brain-of-reese/tests/e2e/test_configurable_brand.py

"""Phase 39 E2E (Playwright): the configurable app name (``BOR_APP_NAME``).

Story: ``.agent/user_stories/configurable-brand.md``
Run in isolation (DB must be up: ``podman compose up -d db``):

    uv run pytest tests/e2e/test_configurable_brand.py -v --no-cov

Contract under test (owner ask — "a way to customize the name for
'Brain of'. Should be an env var"):

* one env var (``BOR_APP_NAME``, default "Brain of Reese") drives the
  display name on EVERY page — the ``<title>``s, the header
  ``.brand-text`` brand, the empty-state greeting, the ``#messages``
  aria-label, and the chat status labels — via the public
  ``GET /api/config`` + the classic ``assets/brand.js`` layer;
* a name starting ``"Brain of "`` keeps the bold split (``Brain of
  <strong>rest</strong>``); the name is HTML-escaped (the story's
  acceptance criteria, AC-2);
* with the variable UNSET the app is byte-identical to today — the
  shared conftest server (the default name every other suite's
  title/label assertions depend on) still renders "Brain of Reese".

Determinism note: this story needs a SECOND app instance — the shared
conftest server keeps the default name, so ``testy_server`` boots the
same env block (same DB, the mock-LLM base URL, the admin auth, the
static dir, the mock-calibrated threshold) with exactly two changes:
``BOR_APP_NAME="Brain of Testy"`` and port ``APP_PORT + 1``. Every
assertion is settled-state: Playwright's ``expect`` retries ride out
the brand.js ``/api/config`` fetch (the name is applied asynchronously
after the fetch answers). The one chat turn is pre-token-labeled — the
``think out loud`` marker makes the mock stream ~2.7k chars of
reasoning at 12 chars / 0.02 s (~4.6 s) before the first content frame,
so the "… is thinking" label is stable for the whole window regardless
of what the turn answers (grounded or deflected — no DB seeding
needed).

Test → story mapping (Playwright Mapping Rule):
1. ``test_api_config_serves_both_names``
2. ``test_index_carries_the_custom_brand``
3. ``test_other_pages_titles_carry_the_name``
4. ``test_chat_status_label_uses_custom_name_pre_token``
5. ``test_shared_server_keeps_the_default_name``
"""
from __future__ import annotations

import os
import subprocess
import sys
from collections.abc import Iterator

import httpx
import pytest
from playwright.sync_api import Page, expect

from e2e.conftest import (
    ADMIN_PASSWORD,
    APP_PORT,
    REPO,
    SESSION_SECRET,
    USE_REAL_LLM,
    _wait_http,
)

TESTY_PORT = APP_PORT + 1
TESTY_URL = f"http://127.0.0.1:{TESTY_PORT}"
TESTY_NAME = "Brain of Testy"
DEFAULT_NAME = "Brain of Reese"


def expect_inner_html(page: Page, selector: str, expected: str, timeout: int = 15_000) -> None:
    """Retrying innerHTML equality (Playwright's Python API ships no
    ``to_have_inner_html`` assertion). The brand layer applies the name
    when its ``/api/config`` fetch answers, so the check must poll —
    ``wait_for_function`` does, on Playwright's own retry clock."""
    page.wait_for_function(
        """([sel, expected]) => {
            const el = document.querySelector(sel);
            return el !== null && el.innerHTML === expected;
        }""",
        arg=[selector, expected],
        timeout=timeout,
    )


@pytest.fixture(scope="session")
def testy_server(mock_llm: int) -> Iterator[str]:
    """A SECOND app instance, booted with the overridden name.

    The shared conftest ``app_server`` keeps the default name (every
    other suite's title/label assertions depend on it) — so this
    fixture copies its env block verbatim (same DB, the mock-LLM base
    URL, ``BOR_ADMIN_PASSWORD``/``BOR_SESSION_SECRET``,
    ``BOR_STATIC_DIR``, ``BOR_RELEVANCE_THRESHOLD``) with exactly two
    changes: ``BOR_APP_NAME="Brain of Testy"`` and a distinct port
    (``APP_PORT + 1`` per the conftest convention). Started after
    ``mock_llm`` is available (its fixture dependency).
    """
    env = dict(os.environ)
    env.pop("DEBUGPY", None)
    env["BOR_ENVIRONMENT"] = "e2e"
    env["BOR_STATIC_DIR"] = str(REPO / "frontend")
    env["BOR_LLM_BASE_URL"] = (
        "https://aipi.reeseapps.com/v1"
        if USE_REAL_LLM
        else f"http://127.0.0.1:{mock_llm}/v1"
    )
    # The mock's token-overlap embeddings have their own score
    # distribution — the same mock-calibrated threshold as the shared
    # instance, so this suite's turn behaves like every other story's.
    env["BOR_RELEVANCE_THRESHOLD"] = "0.30"
    env.setdefault(
        "BOR_DATABASE_URL",
        "postgresql+psycopg://reese:reese@localhost:5432/brain_of_reese",
    )
    # Phase 16: admin auth must be set or create_app() refuses to boot.
    env["BOR_ADMIN_PASSWORD"] = ADMIN_PASSWORD
    env["BOR_SESSION_SECRET"] = SESSION_SECRET
    # The phase-39 override — the whole story:
    env["BOR_APP_NAME"] = TESTY_NAME
    proc = subprocess.Popen(
        [sys.executable, "-m", "uvicorn", "app.main:app",
         "--host", "127.0.0.1", "--port", str(TESTY_PORT), "--log-level", "warning"],
        cwd=REPO,
        env=env,
    )
    try:
        _wait_http(f"{TESTY_URL}/api/health")
        yield TESTY_URL
    finally:
        proc.terminate()
        try:
            proc.wait(timeout=10)
        except subprocess.TimeoutExpired:
            proc.kill()


