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

"""Phase 157 E2E (Playwright): the hero chip's DISPLAY cap (90 chars +
``…``) with the phase-104 full-text contract extended (owner-locked
A1, 2026-10-03 audit: "long opening questions get hard-truncated
mid-word… the chips stretch near full container width before
clipping").

The contract this suite pins end to end (the user-facing half of the
phase — the unit pins in ``tests/unit/test_chip_sizing_question_cap.py``
are the source-level half):

* **capped display** — a >90-char saved-chat opener renders a
  COMPACT chip: the displayed text is
  ``slice(0, 90).trimEnd() + "…"`` — at most 91 chars, ending in
  ``…`` (the phase-104 single-line CSS clip demoted to the secondary
  guard); the display is the EXACT pinned string, asserted
  byte-for-byte on the live DOM, not just by length.
* **full text reachable** — the chip's ``title`` (the hover reveal)
  and, in the CLIPPED case (a 91-char display overflows the
  single-line box — the ≤640px viewport, where the chip's
  ``max-width: 100%`` binds and ``text-overflow: ellipsis``
  engages), the ``aria-label`` accessible name both carry the FULL
  question (phase 104, extended by the cap).
* **the click submits the FULL question** (the load-bearing
  assertion) — clicking a capped chip sends the UN-capped 120-char
  question: the user bubble is the full question, not the 91-char
  display (the cap is display-only — the phase-80 contract).
* **short-chip regression guard** — a <90-char opener renders
  byte-identical to pre-phase on the live DOM: the
  ``textContent`` is the full question with NO ``…`` appended, the
  ``title`` is the full question, and (unclipped at the 1280px
  viewport) no ``aria-label`` — the cap is a no-op below 90.

Seeding idiom (mirrored from the phase-80/103 suite
``tests/e2e/test_suggestion_chips.py``, not invented): the chips are
the session openers read from ``saved_chats`` — each test saves its
chat(s) via the signed-in admin's ``POST /api/chats`` and loads a
FRESH page, whose boot fetch renders the chips. The mock LLM is the
deterministic answer source (no ``E2E_REAL_LLM`` dependency); the
seeded questions are FIXED strings (the pins, not generated
lengths). The click test grounds on the fixture KB (the 120-char
opener carries the "kubernetes" token — the mock-calibrated cosine
clears the 0.30 gate, non-deflected). ``saved_chats`` is global state
on the shared e2e Postgres AND the state this contract reads — the
autouse fixture truncates it before and after EVERY test (including
the click test's auto-saved row), so each test starts from — and
leaves — an empty deployment.

Run in isolation (DB must be up: ``podman compose up -d db``):

    uv run pytest tests/e2e/test_chat_chip_cap.py -v --no-cov
"""
from __future__ import annotations

import asyncio
import json
import time
from collections.abc import Iterator
from pathlib import Path
from threading import Thread
from typing import Any

import pytest
from playwright.sync_api import Browser, Page, expect
from sqlalchemy import text

from app.config import Settings
from app.db import SessionLocal
from app.rag.importer import ImportSummary, import_sources
from app.rag.llm import LLMClient
from e2e.auth_helpers import login

REPO = Path(__file__).resolve().parents[2]
FIXTURES = REPO / "tests" / "fixtures" / "docs"
MOCK_ANSWER_MARKER = "Deterministic mock answer for E2E"

#: The FIXED 120-char opener (90+ — the cap engages deterministically;
#: the suite's pin, not a generated length). It carries the
#: "kubernetes" token (``tests/fixtures/docs/homelab/kubernetes.md``
#: + the deployments fixture), so the click test's turn grounds on
#: the mock-calibrated gate (0.30) — non-deflected, marker answer.
LONG_Q = (
    "What are the correct arguments for running a kubernetes node pool "
    "with qwen 3.8 27b on llama.cpp in the homelab cluster?"
)
assert len(LONG_Q) == 120, "the pin is a FIXED 120-char string"
#: The phase-157 display of LONG_Q (the exact-string pin — char 90 is
#: the "a" of "llama", so the trimEnd is a no-op here; the
#: space-at-char-90 trimEnd case is the unit pins' job). 91 chars.
LONG_Q_CAPPED = LONG_Q[:90].rstrip() + "…"
assert len(LONG_Q_CAPPED) == 91 and LONG_Q_CAPPED.endswith("…")
#: The follow-up (phase-103 semantics: it must NEVER surface as a
#: chip — a follow-up is meaningless as a conversation starter).
LONG_FOLLOW_UP = "What about the node pool's GPU drivers?"
#: Two <90-char openers for the three-chip state (the row fills to
#: the cap of 3 with the long opener NEWEST — the first chip).
SHORT_A = "Which provider is the primary DNS for reeseapps.com?"
SHORT_A_FOLLOW_UP = "What about the secondary DNS for the LAN?"
SHORT_B = "What cron schedule runs the restic prune?"
SHORT_B_FOLLOW_UP = "And where do the restic lock files live?"
#: The STANDALONE <90-char regression-guard question.
SHORT_Q = "How do I rotate the WireGuard keys on the VPN node?"
SHORT_Q_FOLLOW_UP = "What about the peers' allowed-ips?"


