Rectangle> inside a <g>, which keeps the standard class.
const bars = page.locator(".recharts-bar-rectangle");
await expect
.poll(async () => await bars.count(), { timeout: 15_000 })
.toBeGreaterThanOrEqual(2);
// Regression guard (#4734): the deployed KPI / dashboard pills used to
// loop with "A2UI render error: Cannot create component root without a
// type" because the secondary LLM's `render_a2ui` tool call was
// intercepted by the A2UI middleware before our defensive validation
// could drop malformed components. Renaming to `_design_a2ui_surface`
// killed the bypass; assert no A2UI render-error banners are visible.
await expect(
page.getByText(/Cannot create component .* without a type/i),
).toHaveCount(0);
await expect(page.getByText(/Catalog not found/i)).toHaveCount(0);
// Regression guard: only one bar chart surface (one ResponsiveContainer)
// should render — looping renders would stack multiple.
const allCharts = page.locator(".recharts-responsive-container");
await expect
.poll(async () => await allCharts.count(), { timeout: 5_000 })
.toBeLessThanOrEqual(1);
});
// SKIP: KPI dashboard prompt drives `generate_a2ui` to emit a Card +
// multiple Metric tiles. On Railway this path is the slowest of the 4
// pills and regularly exceeds 60s when the secondary LLM stalls. See
// W8-7. Un-skip when the agent deployment stabilises.
test.skip("KPI dashboard pill renders at least 3 Metric tiles", async ({
page,
}) => {
const suggestions = page.locator('[data-testid="copilot-suggestion"]');
await suggestions
.filter({ hasText: "Show a KPI dashboard" })
.first()
.click();
// Each Metric renders an uppercase label with `letterSpacing: 0.12em`
// above a large number. The label styling is the stable fingerprint;
// the text content itself is model-generated and not asserted.
const metricLabels = page.locator(
'div[style*="letter-spacing: 0.12em"], div[style*="letterSpacing: 0.12em"]',
);
await expect
.poll(async () => await metricLabels.count(), { timeout: 90_000 })
.toBeGreaterThanOrEqual(3);
});
// SKIP: StatusReport prompt expects at least one Card + StatusBadge pill.
// Same Railway slowness as KPI — often exceeds 60s. See W8-7.
test.skip("Status report pill renders a Card with a StatusBadge pill", async ({
page,
}) => {
const suggestions = page.locator('[data-testid="copilot-suggestion"]');
await suggestions.filter({ hasText: "Status report" }).first().click();
// StatusBadge style: `borderRadius: 999` (pill), uppercase + 0.1em
// letter-spacing. This combo is unique to the badge renderer.
const badges = page.locator(
'span[style*="border-radius: 999"][style*="letter-spacing: 0.1em"], span[style*="borderRadius: 999"][style*="letterSpacing: 0.1em"]',
);
await expect
.poll(async () => await badges.count(), { timeout: 90_000 })
.toBeGreaterThanOrEqual(1);
});
});