"""Phase 149 task 03 E2E (Playwright, mock-only): the owner's ask, pinned
end to end — while a sync run is in flight, the **button is the loading
spinner** (``is-running`` hides the label parts; the spinning icon is the
whole control) and a NEW bordered text box directly under the button row
(``#sync-progress``) carries the **EXACT full untruncated live label** —
``Syncing… <source/relative/path> (n/m) · chunk n/m`` — wrapping, never
clipping (the long ``source/relative/path`` that overflows the pill is the
fixture's whole reason to exist). At settle the box hides, the pill is
restored byte-identical to pre-phase (``Synced HH:MM`` + the ``fmtSyncResult``
counts line on ``#sync-result``), and ``#sync-result`` stayed the SOLE
aria-live announcer the whole time (the box is presentational,
``aria-hidden``).

The dedicated story suite for ``149_sync_progress_box`` (A16 — one
Playwright file per phase, run in isolation). Tasks 01/02 built the box +
its wiring + its CSS; this suite is the behavioral proof through the REAL
sync path: a real ``git clone`` of a local ``file://`` fixture repo, a real
import, the slowed mock LLM so the in-flight state is observable.

Run in isolation (DB must be up: ``podman compose up -d db``; git on PATH —
a documented environment prerequisite, phase 28):

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

MOCK-ONLY suite: ``E2E_REAL_LLM=1`` is not supported — the gate is the
deterministic mock plus the phase-64 ``tests/e2e/slow_llm.py`` delay proxy
(``SLOW_DELAY_S = "1.5"`` — the proxy delays EVERY LLM call, embeds
included) in front of the mock: each in-flight state outlives the UI's
2 s status poll by ≫ 1×, so every rendered label is observed
deterministically, not as a race.

The suite runs ONE module app (the phase-79 full-inventory pattern — it
binds its own port) over the shared E2E Postgres: the SLOW app with
``BOR_GIT_SOURCES=file://<fixture repo>`` (the env fallback — the autouse
truncate keeps the ``git_sources`` table empty, so the env list IS the
effective source) + its own ``BOR_SOURCES_DIR`` for the checkout.

Timing fixture: a changed sync of the 3-file fixture is ≈ 2 model probes
+ 3 files × (1 content-embed batch + 1 summary chat + 1 summary-chunk
embed) + 1 overview + 7 folder summaries ≈ 19 LLM calls ≈ 30–35 s under
the proxy. The LONG-path file is the FIRST of the three (the sorted walk —
``notes/a-…`` < ``notes/short-…``), so its import span (7–8 embed batches
× 1.5 s + the two summary calls ≈ 15 s) outlives the 2 s poll ~7× — the
full-path + chunk-part windows the assertions observe are ≫ the poll
period (deterministic, not races).

KB fixture — a real git repo (host temp dir via ``tmp_path_factory``; the
app runs on the same host) with EXACTLY THREE markdown files:

* ``notes/a-rather-deeply-nested-folder/structure/for/the/sync/
  fixture-document.md`` — the LONG relative path (the clipping trigger the
  owner hit: a ``source/relative/path`` that overflows the pre-phase pill
  at desktop width), and a multi-chunk body (7 single-paragraph sections
  — one chunk apiece at the code-default 2000-char target, so the embed
  span is many 1.5 s token-budgeted batches: the wide in-flight window);
* ``notes/short-one.md`` + ``notes/short-two.md`` — short paths, one chunk
  each (the sync has several files → the ``(n/3)`` count part and the
  per-file chunk restart are exercised too).

The source name is the checkout's directory basename (the ``repo_name``
rule, phase 28): ``progressbox`` — so the in-flight label's file part is
``progressbox/notes/a-rather-deeply-nested-folder/structure/for/the/sync/
fixture-document.md``.

Autouse cleanup (the phase-96 pattern): before each test wait for no
running sync on the app, then truncate the shared registry + KB tables
(chunks/documents/query_log/steering_notes/kb_overview/git_sources/
folder_summaries) — each test runs its own sync of the env-fallback source;
after each test the same truncate. The checkout in ``BOR_SOURCES_DIR``
survives across tests on purpose (the sync's pull is a fast no-op after
the first clone).

Test → observable mapping (Playwright Mapping Rule — the rendered DOM is
the observable; every in-flight snapshot is ONE in-browser evaluation that
verifies the awaited label state and reads the numbers together, so they
can never straddle a 2 s poll tick — the single ``setSyncLabel`` writer
updates the box + the button's aria-label/title + ``#sync-result`` in one
JS task):