@pytest.fixture(autouse=True)
def clean_chats(db_ready: None) -> Iterator[None]:
    """``saved_chats`` is the state the onboarding chips read (the
    phase-103 openers) AND the click test's turn auto-saves a row —
    truncate it before and after EVERY test so each starts from (and
    leaves) an empty deployment (the seed-chip state), exactly the
    phase-80/103/104 autouse pattern."""
    with SessionLocal() as db:
        db.execute(text("TRUNCATE saved_chats"))
        db.commit()
    yield
    with SessionLocal() as db:
        db.execute(text("TRUNCATE saved_chats"))
        db.commit()


async def _import_fixtures(mock_port: int) -> ImportSummary:
    kwargs: dict[str, Any] = {"_env_file": None, "llm_base_url": f"http://127.0.0.1:{mock_port}/v1"}
    settings = Settings(**kwargs)  # pyright: ignore[reportCallIssue]
    return await import_sources([FIXTURES], LLMClient(settings))


def _run_in_thread(coro: Any) -> Any:
    """Run a coroutine on a worker thread (Playwright owns the test loop)."""
    box: dict[str, Any] = {}

    def runner() -> None:
        try:
            box["value"] = asyncio.run(coro)
        except BaseException as e:  # noqa: BLE001 — re-raised on the test thread
            box["error"] = e

    t = Thread(target=runner)
    t.start()
    t.join()
    if "error" in box:
        raise box["error"]
    return box["value"]


def _seed_kb(mock_port: int) -> ImportSummary:
    """Deterministic KB: truncate the KB tables, import the fixture
    docs (needed by the click test's grounded answer).
    ``saved_chats`` is the autouse fixture's job."""
    with SessionLocal() as db:
        db.execute(text("TRUNCATE chunks, documents, query_log"))
        db.commit()
    summary = _run_in_thread(_import_fixtures(mock_port))
    assert summary is not None and summary.added == 13  # A9 formats (phase 47 added quadlet+j2)
    return summary


def _user(q: str) -> dict[str, Any]:
    return {"who": "user", "text": q}


def _brain(text: str = "Grounded mock brain reply.") -> dict[str, Any]:
    return {"who": "brain", "text": text}


def _save_chat(page: Page, app_url: str, messages: list[dict[str, Any]]) -> dict[str, Any]:
    """Save one conversation as the signed-in admin (``POST /api/chats``)
    and return the 201 body. ``updated_at`` is the API's stamp (server
    ``now()`` at INSERT) — the save ORDER is what makes the chip-walk
    order deterministic in the multi-chat test."""
    r = page.request.post(
        f"{app_url}/api/chats",
        data=json.dumps({"messages": messages}),
        headers={"Content-Type": "application/json"},
        timeout=10_000,
    )
    assert r.status == 201, r.text
    return r.json()


def _chips(page: Page) -> Any:
    return page.locator("#suggestions .suggestion-chip")


def test_capped_display(page: Page, app_url: str, db_ready: None) -> None:
    """The >90-char opener renders a COMPACT chip: the displayed text
    is the EXACT ``slice(0, 90).trimEnd() + "…"`` pin (91 chars,
    ending in ``…``), and the row is the phase-80/103 three-chip
    contract — the three seeded openers, newest ``updated_at``
    first; the <90-char openers render byte-identical (no ``…``), and
    none of the chats' follow-ups surfaces."""
    page.set_default_timeout(30_000)
    login(page, app_url, next="/")

    # Save oldest→newest: the API stamps ``updated_at`` (server
    # now()), so save order IS walk order (newest first). The short
    # pauses keep the three stamps strictly apart.
    stamps: list[str] = []
    for opener, follow_up in (
        (SHORT_A, SHORT_A_FOLLOW_UP),
        (SHORT_B, SHORT_B_FOLLOW_UP),
        (LONG_Q, LONG_FOLLOW_UP),
    ):
        body = _save_chat(
            page,
            app_url,
            [
                _user(opener), _brain(),
                _user(follow_up), _brain(),
            ],
        )
        stamps.append(body["updated_at"])
        time.sleep(0.05)
    assert stamps == sorted(stamps) and len(set(stamps)) == 3, (
        "the three API-stamped updated_at values must be strictly increasing"
    )

