"""Phase 86 E2E (Playwright): the table pages are viewport-width — the
owner's "the page is the width of the table" bug, pinned.
Source: ``TODO.md`` L4 (owner bug report, 2026-09-07): "The history page
appears to be the width of the table despite the table being scrollable.
On mobile this results in half the page being blank and awkwardly
scrollable."
Story: n/a (owner bug report — ``TODO.md`` L4 is the story; no
``.agents/user_stories/`` file).
Run in isolation (DB must be up: ``podman compose up -d db``):
uv run pytest tests/e2e/test_history_page_width.py -v --no-cov
The bug (confirmed by live reproduction, 375×812, signed-in admin):
the History and Tokens tables carry ``position: absolute``
``.visually-hidden`` elements (the ``
`` + the Actions column
header span). No ancestor in the chain was positioned, so their 1px
boxes were laid out against the INITIAL containing block — the Actions
span at the 640px table's right edge — and leaked into the DOCUMENT's
scrollable overflow, bypassing the card's own ``overflow-x: auto``
scrolling. Measured ``documentElement.scrollWidth`` at 375px:
``/history.html`` **626**, ``/tokens.html`` **618** (the byte-identical
defect, folded in per owner decision A3), ``/sources.html`` **375**
(clean — the RAG table's headers are visible text). The fix is ONE CSS
property (phase 86, task 01): ``position: relative`` on the shared
``.table-wrap`` card, which becomes the containing block for those
spans. This suite pins the FIXED contract: the document never pans
past the viewport, the 640px tables still scroll INSIDE their cards
(the phase-07 / AGENTS.md-rule-5 full-width contract), the a11y spans
stay in the DOM, and the RAG view + the desktop layout are unchanged.
Counterfactual (verified red→green in the phase session): with the
task-01 rule reverted (no ``position: relative`` on ``.table-wrap``),
``test_history_page_is_viewport_width`` fails at its
``scrollWidth <= innerWidth`` pin (626 > 375); restored, it is green.
Test → contract mapping (Playwright Mapping Rule):
1. ``test_history_page_is_viewport_width`` — THE TODO.md L4 pin: admin
at 375×812, direct ``/history.html`` —
``documentElement.scrollWidth <= innerWidth`` (626 → ≤375), the
table still scrolls INSIDE ``#history-table-wrap``
(``scrollWidth > clientWidth`` — the 640px ``min-width``), the card
is the container's content width (≤ the viewport), and the
visually-hidden Actions header + caption survived the fix (the a11y
names are intact).
2. ``test_tokens_page_is_viewport_width`` — the folded-in identical
defect (618 pre-fix): the same two pins on direct ``/tokens.html``
for ``#tokens-table-wrap``.
3. ``test_spa_switch_keeps_page_width`` — the invariant survives the
router, not just direct loads: at 375px on ``/history.html`` the
mobile menu (a REAL ``#nav-toggle`` click — the phase-85 contract)
opens, the Tokens link switches the view IN THE SAME DOCUMENT
(the phase-76 window-sentinel proof — a real navigation would wipe
``window`` globals), and the Tokens page is still viewport-width
with its table still scrolling in-card.
4. ``test_rag_view_regression`` — ``/sources.html`` at 375px: no
overflow (it was already clean at 375 pre-fix — pinned so the
shared-rule change cannot regress it) AND the RAG table is still
full-width inside its card (in-card scroll — the 640px
``min-width`` still engages).
5. ``test_desktop_unchanged`` — 1280×800 regression on ALL THREE
views: no document overflow at any width, and the tables render at
container width (the cards need no internal scroll —
``scrollWidth <= clientWidth + 1``, the card is far wider than the
mobile 346px).
House conventions: a fresh 375×812 page per mobile test via the
session ``browser`` fixture (the conftest ``page`` is 1280×800),
``e2e.auth_helpers.login`` for the real form login, ``_wait_settled_admin``
copied from ``test_mobile_hamburger_nav.py`` (suites stay
self-contained — helpers copied, only ``auth_helpers`` imported).
Data: the History/Tokens contract holds with EMPTY tables (static
thead + the 640px ``min-width`` — no rows needed, and the suite never
depends on saved chats or issued tokens — test 5 ENFORCES that state,
resetting ``saved_chats`` (TRUNCATE, the house precedent) and the
``e2e-``-labeled tokens (label-scoped delete) before pinning, because
other suites leave chats on the shared dev DB and a single History
row pushes the table's auto-layout minimum width past the 1110px
desktop card). The RAG view, by contrast,
HIDES its table and shows the "Nothing indexed yet" empty state on an
empty KB, so tests 4 and 5 seed the KB the house way (truncate +
import the 13 fixture docs, deterministic mock embeddings) and pin
nothing about the rows themselves — only that the table is up and
full-width.