1. ``test_running_button_is_spinner_only`` — click **Sync sources**; poll
   until ``#sync-progress`` is visible (a run is in flight): ``#sync-btn``
   carries ``is-running`` AND the icon ``is-spinning`` (the spinner IS the
   button), ``#sync-label`` is NOT visible (the label parts are hidden —
   the button is the icon), the button's ``aria-label`` EQUALS the box's
   current text (written by the same call — never stale), the box is
   ``aria-hidden`` (presentational — ``#sync-result`` is the sole live
   region), the button is disabled + ``aria-busy`` (the pre-phase running
   mechanics), and the Cancel button is visible (the phase-134 contract
   unchanged).
2. ``test_box_carries_the_full_untruncated_label`` — during the
   in-flight window of the LONG-path file (poll until the box text is
   ``Syncing… <full long path> (1/3)`` — the path string IN FULL, the
   pre-phase ellipsis would have truncated it): the box's
   ``scrollWidth <= clientWidth`` AND the text's ``scrollWidth <=
   clientWidth`` (the wrap engaged — nothing is clipped), ``#sync-result``
   (the announcer) carries the SAME full string (A4 — one source, two
   surfaces), the button's ``aria-label`` equals it too, and the box ink
   on the panel background clears WCAG 2.1 AA (≥ 4.5:1). Then a retry loop
   over the slow embed window pins the phase-129 chunk part IN THE BOX
   (``Syncing… <full long path> (1/3) · chunk n/m`` — the part is only
   present while an embed batch is in flight; the proxy's 1.5 s delay
   makes the ~15 s window ≫ the 2 s poll).
3. ``test_mobile_box_full_width`` — a fresh page at the house mobile
   viewport (390×844 — the ``test_sync_label_mobile.py`` constant): the
   same in-flight assertions — the box is FULL WIDTH (its width ≈ the
   container's content width minus the container's padding — no override
   needed, the block rides the 72rem container), no horizontal clipping
   (``scrollWidth <= clientWidth``), the icon-only button keeps the
   mobile full-width rule, the label wraps (the full long path is a
   substring — the old mobile-stuffing ellipsis is gone).
4. ``test_settled_state_restores_the_pill`` — in flight first (the box is
   visible), then let the run settle (the house settle idiom): the box is
   HIDDEN (run-scoped only), the button shows the settle label
   ``Synced HH:MM`` (the pre-phase byte-identical settle copy) with the
   count + chunk parts hidden and empty, ``is-running`` / ``is-spinning``
   are gone, the ``aria-label`` is back to ``"Sync sources"``, and the
   counts line on ``#sync-result`` matches the pre-phase ``fmtSyncResult``
   shape (``N added``).
5. ``test_reattach_shows_the_box`` — in flight (the box visible), RELOAD
   the page (the admin session cookie persists): the load-time re-attach
   (``initSyncButton``) finds the RUNNING sync → the box is VISIBLE with
   the current label and the button is icon-only again (the never-stale
   contract — a page load mid-run shows the truth, not an idle pill).
"""
from __future__ import annotations

import json
import os
import re
import subprocess
import sys
import time
from collections.abc import Iterator
from pathlib import Path
from typing import Any

import httpx
import pytest
from playwright.sync_api import Page, ViewportSize, expect
from sqlalchemy import text

from app.config import Settings as _Settings
from app.db import SessionLocal
from e2e.auth_helpers import login
from e2e.conftest import (
    ADMIN_PASSWORD,
    MOCK_PORT,
    SESSION_SECRET,
    USE_REAL_LLM,
    _wait_http,
)

REPO = Path(__file__).resolve().parents[2]

# Phase 79 (task 04, full inventory): the conftest session app owns its
# port in a combined run — this module's app binds its own port
# instead (a same-port second uvicorn dies on bind and would drive the
# wrong server). Env-overridable.
APP_PORT = int(os.environ.get("E2E_APP_PORT_SYNCPROGRESS", "8166"))
APP_URL = f"http://127.0.0.1:{APP_PORT}"

#: The slow-LLM proxy's port (the conftest's mock LLM stays on
#: MOCK_PORT — the proxy fronts it).
SLOW_PORT = int(os.environ.get("E2E_SLOW_LLM_PORT_SYNCPROGRESS", "8908"))
SLOW_URL = f"http://127.0.0.1:{SLOW_PORT}"

#: Per-LLM-request delay on the proxy (the phase-64/98 rationale,
#: module docstring): the proxy delays EVERY call — embeds included —
#: so the in-flight windows outlive the UI's 2 s poll by ≫ 1×.
SLOW_DELAY_S = "1.5"

# --------------------------------------------------------------------------
# Fixture constants (deterministic, token-controlled)
# --------------------------------------------------------------------------

#: The fixture repo's directory name — the source name the import
#: records for it (the ``repo_name`` basename rule, phase 28).
SOURCE = "progressbox"

#: The LONG relative path — the clipping trigger the owner hit (a
#: ``source/relative/path`` that overflows the pre-phase pill even on
#: desktop) — and the two short-path siblings (the sync has several
#: files: the ``(n/3)`` count part + the per-file chunk restart).
LONG_REL = (
    "notes/a-rather-deeply-nested-folder/structure/for/the/sync/"
    "fixture-document.md"
)
SHORT_RELS = ("notes/short-one.md", "notes/short-two.md")
N_FILES = 3

#: The in-flight label's file part — the status endpoint's
#: ``current_file`` = ``<source>/<rel>`` (the sync's progress hook),
#: verbatim. The long file is the FIRST of the three (the sorted walk —
#: ``notes/a-…`` < ``notes/short-…``), so its count part is ``(1/3)``.
LONG_PATH = f"{SOURCE}/{LONG_REL}"

#: The long file's shape (module docstring): an H1 preamble (one chunk)
#: + 7 single-paragraph sections (~1500 chars each — one chunk apiece at
#: the code-default 2000-char target) → ~8 chunks → ~8 token-budgeted
#: embed batches × 1.5 s (the wide in-flight window the assertions
#: observe — deterministic, not a race).
N_LONG_SECTIONS = 7

SHORT_MD = (
    "# Short fixture {n}\n\nA tiny sync fixture note with a short path "
    "({n}).\n"
)

#: The house mobile viewport (the ``test_sync_label_mobile.py``
#: constant — the phone-width leg of the owner's ask).
MOBILE_VIEWPORT: ViewportSize = {"width": 390, "height": 844}

#: "Synced HH:MM" — the local-time last-result label (sources.js's
#: fmtSyncTime), any hour/minute (the test_sync_button.py pattern).
SYNCED_LABEL = re.compile(r"Synced \d{1,2}:\d{2}$")
#: The settled boot label — idle ("Sync sources") or the RE-ATTACHED
#: terminal from a previous test's sync (the module app is shared, its
#: in-memory sync state survives the test boundary, and the boot
#: re-attach renders the last result).
IDLE_OR_SETTLED = re.compile(r"^(Sync sources|Synced \d{1,2}:\d{2})$")

#: Generous settle budgets: a changed sync against the slowed LLM is
#: ≈ 30–35 s; the UI's 2 s poll settles at most one tick after the
#: terminal state lands.
SETTLE_TIMEOUT_MS = 120_000