    # A FRESH page load (a new boot fetch): EXACTLY the three
    # openers, newest first — the long one capped, the shorts not.
    page.goto(app_url + "/")
    chips = _chips(page)
    expect(chips.first).to_be_visible(timeout=30_000)
    expect(chips).to_have_count(3, timeout=15_000)

    # The CAP (the phase-157 A1 contract, live DOM, byte-for-byte):
    # the 120-char opener shows as EXACTLY slice(0, 90).trimEnd() +
    # "…" — 91 chars, ending in the ellipsis, not the 120-char
    # pre-phase display (which stretched the chip near full row
    # width before the CSS clip engaged — the audit's observation).
    display = chips.nth(0).text_content()
    assert display is not None
    assert len(display) <= 91, (
        f"a >90-char opener must display ≤91 chars — the cap + the "
        f"… — got {len(display)}"
    )
    assert display.endswith("…"), "a capped chip ends with the ellipsis"
    assert display == LONG_Q_CAPPED, (
        "the display is the EXACT slice(0, 90).trimEnd() + '…' pin, "
        "not just a length"
    )
    # The row order (phase 80/103 — newest updated_at first, cap 3)
    # and the <90-char openers' byte-identity (no cap, no "…").
    assert chips.nth(1).text_content() == SHORT_B, "a <90-char opener is byte-identical (no …)"
    assert chips.nth(2).text_content() == SHORT_A, "a <90-char opener is byte-identical (no …)"
    for follow_up in (SHORT_A_FOLLOW_UP, SHORT_B_FOLLOW_UP, LONG_FOLLOW_UP):
        assert follow_up not in [
            chips.nth(i).text_content() for i in range(3)
        ], "a chat's follow-up must never chip"


def test_full_text_reachable(browser: Browser, app_url: str, db_ready: None) -> None:
    """The full text stays one hover away AND accessible in the
    CLIPPED case: at the ≤640px viewport the 91-char display
    overflows the single-line box (the chip's ``max-width: 100%``
    binds, ``text-overflow: ellipsis`` engages — the phase-104 CSS,
    the secondary guard), so the phase-104 idiom applies: the
    ``title`` carries the FULL question (universal — every chip) and
    the ``aria-label`` (set ONLY when visually clipped, the
    ``scrollWidth > clientWidth`` check) carries the FULL question
    — never the 91-char display (phase 104, extended by the cap)."""
    page = browser.new_page(viewport={"width": 375, "height": 720})
    try:
        page.set_default_timeout(30_000)
        login(page, app_url, next="/")

        # ONE multi-turn chat: the 120-char opener + the follow-up
        # (phase-103: the follow-up never surfaces).
        _save_chat(
            page,
            app_url,
            [
                _user(LONG_Q), _brain(),
                _user(LONG_FOLLOW_UP), _brain(),
            ],
        )

        # A FRESH page load: EXACTLY ONE chip — the opener.
        page.goto(app_url + "/")
        chips = _chips(page)
        expect(chips.first).to_be_visible(timeout=30_000)
        expect(chips).to_have_count(1, timeout=15_000)
        chip = chips.first
        # Precondition: the CLIPPED case at this viewport (the 91-char
        # display overflows the single-line box — pin the state the
        # aria-label branch keys on, so a layout regression cannot
        # silently flip the attribute's contract).
        dims = chip.evaluate(
            "el => ({ sw: el.scrollWidth, cw: el.clientWidth })"
        )
        assert dims["sw"] > dims["cw"], (
            "at the ≤640px viewport the 91-char display must overflow "
            "the single-line box (the clipped case the aria-label "
            "idiom governs)"
        )
        # The hover reveal (phase 104, universal): the FULL question.
        assert chip.get_attribute("title") == LONG_Q, (
            "the title tooltip must carry the FULL question (phase 104)"
        )
        # The clipped-case accessible name (phase 104 idiom — the
        # scrollWidth > clientWidth check): the FULL question.
        assert chip.get_attribute("aria-label") == LONG_Q, (
            "a clipped chip's aria-label must carry the FULL "
            "question — never the 91-char display"
        )
    finally:
        page.close()


def test_click_submits_full(
    page: Page, app_url: str, mock_llm: int, db_ready: None
) -> None:
    """The load-bearing assertion: clicking a CAPPED chip submits the
    FULL (120-char) question — the cap is display-only. The user
    bubble is the EXACT full question (byte-for-byte, not the 91-char
    display), and the turn is the grounded mock's (the opener carries
    the "kubernetes" token — non-deflected, marker answer); the turn
    clears the input and the send button recovers (never stale)."""
    _seed_kb(mock_llm)
    page.set_default_timeout(30_000)
    login(page, app_url, next="/")