"""
from __future__ import annotations
import asyncio
from pathlib import Path
from threading import Thread
from typing import Any
from playwright.sync_api import Browser, JSHandle, Page, ViewportSize, 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"
MOBILE: ViewportSize = {"width": 375, "height": 812} # the bug-report phone viewport
DESKTOP: ViewportSize = {"width": 1280, "height": 800} # the conftest page size
async def _import_fixtures(mock_port: int) -> ImportSummary:
"""The house KB seed (copied from ``test_api_tokens.py``): import
the fixture docs against the mock LLM's embeddings."""
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's sync API keeps an asyncio loop running on the test
thread, so ``asyncio.run`` cannot be called directly from a test
body.
"""
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:
"""House reset (truncate the KB + the prompt-shaping tables so the
mock answers stay deterministic) + the fixture re-import — the
RAG view hides its table on an empty KB, so the two tests that
pin it need at least the fixture docs present. ``saved_chats`` and
``api_tokens`` are deliberately NOT touched (the house pattern)."""
with SessionLocal() as db:
db.execute(text("TRUNCATE chunks, documents, query_log, steering_notes"))
db.commit()
return _run_in_thread(_import_fixtures(mock_port))
def _mobile_page(browser: Browser) -> Page:
"""A fresh 375×812 page (the conftest ``page`` is 1280×800)."""
return browser.new_page(viewport=MOBILE)
def _wait_settled_admin(page: Page) -> None:
"""Wait until whoami has resolved for the admin: the whoami reveal
has un-hidden the admin-only nav links — the viewport-independent
settled signal (copied from ``test_mobile_hamburger_nav.py``; the
sign-out control is viewport-dependent and not a cross-viewport
probe)."""
page.wait_for_function(
"() => !document.querySelector('#nav-sources').hasAttribute('hidden')",
timeout=10_000,
)
def _open_menu(page: Page) -> None:
"""Open the mobile menu with a REAL click on the toggle (the
phase-85 contract: the gate sits below the header, so a real tap
reaches the toggle; for a signed-in admin the gate is hidden
anyway)."""
page.click("#nav-toggle")
expect(page.locator("#nav-toggle")).to_have_attribute("aria-expanded", "true")
def _wrap_handle(page: Page, view: str) -> JSHandle:
"""The table card (``.table-wrap``) of ``view``: the id'd cards for
history/tokens, the RAG card found through its ``.docs-table``
(that card is the only one without an id)."""
if view == "sources":
return page.evaluate_handle(
"() => document.querySelector('.docs-table').closest('.table-wrap')"
)
return page.evaluate_handle(f"() => document.querySelector('#{view}-table-wrap')")
def _reset_history_and_e2e_tokens() -> None:
"""The desktop pin's data state (see ``test_desktop_unchanged``):
an empty History table + no e2e-issued tokens — the row-width
interaction (auto table layout vs the 72rem card) is not the
contract under test, and other suites leave saved chats on the
shared dev DB. ``saved_chats``: TRUNCATE (the
``test_bottom_chat_actions.py`` precedent); ``api_tokens``:
label-scoped ``e2e-`` delete only (the ``test_api_tokens.py``
pattern — the dev DB may hold the owner's real tokens)."""
with SessionLocal() as db:
db.execute(text("TRUNCATE saved_chats"))
db.execute(text("DELETE FROM api_tokens WHERE label LIKE 'e2e-%'"))
db.commit()
def _assert_viewport_width(page: Page, label: str) -> None:
"""THE pin (TODO.md L4): the document itself never pans past the
viewport — pre-fix, /history.html measured 626 and /tokens.html
618 at this 375px viewport."""
report = page.evaluate(
"() => ({ doc: document.documentElement.scrollWidth, inner: window.innerWidth })"
)
assert report["doc"] <= report["inner"], (
f"{label}: the document panned to {report['doc']}px at a "
f"{report['inner']}px viewport — the TODO.md L4 defect "
'(the page is "the width of the table"); the .table-wrap card '
"must be the containing block for the positioned "
".visually-hidden spans (position: relative, phase 86)"
)
def _assert_in_card_scroll(page: Page, wrap: JSHandle, label: str) -> None:
"""The phase-07 / AGENTS.md-rule-5 contract the fix preserves: the
640px table scrolls INSIDE the card (``overflow-x: auto``), the
card is the container's content width (≤ the viewport), and the
table's ``min-width: 640px`` is what the in-card scroll is over."""