#: The module's wall budget: five full 3-file syncs against the slowed
#: mock (≈ 30–35 s each) must stay well under this — a breach is a
#: regression canary for the sync pipeline's speed, not a flake
#: tolerance.
WALL_BUDGET_S = 600.0

# ---------------------------------------------------------------------------
# The in-browser snapshot functions (Playwright Mapping Rule — the
# rendered DOM is the observable).
#
# Each returns null while the awaited label state is absent, and an
# ATOMIC snapshot once it is present: the box's text + the button's
# aria-label / classes + the geometry — read in the SAME in-browser
# evaluation that verified the state, so the numbers can never straddle
# a 2 s poll tick (the single setSyncLabel writer updates the box, the
# button, and the announcer together in one JS task — A4: one source,
# two surfaces).
# ---------------------------------------------------------------------------

#: Tests 1 + 5 — ANY in-flight state (a run is in flight): the box is
#: VISIBLE with non-empty text, and the button is in its running state
#: (is-running + the spinning icon + the hidden label parts + the full
#: live label as aria-label + disabled + aria-busy) with the Cancel
#: control visible (the phase-134 contract). The box's aria-hidden
#: attribute is read too (presentational — the one-live-region pin).
_JS_RUNNING_SNAPSHOT = """
() => {
    const box = document.querySelector('#sync-progress');
    const textEl = document.querySelector('#sync-progress-text');
    const btn = document.querySelector('#sync-btn');
    const label = document.querySelector('#sync-label');
    const icon = btn ? btn.querySelector('.sync-icon') : null;
    const cancel = document.querySelector('#sync-cancel-btn');
    if (!box || !textEl || !btn || !label || !icon || !cancel) return null;
    if (box.hidden || !textEl.textContent) return null;
    const cs = (el) => getComputedStyle(el);
    return {
        boxText: textEl.textContent,
        ariaHidden: box.getAttribute('aria-hidden'),
        ariaLabel: btn.getAttribute('aria-label'),
        isRunning: btn.classList.contains('is-running'),
        isSpinning: icon.classList.contains('is-spinning'),
        labelHidden: cs(label).display === 'none',
        cancelVisible: !cancel.hidden && cs(cancel).display !== 'none',
        btnDisabled: btn.disabled,
        ariaBusy: btn.getAttribute('aria-busy'),
    };
}
"""

#: Test 2, wait A — the in-flight window of the LONG-path file: the box
#: is VISIBLE and its text is the FULL label for the long file —
#: "Syncing… <full long path> (1/3)" (+ the optional phase-129
#: " · chunk n/m" part while an embed batch is in flight — the
#: pre-first-batch frame has none) — the complete
# source/relative/path IN the text (the pre-phase ellipsis would have
#: truncated it). Returns the text + the wrap proof (scrollWidth vs
#: clientWidth of box and text) + the announcer + the aria-label + the
#: computed ink/panel colors (the AA pin).
_JS_LONG_FILE_SNAPSHOT = """
(longPath) => {
    const box = document.querySelector('#sync-progress');
    const textEl = document.querySelector('#sync-progress-text');
    const btn = document.querySelector('#sync-btn');
    const result = document.querySelector('#sync-result');
    if (!box || !textEl || !btn || !result || box.hidden) return null;
    const t = textEl.textContent;
    const prefix = 'Syncing… ' + longPath + ' (';
    if (!t.startsWith(prefix)) return null;
    const tail = t.slice(prefix.length);
    if (!/^\\d+\\/3\\)( · chunk \\d+\\/\\d+)?$/.test(tail)) return null;
    const cs = (el) => getComputedStyle(el);
    return {
        boxText: t,
        resultText: result.textContent,
        ariaLabel: btn.getAttribute('aria-label'),
        boxScrollWidth: box.scrollWidth,
        boxClientWidth: box.clientWidth,
        textScrollWidth: textEl.scrollWidth,
        textClientWidth: textEl.clientWidth,
        ink: cs(textEl).color,
        bg: cs(box).backgroundColor,
    };
}
"""

#: Test 2, wait B — the phase-129 chunk part IN THE BOX during the slow
#: embed window: the box text is "Syncing… <full long path> (1/3) ·
#: chunk n/m" (the part is only present while an in-flight file's embed
#: batch is in flight — the proxy's 1.5 s delay makes the ~15 s
#: long-file window ≫ the 2 s poll, so the retry loop resolves
#: deterministically).
_JS_LONG_FILE_CHUNK_SNAPSHOT = """
(longPath) => {
    const box = document.querySelector('#sync-progress');
    const textEl = document.querySelector('#sync-progress-text');
    if (!box || !textEl || box.hidden) return null;
    const t = textEl.textContent;
    if (!t.startsWith('Syncing… ' + longPath + ' (1/3) · chunk ')) return null;
    const m = t.match(/· chunk (\\d+)\\/(\\d+)$/);
    if (!m) return null;
    return {
        boxText: t,
        chunkN: Number(m[1]),
        chunkM: Number(m[2]),
        boxScrollWidth: box.scrollWidth,
        boxClientWidth: box.clientWidth,
        textScrollWidth: textEl.scrollWidth,
        textClientWidth: textEl.clientWidth,
    };
}
"""