    # ONE multi-turn chat: the 120-char opener + the follow-up.
    _save_chat(
        page,
        app_url,
        [
            _user(LONG_Q), _brain(),
            _user(LONG_FOLLOW_UP), _brain(),
        ],
    )

    # A FRESH page load: EXACTLY ONE chip — the capped opener.
    page.goto(app_url + "/")
    chips = _chips(page)
    expect(chips.first).to_be_visible(timeout=30_000)
    expect(chips).to_have_count(1, timeout=15_000)
    chip = chips.first
    # Precondition: the chip that gets clicked IS the capped display
    # (the 91-char pin — the click test is about the CAPPED chip).
    assert chip.text_content() == LONG_Q_CAPPED

    # One tap = one question: the click fills AND submits — with the
    # FULL text (the phase-80 contract, the load-bearing assertion).
    chip.click()
    expect(page.locator("#empty-state")).to_be_hidden()
    expect(page.locator("#message-input")).to_have_value("")  # submitted, not queued
    bubble = page.locator(".msg.user .bubble")
    expect(bubble).to_have_count(1, timeout=30_000)
    # The user bubble is the EXACT FULL question — byte-for-byte, not
    # the 91-char capped display (the cap is display-only).
    assert bubble.inner_text() == LONG_Q, (
        "the capped chip must submit the FULL question — the user "
        "bubble is the un-capped 120-char text, not the display"
    )
    assert len(bubble.inner_text()) == 120

    # The grounded mock reply follows (the seeded KB answers this
    # topic — the "kubernetes" token grounds, non-deflected).
    brain = page.locator(".msg.brain .bubble").first
    expect(brain).to_contain_text(MOCK_ANSWER_MARKER, timeout=30_000)
    expect(page.locator(".msg.brain.is-deflected")).to_have_count(0)

    # Never stale: the turn cleared the input and the send button
    # recovered.
    expect(page.locator("#send-btn")).to_be_enabled()
    expect(page.locator("#send-label")).to_have_text("Send")


def test_short_chip_unchanged(page: Page, app_url: str, db_ready: None) -> None:
    """The regression guard on the live DOM: a <90-char opener renders
    byte-identical to pre-phase — the ``textContent`` is the full
    question with NO ``…`` appended (the cap is a no-op below 90),
    the ``title`` is the full question (phase 104, unchanged), and
    (unclipped at the 1280px viewport) no ``aria-label`` (the
    ``scrollWidth > clientWidth`` check — the attribute is set only
    when visually clipped, phase 104)."""
    page.set_default_timeout(30_000)
    login(page, app_url, next="/")

    # ONE multi-turn chat: the 49-char opener + the follow-up.
    _save_chat(
        page,
        app_url,
        [
            _user(SHORT_Q), _brain(),
            _user(SHORT_Q_FOLLOW_UP), _brain(),
        ],
    )

    # A FRESH page load: EXACTLY ONE chip — the short opener.
    page.goto(app_url + "/")
    chips = _chips(page)
    expect(chips.first).to_be_visible(timeout=30_000)
    expect(chips).to_have_count(1, timeout=15_000)
    chip = chips.first
    # Byte-identical to pre-phase: the full question, no "…".
    assert chip.text_content() == SHORT_Q, (
        "a <90-char chip's display is byte-identical to pre-phase — "
        "the full question, no ellipsis appended"
    )
    assert "…" not in (chip.text_content() or "")
    # The hover reveal: the full question (phase 104, unchanged).
    assert chip.get_attribute("title") == SHORT_Q
    # Unclipped at the 1280px viewport (pin the state — the
    # pre-phase display fit, so the cap must not change the
    # aria-label contract either): no aria-label.
    dims = chip.evaluate(
        "el => ({ sw: el.scrollWidth, cw: el.clientWidth })"
    )
    assert dims["sw"] <= dims["cw"], (
        "a 49-char chip must NOT be clipped at the 1280px viewport "
        "(the pre-phase state — byte-identical)"
    )
    assert chip.get_attribute("aria-label") is None, (
        "an unclipped chip stays attribute-free (phase 104)"
    )
    # The follow-up never surfaces (phase 103 — the row is the
    # opener only).
    assert SHORT_Q_FOLLOW_UP not in [chip.text_content()]