wehub-resource-sync
76 lines
3.3 KiB
Python
76 lines
3.3 KiB
Python
"""Smoke test: build SPA, spawn server, send a message, see a response.
|
|
|
|
This is the bootstrap test for ``tests/e2e_ui/`` — it verifies the
|
|
full chain (Vite build → static mount → React boot → SSE stream →
|
|
DOM render) is wired correctly. It is intentionally minimal; richer
|
|
coverage (multi-turn, refresh hydration, stop-cancel, deep-link
|
|
errors) lands in follow-up tests once stable selectors are in place.
|
|
|
|
Selectors are accessibility-first where they're stable: the textarea
|
|
is found by its placeholder, the Send button by its accessible name
|
|
(a hidden ``<span class="sr-only">Send</span>`` per
|
|
``web/src/pages/ChatPage.tsx``). Real message bubbles use
|
|
``data-testid="message-bubble"`` + ``data-role={user|assistant}``.
|
|
Without the testid we can't distinguish the streaming "Working…"
|
|
shimmer (also rendered as ``<Message from="assistant">``) from a
|
|
real assistant bubble — the test would pass on the spinner.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import re
|
|
|
|
from playwright.sync_api import Page, expect
|
|
|
|
|
|
def test_send_message_renders_assistant_response(
|
|
page: Page,
|
|
seeded_session: tuple[str, str],
|
|
) -> None:
|
|
"""
|
|
Open a pre-created session, type a prompt, click Send, and assert
|
|
an assistant bubble renders with non-empty text.
|
|
|
|
A failure here means one of:
|
|
|
|
- The SPA didn't boot (build missing or static mount broken in
|
|
``omnigent/server/app.py``'s ``_SPAStaticFiles``).
|
|
- The composer is mis-wired (``ChatPage.tsx`` regression).
|
|
- The agent never received the request (server / runtime
|
|
regression — check the live_server log).
|
|
- The LLM never responded (Databricks credentials missing or
|
|
hello_world model unavailable).
|
|
- The SDK reducer didn't render output (TS reducer parity drift
|
|
vs ``omnigent_client/_stream.py`` — see
|
|
``web/README.md`` § Reducer parity).
|
|
|
|
Starts from ``/c/<id>`` rather than ``/`` because the home route
|
|
no longer renders a composer — see :func:`seeded_session`.
|
|
|
|
:param page: Playwright page fixture (function-scoped, fresh
|
|
browser context per test).
|
|
:param seeded_session: ``(base_url, session_id)`` from the
|
|
pre-created session fixture.
|
|
"""
|
|
base_url, session_id = seeded_session
|
|
page.goto(f"{base_url}/c/{session_id}")
|
|
|
|
composer = page.get_by_placeholder("Ask the agent anything…")
|
|
expect(composer).to_be_visible()
|
|
composer.fill("Say 'pong' in one word.")
|
|
page.get_by_role("button", name="Send", exact=True).click()
|
|
|
|
# The URL should already match /c/<id> since we started there.
|
|
expect(page).to_have_url(re.compile(rf"/c/{re.escape(session_id)}"))
|
|
|
|
# Wait for a real assistant bubble (NOT the "Working…" shimmer
|
|
# — that has data-testid="working-indicator" instead). 60s budget
|
|
# covers cold-start LLM latency under Databricks routing without
|
|
# masking a true hang. ``re.compile(r"\S")`` matches any non-
|
|
# whitespace character — a rendered bubble whose MessageContent
|
|
# is empty would mean the streaming reducer fired but produced
|
|
# no text, itself a regression worth surfacing.
|
|
assistant = page.locator('[data-testid="message-bubble"][data-role="assistant"]').first
|
|
expect(assistant).to_be_visible(timeout=60_000)
|
|
expect(assistant).to_have_text(re.compile(r"\S"), timeout=60_000)
|