#: Test 3 — the mobile (390×844) in-flight window of the long file: the
#: box text is the full label (the long path IN the text), the box is
#: FULL WIDTH (its bounding-box width ≈ the container's content width —
#: the block rides the 72rem container, no override), no horizontal
#: clipping (scrollWidth <= clientWidth), the icon-only button keeps the
#: mobile full-width rule (its width:100% resolves against its
#: containing block — the .page-head card's content width, NOT the
#: container's), and the label parts are hidden.
_JS_MOBILE_SNAPSHOT = """
(longPath) => {
    const box = document.querySelector('#sync-progress');
    const textEl = document.querySelector('#sync-progress-text');
    const btn = document.querySelector('#sync-btn');
    const label = document.querySelector('#sync-label');
    const icon = btn ? btn.querySelector('.sync-icon') : null;
    const container = document.querySelector('#view-rag .container');
    const head = document.querySelector('#view-rag .page-head');
    if (!box || !textEl || !btn || !label || !icon || !container || !head) return null;
    if (box.hidden) return null;
    const t = textEl.textContent;
    if (!t.startsWith('Syncing… ' + longPath + ' (')) return null;
    const cs = (el) => getComputedStyle(el);
    const rect = (el) => {
        const b = el.getBoundingClientRect();
        return { left: b.left, right: b.right, width: b.width, height: b.height };
    };
    // clientWidth already excludes the border — content = clientWidth
    // minus the horizontal padding only.
    const contentWidth = (el) => {
        const s = cs(el);
        return el.clientWidth - parseFloat(s.paddingLeft) - parseFloat(s.paddingRight);
    };
    return {
        boxText: t,
        isRunning: btn.classList.contains('is-running'),
        isSpinning: icon.classList.contains('is-spinning'),
        labelHidden: cs(label).display === 'none',
        boxRect: rect(box),
        btnRect: rect(btn),
        containerContentWidth: contentWidth(container),
        headContentWidth: contentWidth(head),
        boxScrollWidth: box.scrollWidth,
        boxClientWidth: box.clientWidth,
    };
}
"""


# ---------------------------------------------------------------------------
# Fixtures
# ---------------------------------------------------------------------------


def _git(cwd: Path, *args: str) -> None:
    """Run git in *cwd*; a non-zero exit fails the fixture loudly."""
    proc = subprocess.run(["git", *args], cwd=cwd, capture_output=True, text=True)
    if proc.returncode != 0:
        raise AssertionError(f"git {' '.join(args)} failed: {proc.stderr.strip()}")


def _commit(repo: Path, message: str) -> None:
    """A fixture commit (never signed — the house ``commit.gpgsign``
    flag, the test_sync_button.py idiom)."""
    _git(
        repo,
        "-c", "user.email=e@x", "-c", "user.name=t",
        "-c", "commit.gpgsign=false",
        "commit", "-qm", message,
    )


def _long_md() -> str:
    """The long-path fixture doc (see the module docstring): an H1
    preamble (one chunk) + 7 single-paragraph sections (~1500 chars
    each — one chunk apiece at the code-default 2000-char target — no
    randomness, so the chunk count is stable run to run)."""
    parts = ["# Long-path fixture document\n"]
    parts.append(
        "Preamble for the deeply nested sync fixture document. This "
        "paragraph sits under the H1 and chunks on its own.\n"
    )
    for k in range(1, N_LONG_SECTIONS + 1):
        parts.append(f"\n## Nesting {k}\n")
        base = (
            f"Nesting paragraph {k} of the long-path sync fixture carries "
            "deterministic filler for the chunker: alpha bravo charlie "
            "delta echo foxtrot golf hotel india juliet kilo lima "
        )
        parts.append((base * 30)[:1500] + "\n\n")
    return "".join(parts)


@pytest.fixture(scope="module")
def progressbox_repo(tmp_path_factory: pytest.TempPathFactory) -> Path:
    """A real git repo the sync must clone over ``file://`` (see the
    module docstring): the long-path doc (the clipping trigger,
    multi-chunk body) + the two short-path docs. The directory NAME is
    the source name (the ``repo_name`` basename rule, phase 28).
    ``tmp_path`` is function-scoped while the module-scoped app fixture
    needs the repo for the module's lifetime, so it is built under
    ``tmp_path_factory`` via real ``git`` subprocess calls."""
    root = tmp_path_factory.mktemp("bor_syncprogress")
    repo = root / SOURCE
    long_file = repo / LONG_REL
    long_file.parent.mkdir(parents=True)
    long_file.write_text(_long_md(), encoding="utf-8")
    for i, rel in enumerate(SHORT_RELS, start=1):
        (repo / rel).write_text(SHORT_MD.format(n=i), encoding="utf-8")
    _git(repo, "init", "-q")
    _git(repo, "add", "-A")
    _commit(repo, "one")
    assert (repo / ".git").is_dir()
    assert len(list(repo.rglob("*.md"))) == N_FILES
    return repo


def _app_env(
    base_url: str, repo: Path, tmp_path_factory: pytest.TempPathFactory
) -> dict[str, str]:
    """The per-module app env (the conftest pattern): the ``base_url``
    picks the LLM (the slow proxy in front of the mock), and
    ``BOR_GIT_SOURCES`` is this suite's own ``file://`` fixture repo —
    the sync's SUBJECT (the env fallback: the autouse truncate keeps the
    ``git_sources`` table empty, so the env list is the effective
    source). The env var also overrides the owner's local (gitignored)
    ``.env`` ``BOR_GIT_SOURCES`` (process env ranks above the file)."""
    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 base_url
    )
    # Mock-calibrated gate values (conftest pattern): the floors ride
    # WITH the 0.30 threshold — a floor above it fails the startup
    # validator (phase 112/113).
    env["BOR_RELEVANCE_THRESHOLD"] = "0.30"
    env["BOR_LEXICAL_SUPPORT_FLOOR"] = "0.15"
    env["BOR_SOURCE_USEFULNESS_FLOOR"] = "0.15"
    # Phase 67: instant retry waits + the code-default budget (the
    # conftest leak-guard pattern).
    env["BOR_LLM_RETRY_DELAY"] = "0"
    env["BOR_LLM_RETRIES"] = str(_Settings.model_fields["llm_retries"].default)
    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 sync's subject (see the docstring) + its own checkout dir.
    env["BOR_GIT_SOURCES"] = f"file://{repo}"
    env["BOR_SOURCES_DIR"] = str(tmp_path_factory.mktemp("bor_checkouts"))
    # Leak guards (conftest pattern): an operator's local (gitignored)
    # .env cannot leak corpus-specific settings into the app under test.
    env["BOR_DOCS_REPO"] = ""
    env["BOR_SUGGESTIONS"] = json.dumps(
        _Settings.model_fields["suggestions"].default
    )
    env["BOR_INPUT_PLACEHOLDER"] = _Settings.model_fields["input_placeholder"].default
    env["BOR_FOOTER_TEXT"] = _Settings.model_fields["footer_text"].default
    return env