# ---------------------------------------------------------------------------
# 1. The endpoint the brand layer reads — overridden on the second
#    instance, default on the shared one
# ---------------------------------------------------------------------------


def test_api_config_serves_both_names(testy_server: str, app_server: str) -> None:
    r = httpx.get(f"{TESTY_URL}/api/config", timeout=5)
    assert r.status_code == 200
    body = r.json()
    assert set(body) == {"app_name", "version"}
    assert body["app_name"] == TESTY_NAME

    # The shared conftest instance keeps the default (the other
    # suites' title/label contract rides on it).
    r2 = httpx.get(f"{app_server}/api/config", timeout=5)
    assert r2.status_code == 200
    assert r2.json()["app_name"] == DEFAULT_NAME


# ---------------------------------------------------------------------------
# 2. The chat page — title, header brand, greeting, stream aria-label
# ---------------------------------------------------------------------------


def test_index_carries_the_custom_brand(page: Page, testy_server: str) -> None:
    page.goto(testy_server + "/")
    # The brand layer applies the name when /api/config answers — the
    # retrying expectations ride out that fetch.
    expect(page).to_have_title(TESTY_NAME, timeout=15_000)
    # The "Brain of …" name keeps the bold split (AC-2) — and the name
    # arrives escaped (Testy has no special characters, but the markup
    # shape is the contract).
    expect_inner_html(page, ".brand-text", "Brain of <strong>Testy</strong>")
    # The empty-state greeting (the TreeWalker prose pass).
    expect(page.locator("#empty-state h1")).to_have_text(
        f"Hey! I'm {TESTY_NAME}.", timeout=15_000
    )
    # The message stream's accessible label (the attribute pass).
    expect(page.locator("#messages")).to_have_attribute(
        "aria-label", f"Conversation with {TESTY_NAME}", timeout=15_000
    )


# ---------------------------------------------------------------------------
# 3. The other pages — their <title>s carry the name too
# ---------------------------------------------------------------------------


def test_other_pages_titles_carry_the_name(page: Page, testy_server: str) -> None:
    page.goto(testy_server + "/sources.html")
    expect(page).to_have_title(f"Sources · {TESTY_NAME}", timeout=15_000)

    page.goto(testy_server + "/login.html")
    expect(page).to_have_title(f"Sign in · {TESTY_NAME}", timeout=15_000)


# ---------------------------------------------------------------------------
# 4. The chat status label — a pre-token turn on the custom instance
# ---------------------------------------------------------------------------


def test_chat_status_label_uses_custom_name_pre_token(
    page: Page, testy_server: str, db_ready: None
) -> None:
    page.goto(testy_server + "/")
    expect(page.locator("#send-btn")).to_be_enabled(timeout=15_000)

    # The ``think out loud`` marker (mock_llm.py) makes the answer
    # stream ~2.7k chars of reasoning FIRST — the pre-token window.
    # Works on the default (possibly empty) KB: a deflected answer is
    # fine, the label assertion is pre-token, no seeding required.
    page.fill("#message-input", "think out loud — what is the capital of France?")
    page.click("#send-btn")

    # The pre-token status label carries the configured name (the
    # ~4.6 s reasoning stream keeps the state "thinking" — the label
    # is stable for the whole window).
    expect(page.locator("#send-status")).to_have_text(
        f"{TESTY_NAME} is thinking", timeout=15_000
    )

    # The turn always finishes — the button is never left stale
    # (PLAN §7.4), whatever the turn answered.
    expect(page.locator("#send-label")).to_have_text("Send", timeout=60_000)


# ---------------------------------------------------------------------------
# 5. The no-op regression — the shared default-name instance is
#    byte-identical to today
# ---------------------------------------------------------------------------


def test_shared_server_keeps_the_default_name(page: Page, app_server: str) -> None:
    page.goto(app_server + "/")
    expect(page).to_have_title(DEFAULT_NAME, timeout=15_000)
    # With the variable unset the brand pass is a no-op: the header
    # renders exactly the pre-phase-39 markup.
    expect_inner_html(page, ".brand-text", "Brain of <strong>Reese</strong>")