@pytest.fixture(scope="module")
def slow_llm(mock_llm: int) -> Iterator[int]:
    """The delay-injecting reverse proxy in front of the mock LLM
    (``tests/e2e/slow_llm.py``) — the timing fixture for the whole
    suite: it delays EVERY LLM call, embeds included, so the in-flight
    windows outlive the UI's 2 s poll by ≫ 1× (see ``SLOW_DELAY_S``)."""
    env = dict(os.environ)
    env.pop("DEBUGPY", None)
    env["SLOW_LLM_DELAY_S"] = SLOW_DELAY_S
    env["E2E_MOCK_PORT"] = str(MOCK_PORT)
    proc = subprocess.Popen(
        [sys.executable, "-m", "uvicorn", "tests.e2e.slow_llm:app",
         "--host", "127.0.0.1", "--port", str(SLOW_PORT), "--log-level", "warning"],
        cwd=REPO,
        env=env,
    )
    try:
        _wait_http(f"{SLOW_URL}/v1/models")
        yield SLOW_PORT
    finally:
        proc.terminate()
        try:
            proc.wait(timeout=10)
        except subprocess.TimeoutExpired:
            proc.kill()


@pytest.fixture(scope="module")
def app_server(
    mock_llm: int,
    slow_llm: int,
    progressbox_repo: Path,
    tmp_path_factory: pytest.TempPathFactory,
) -> Iterator[str]:
    """The SLOW app under test (the whole suite):
    ``BOR_LLM_BASE_URL`` is the slow proxy in front of the mock (the
    timing fixture), ``BOR_GIT_SOURCES`` the ``file://`` fixture repo
    (the sync's subject — see the module docstring), and
    ``BOR_SOURCES_DIR`` its own checkout dir. ``progressbox_repo`` is a
    dependency only for the fixture ordering (the repo exists before
    the first test)."""
    env = _app_env(f"{SLOW_URL}/v1", progressbox_repo, tmp_path_factory)
    proc = subprocess.Popen(
        [sys.executable, "-m", "uvicorn", "app.main:app",
         "--host", "127.0.0.1", "--port", str(APP_PORT), "--log-level", "warning"],
        cwd=REPO,
        env=env,
    )
    try:
        _wait_http(f"{APP_URL}/api/health")
        yield APP_URL
    finally:
        proc.terminate()
        try:
            proc.wait(timeout=10)
        except subprocess.TimeoutExpired:
            proc.kill()


@pytest.fixture(scope="module")
def app_url(app_server: str) -> str:
    return app_server


def _truncate_all() -> None:
    """Fresh registry + KB (the E2E isolation pattern): the E2E suites
    share one Postgres, so a leftover git_sources row would outrank
    this app's ``BOR_GIT_SOURCES`` env fallback (effective_sources
    prefers the DB list) and a leftover document would pollute the file
    counts the sync pins exactly."""
    with SessionLocal() as db:
        db.execute(
            text(
                "TRUNCATE chunks, documents, query_log, steering_notes, "
                "kb_overview, git_sources, folder_summaries"
            )
        )
        db.commit()  # without the commit the TRUNCATE rolls back (the house pattern)


def _wait_sync_settled(base_url: str, timeout_s: float = 180.0) -> None:
    """No background sync may leak across tests (the run state lives in
    the app's memory, and a still-running run would keep importing into
    the NEXT test's truncated KB): wait for the app's sync status to be
    non-running BEFORE the truncate. Own admin session (the endpoint is
    admin-only) — usually a no-op: the tests settle only after their
    run's terminal state."""
    with httpx.Client(base_url=base_url, timeout=5.0) as client:
        r = client.post("/api/login", json={"password": ADMIN_PASSWORD})
        assert r.status_code == 204, r.text
        deadline = time.monotonic() + timeout_s
        while time.monotonic() < deadline:
            r = client.get("/api/sync/status")
            assert r.status_code == 200, r.text
            body = r.json()
            if body["state"] != "running":
                # Diagnostic: a FAILED terminal between tests is a real
                # anomaly for this suite (every sync here must succeed)
                # — surface the error text so a cascade is diagnosable
                # instead of showing up as a later test's blocked click.
                if body["state"] == "failed":
                    print(f"sync progress box: terminal FAILED at a test "
                          f"boundary: {body.get('error')}")
                return
            time.sleep(0.2)
    raise AssertionError(f"a sync was still running at a test boundary: {base_url}")


@pytest.fixture(autouse=True)
def _clean(app_url: str, db_ready: None) -> Iterator[None]:
    """Per-test KB isolation (the phase-96 cleanup pattern): the
    shared E2E Postgres is truncated before AND after every test —
    each test runs its own sync of the env-fallback source, so the KB
    rows of the fixture source are exactly this test's own. The
    pre-wait keeps a leaked run from the previous test (the module app
    is shared) from racing the truncate."""
    _wait_sync_settled(app_url)
    _truncate_all()
    yield
    _truncate_all()


@pytest.fixture(scope="module", autouse=True)
def _wall_time_guard() -> Iterator[None]:
    """The module's wall-time budget: the whole module — app boot +
    FIVE full 3-file syncs against the slowed mock LLM — stays under
    ``WALL_BUDGET_S``. A breach is a regression canary (the sync
    pipeline slowed), not a flake tolerance."""
    start = time.monotonic()
    yield
    elapsed = time.monotonic() - start
    assert elapsed < WALL_BUDGET_S, (
        f"the module took {elapsed:.1f}s — over the {WALL_BUDGET_S:.0f}s wall "
        "budget (the five small-fixture syncs against the slowed mock LLM "
        "are a speed canary)"
    )


# ---------------------------------------------------------------------------
# Helpers
# ---------------------------------------------------------------------------


def _parse_rgb(css_color: str) -> tuple[int, int, int]:
    """A computed ``rgb(r, g, b)`` / ``rgba(…)`` color → the triple."""
    m = re.match(r"rgba?\(\s*(\d+)[,\s]+(\d+)[,\s]+(\d+)", css_color)
    assert m, f"not a computable rgb color: {css_color!r}"
    return (int(m.group(1)), int(m.group(2)), int(m.group(3)))


def _wcag_luminance(rgb: tuple[float, float, float]) -> float:
    """WCAG 2.1 relative luminance (the same math the theme suites run)."""

    def lin(c: float) -> float:
        s = c / 255
        return s / 12.92 if s <= 0.03928 else ((s + 0.055) / 1.055) ** 2.4

    r, g, b = rgb
    return 0.2126 * lin(r) + 0.7152 * lin(g) + 0.0722 * lin(b)


def _wcag_ratio(
    fg: tuple[float, float, float], bg: tuple[float, float, float]
) -> float:
    l1, l2 = _wcag_luminance(fg), _wcag_luminance(bg)
    return (max(l1, l2) + 0.05) / (min(l1, l2) + 0.05)


def _start_sync(page: Page, app_url: str) -> None:
    """Land the admin on the sources view, wait for the settled boot
    label (the phase-64 pattern — also proves the boot re-attach
    completed, so no later re-render can clobber the running state),
    then click **Sync sources** and pin the click's immediate 202
    state: the button IS the spinner (disabled, aria-busy,
    ``is-running``, the spinning icon) and the box is revealed with
    the bare label (phase 149 — the running detail rides the box)."""
    login(page, app_url)  # lands on /sources.html (the button's home)

    btn = page.locator("#sync-btn")
    expect(btn).to_be_visible(timeout=30_000)
    expect(page.locator("#sync-label")).to_have_text(IDLE_OR_SETTLED)
    expect(btn).to_be_enabled()

    btn.click()

    # The click's immediate state (the 202 moment — the bare prefix,
    # the phase-64 contract): the button is the loading spinner and
    # the box carries the bare label (phase 149 — one source, two
    # surfaces).
    expect(btn).to_be_disabled()
    expect(btn).to_have_attribute("aria-busy", "true")
    expect(btn).to_have_class(re.compile(r"\bis-running\b"))
    expect(btn.locator(".sync-icon")).to_have_class(re.compile(r"\bis-spinning\b"))
    expect(page.locator("#sync-progress")).to_have_text("Syncing…")
    expect(page.locator("#sync-error-banner")).to_be_hidden()


def _wait_terminal(app_url: str, timeout_s: float = 120.0) -> dict[str, Any]:
    """Poll the status endpoint (own admin session) until the UI-
    triggered run reaches a terminal state — the source of truth for
    the settle (the UI's 2 s poll renders it at most one tick later).
    A FAILED terminal fails the test with the terminal body (the
    ``detail`` / error text) — never an opaque label timeout."""
    with httpx.Client(base_url=app_url, timeout=30.0) as client:
        r = client.post("/api/login", json={"password": ADMIN_PASSWORD})
        assert r.status_code == 204, r.text
        deadline = time.monotonic() + timeout_s
        body: dict[str, Any] = {}
        while time.monotonic() < deadline:
            r = client.get("/api/sync/status")
            assert r.status_code == 200, r.text
            body = r.json()
            if body["state"] in ("success", "failed"):
                assert body["state"] == "success", body
                return body
            time.sleep(0.2)
    raise AssertionError(f"the run never reached a terminal state: {body}")


def _settle(page: Page, app_url: str) -> None:
    """The phase-32 success settle (the pre-phase byte-identical
    terminal shape, preserved by phase 149): the run reaches a SUCCESS
    terminal at the endpoint, then the button shows "Synced HH:MM"
    (re-enabled, the running state fully cleared), the box is HIDDEN
    (run-scoped), and no error."""
    _wait_terminal(app_url)
    expect(page.locator("#sync-label")).to_have_text(SYNCED_LABEL, timeout=30_000)
    btn = page.locator("#sync-btn")
    expect(btn).to_be_enabled()
    expect(btn).not_to_have_attribute("aria-busy")
    expect(btn).not_to_have_class(re.compile(r"\bis-running\b"))
    expect(btn.locator(".sync-icon")).not_to_have_class(re.compile(r"\bis-spinning\b"))
    expect(page.locator("#sync-progress")).to_be_hidden()
    expect(page.locator("#sync-error-banner")).to_be_hidden()


# ---------------------------------------------------------------------------
# 1. In flight: the button is the spinner only (no visible label text),
#    the aria-label carries the full live label, and Cancel rides along
# ---------------------------------------------------------------------------


def test_running_button_is_spinner_only(
    page: Page, app_url: str, db_ready: None
) -> None:
    """A fresh admin page on ``/sources.html`` clicks **Sync sources**;
    the page's OWN 2 s poll keeps the run in flight long enough (the
    slow proxy guarantees a ≫ 30 s window) that the in-flight state is
    observed deterministically: ``#sync-btn`` carries ``is-running``
    AND the icon ``is-spinning`` (the spinner IS the loading state),
    ``#sync-label`` is NOT visible (the label parts are hidden — the
    button is the icon, at any viewport), the button's ``aria-label``
    EQUALS the box's current text (written by the same single-writer
    call — never stale), the box is ``aria-hidden`` (presentational —
    ``#sync-result`` remains the sole aria-live region), the running
    mechanics (disabled + ``aria-busy``) are preserved, and the Cancel
    button is visible (the phase-134 contract unchanged)."""
    _start_sync(page, app_url)

    # Poll (the generous timeout — the slow proxy's in-flight window is
    # ~30 s) until a run is in flight: the box is visible + non-empty.
    snap = page.wait_for_function(
        _JS_RUNNING_SNAPSHOT, timeout=SETTLE_TIMEOUT_MS
    ).json_value()

    # The button is the loading spinner (the owner's ask).
    assert snap["isRunning"] is True, snap
    assert snap["isSpinning"] is True, snap
    # The label parts are hidden — the button is the icon (no visible
    # label text).
    assert snap["labelHidden"] is True, snap
    assert page.locator("#sync-label").is_visible() is False
    assert page.locator("#sync-progress").is_visible()
    # The icon-only button keeps an accessible name — the FULL live
    # label, equal to the box's text (the same writer call wrote both).
    assert snap["boxText"], snap
    assert snap["ariaLabel"] == snap["boxText"], snap
    # The box is presentational — #sync-result is the SOLE live region.
    assert snap["ariaHidden"] == "true", snap
    # The pre-phase running mechanics are preserved.
    assert snap["btnDisabled"] is True, snap
    assert snap["ariaBusy"] == "true", snap
    # The Cancel control rides the running state (phase 134 — unchanged).
    assert snap["cancelVisible"] is True, snap

    _settle(page, app_url)


# ---------------------------------------------------------------------------
# 2. The box carries the EXACT full untruncated label (long path
#    included — wrapped, never clipped), the announcer carries the same
#    string, and the chunk part appears in the box during the embed
#    window
# ---------------------------------------------------------------------------


def test_box_carries_the_full_untruncated_label(
    page: Page, app_url: str, db_ready: None
) -> None:
    """During the in-flight window of the LONG-path file (poll until the
    box text is ``Syncing… <full long path> (1/3)`` — the slow proxy
    makes the ~15 s span deterministic): the box's text contains the
    FULL label shape with the complete ``source/relative/path`` as a
    substring (the pre-phase ellipsis would have truncated it — the
    wrap means nothing is dropped), the box did NOT overflow
    horizontally (``scrollWidth <= clientWidth`` — the wrap engaged)
    neither as a panel nor in its text, ``#sync-result`` (the
    announcer) carries the SAME full string (A4 — one source, two
    surfaces), the button's ``aria-label`` equals it, and the box ink
    on the panel clears WCAG 2.1 AA (≥ 4.5:1). Then a retry loop over
    the slow embed window pins the phase-129 chunk part IN THE BOX
    (``· chunk n/m`` — the part is only present while an embed batch is
    in flight; the proxy's 1.5 s delay makes the window wide)."""
    _start_sync(page, app_url)

    # Wait A — the in-flight window of the long file (any frame of it —
    # the pre-first-batch frame has no chunk part yet).
    snap = page.wait_for_function(
        _JS_LONG_FILE_SNAPSHOT, arg=LONG_PATH, timeout=SETTLE_TIMEOUT_MS
    ).json_value()

    # The FULL label shape — the complete long path IN the text (the
    # clipping trigger, untruncated).
    expected_prefix = f"Syncing… {LONG_PATH} (1/3)"
    assert snap["boxText"].startswith(expected_prefix), snap["boxText"]
    assert LONG_PATH in snap["boxText"], snap["boxText"]
    # The wrap engaged — no horizontal overflow (the whole point of the
    # box: the path WRAPS, it is never clipped).
    assert snap["boxScrollWidth"] <= snap["boxClientWidth"], snap
    assert snap["textScrollWidth"] <= snap["textClientWidth"], snap
    # A4: the announcer carries the SAME full string (one source, two
    # surfaces) — and the icon-only button's accessible name does too.
    assert snap["resultText"] == snap["boxText"], snap
    assert snap["ariaLabel"] == snap["boxText"], snap
    # WCAG 2.1 AA: the box's mono ink on its panel background ≥ 4.5:1
    # (the theme's pinned identity pair — observed, not assumed).
    ratio = _wcag_ratio(_parse_rgb(snap["ink"]), _parse_rgb(snap["bg"]))
    assert ratio >= 4.5, f"{snap['ink']} on {snap['bg']}: {ratio:.2f}:1 < 4.5:1 (AA)"

    # Wait B — the phase-129 chunk part IN THE BOX (the retry loop over
    # the slow embed window — the part is only present while an embed
    # batch is in flight; the ~15 s long-file window outlives the 2 s
    # poll by ≫ 1×).
    snap2 = page.wait_for_function(
        _JS_LONG_FILE_CHUNK_SNAPSHOT, arg=LONG_PATH, timeout=SETTLE_TIMEOUT_MS
    ).json_value()
    assert snap2["chunkN"] >= 1, snap2
    assert snap2["chunkM"] >= 1, snap2
    # The longer label (with the chunk part) still wraps — no clipping.
    assert snap2["boxScrollWidth"] <= snap2["boxClientWidth"], snap2
    assert snap2["textScrollWidth"] <= snap2["textClientWidth"], snap2

    _settle(page, app_url)


# ---------------------------------------------------------------------------
# 3. Mobile (390×844): the box is full-width, the button is icon-only
#    (the mobile full-width rule), and the label wraps
# ---------------------------------------------------------------------------


def test_mobile_box_full_width(page: Page, app_url: str, db_ready: None) -> None:
    """A fresh page at the house MOBILE viewport (390×844) clicks
    **Sync sources**; during the in-flight window of the LONG-path
    file: the box is FULL WIDTH (its bounding-box width ≈ the
    container's content width — a block in the same container rides
    the 100%-width phone shell with no override), no horizontal
    clipping (``scrollWidth <= clientWidth``), the button is
    icon-only (``is-running`` + the spinning icon, the label parts
    hidden) and keeps the mobile full-width rule, and the label wraps
    (the full long path is a substring — the old mobile-stuffing
    ellipsis is gone)."""
    page.set_viewport_size(MOBILE_VIEWPORT)
    _start_sync(page, app_url)

    snap = page.wait_for_function(
        _JS_MOBILE_SNAPSHOT, arg=LONG_PATH, timeout=SETTLE_TIMEOUT_MS
    ).json_value()

    # The label wraps — the full long path is IN the box's text (the
    # old mobile-stuffing ellipsis is gone).
    assert snap["boxText"].startswith(f"Syncing… {LONG_PATH} ("), snap["boxText"]
    assert LONG_PATH in snap["boxText"], snap["boxText"]
    # The wrap engaged — no horizontal clipping at the phone width.
    assert snap["boxScrollWidth"] <= snap["boxClientWidth"], snap
    # The box is FULL WIDTH: its width ≈ the container's content width
    # (the 2px slack covers float rounding only).
    assert abs(snap["boxRect"]["width"] - snap["containerContentWidth"]) <= 2, snap
    # …and it is inside the viewport (the owner's clipped-detail
    # report, directly negated).
    assert snap["boxRect"]["left"] >= 0, snap["boxRect"]
    assert snap["boxRect"]["right"] <= MOBILE_VIEWPORT["width"] + 1, snap["boxRect"]
    # The button is icon-only (the spinner) and keeps the mobile
    # full-width rule: its width:100% resolves against its containing
    # block — the .page-head card's content width (the card's padding
    # + border inset it from the container).
    assert snap["isRunning"] is True, snap
    assert snap["isSpinning"] is True, snap
    assert snap["labelHidden"] is True, snap
    assert abs(snap["btnRect"]["width"] - snap["headContentWidth"]) <= 2, snap

    _settle(page, app_url)


# ---------------------------------------------------------------------------
# 4. Settle: the box hides, the "Synced HH:MM" pill is restored
#    byte-identical, and the counts line is unchanged
# ---------------------------------------------------------------------------


def test_settled_state_restores_the_pill(
    page: Page, app_url: str, db_ready: None
) -> None:
    """A sync is triggered and observed IN FLIGHT (the box is visible —
    so the settle below is a real transition, not a no-op), then let to
    settle (the house settle idiom): the box is HIDDEN (run-scoped
    only), the button shows the settle label ``Synced HH:MM`` (the
    pre-phase byte-identical settle copy) with the label parts
    restored — the count + chunk parts hidden and empty —,
    ``is-running`` / ``is-spinning`` are gone, the ``aria-label`` is
    back to ``"Sync sources"`` (the ship value), and the counts line
    on ``#sync-result`` matches the pre-phase ``fmtSyncResult`` shape
    (``N added`` — the 3 fixture files)."""
    _start_sync(page, app_url)

    # In flight first (the box is visible + the button is the spinner).
    page.wait_for_function(_JS_RUNNING_SNAPSHOT, timeout=SETTLE_TIMEOUT_MS).json_value()

    # The settle (the house idiom): the terminal lands at the endpoint,
    # then the UI's 2 s poll renders the phase-32 terminal shape.
    _wait_terminal(app_url)
    expect(page.locator("#sync-label")).to_have_text(SYNCED_LABEL, timeout=30_000)

    # The pill is restored (the pre-phase byte-identical settle copy).
    expect(page.locator("#sync-label")).to_be_visible()
    btn = page.locator("#sync-btn")
    expect(btn).to_be_enabled()
    expect(btn).not_to_have_attribute("aria-busy")
    expect(btn).to_have_attribute("aria-label", "Sync sources")
    expect(btn).not_to_have_class(re.compile(r"\bis-running\b"))
    expect(btn.locator(".sync-icon")).not_to_have_class(re.compile(r"\bis-spinning\b"))
    # The box is run-scoped: hidden at settle.
    expect(page.locator("#sync-progress")).to_be_hidden()
    # The label parts are restored — the settle label carries no count
    # or chunk part.
    count = page.locator(".sync-label-count")
    expect(count).to_be_hidden()
    expect(count).to_have_text("")
    chunk = page.locator(".sync-label-chunk")
    expect(chunk).to_be_hidden()
    expect(chunk).to_have_text("")
    # The counts line is unchanged (the pre-phase fmtSyncResult pin).
    expect(page.locator("#sync-result")).to_have_text(f"{N_FILES} added")
    expect(page.locator("#sync-error-banner")).to_be_hidden()


# ---------------------------------------------------------------------------
# 5. A page load mid-run re-attaches with the box visible (never stale)
# ---------------------------------------------------------------------------


def test_reattach_shows_the_box(page: Page, app_url: str, db_ready: None) -> None:
    """A sync is triggered and observed IN FLIGHT (the box is visible),
    then the page is RELOADED (the admin session cookie persists): the
    load-time re-attach (``initSyncButton``) finds the RUNNING sync →
    the box is VISIBLE with the current label and the button is
    icon-only again (``is-running`` + the spinning icon, the label
    parts hidden, the ``aria-label`` equal to the box's text, the
    Cancel control visible) — the never-stale contract: a page load
    mid-run shows the truth, not an idle pill."""
    _start_sync(page, app_url)

    # In flight first (the pre-reload state: the box is visible).
    page.wait_for_function(_JS_RUNNING_SNAPSHOT, timeout=SETTLE_TIMEOUT_MS).json_value()

    # Reload mid-run — the run keeps going on the server; the admin
    # session (the cookie) survives the reload.
    page.reload(wait_until="domcontentloaded")

    # The re-attach: the box is visible with the CURRENT label and the
    # button is the spinner again (any in-flight state — the run is
    # ~30 s and the reload lands within a few seconds of the start).
    snap = page.wait_for_function(
        _JS_RUNNING_SNAPSHOT, timeout=SETTLE_TIMEOUT_MS
    ).json_value()
    assert snap["isRunning"] is True, snap
    assert snap["isSpinning"] is True, snap
    assert snap["labelHidden"] is True, snap
    assert snap["boxText"], snap
    assert snap["ariaLabel"] == snap["boxText"], snap
    assert snap["cancelVisible"] is True, snap

    _settle(page, app_url)