report = wrap.evaluate(
"""el => {
const c = el.closest('.container');
const cs = getComputedStyle(c);
const ws = getComputedStyle(el);
return {
scroll: el.scrollWidth,
client: el.clientWidth,
// The container's content width minus the card's own
// borders (clientWidth excludes them) — the card fills
// the row edge to edge.
content: c.clientWidth - parseFloat(cs.paddingLeft) - parseFloat(cs.paddingRight)
- parseFloat(ws.borderLeftWidth) - parseFloat(ws.borderRightWidth),
inner: window.innerWidth,
table: el.querySelector('table').scrollWidth,
};
}"""
)
assert report["table"] >= 640, (
f"{label}: the table must keep its 640px min-width, got {report['table']}px"
)
assert report["scroll"] > report["client"], (
f"{label}: the 640px table must still scroll INSIDE the card "
f"(card {report['client']}px < table {report['table']}px)"
)
assert report["client"] <= report["inner"], (
f"{label}: the card must not exceed the viewport "
f"({report['client']}px > {report['inner']}px)"
)
assert abs(report["client"] - report["content"]) <= 1, (
f"{label}: the card must fill the container's content width "
f"(card {report['client']}px vs container {report['content']}px)"
)
def _assert_hidden_a11y_spans(page: Page, wrap_sel: str) -> None:
"""The a11y names survived the fix: the visually-hidden Actions
column header and the table caption are still in the DOM (the fix
repositioned their CONTAINING BLOCK — it did not remove them)."""
expect(page.locator(f"{wrap_sel} th .visually-hidden")).to_have_count(1)
expect(page.locator(f"{wrap_sel} th .visually-hidden")).to_have_text("Actions")
expect(page.locator(f"{wrap_sel} caption.visually-hidden")).to_have_count(1)
# ---------------------------------------------------------------------------
# 1. THE TODO.md L4 pin: /history.html is viewport-width at 375px
# ---------------------------------------------------------------------------
def test_history_page_is_viewport_width(
browser: Browser, app_url: str, db_ready: None
) -> None:
"""THE TODO.md L4 pin: admin at 375×812, direct ``/history.html`` —
the document is viewport-width (pre-fix ``scrollWidth`` 626 — the
page panned ~250px into a blank region, "half the page being
blank"), the 640px table still scrolls INSIDE
``#history-table-wrap`` (the phase-07 / AGENTS.md-rule-5
full-width contract), and the visually-hidden Actions header +
caption are still in the DOM (the a11y names survived the fix)."""
page = _mobile_page(browser)
try:
login(page, app_url, next="/history.html")
_wait_settled_admin(page)
expect(page.locator("#history-table-wrap")).to_be_visible(timeout=15_000)
_assert_viewport_width(page, "/history.html (direct)")
_assert_in_card_scroll(page, _wrap_handle(page, "history"), "/history.html")
_assert_hidden_a11y_spans(page, "#history-table-wrap")
finally:
page.close()
# ---------------------------------------------------------------------------
# 2. The folded-in identical defect: /tokens.html (618 pre-fix)
# ---------------------------------------------------------------------------
def test_tokens_page_is_viewport_width(
browser: Browser, app_url: str, db_ready: None
) -> None:
"""The owner-confirmed A3 fold-in: the Tokens view carries the
byte-identical defect (measured ``scrollWidth`` 618 pre-fix — the
same visually-hidden Actions header + caption, no positioned
ancestor) and is fixed by the SAME one-rule ``.table-wrap`` fix.
Same two pins as test 1, on direct ``/tokens.html`` for
``#tokens-table-wrap``."""
page = _mobile_page(browser)
try:
login(page, app_url, next="/tokens.html")
_wait_settled_admin(page)
expect(page.locator("#tokens-table-wrap")).to_be_visible(timeout=15_000)
_assert_viewport_width(page, "/tokens.html (direct)")
_assert_in_card_scroll(page, _wrap_handle(page, "tokens"), "/tokens.html")
_assert_hidden_a11y_spans(page, "#tokens-table-wrap")
finally:
page.close()
# ---------------------------------------------------------------------------
# 3. The invariant survives the router (not just direct loads)
# ---------------------------------------------------------------------------
def test_spa_switch_keeps_page_width(
browser: Browser, app_url: str, db_ready: None
) -> None:
"""At 375px on ``/history.html``: open the mobile menu (a REAL
``#nav-toggle`` click — the phase-85 contract; at this width the
nav links live in the dropdown), click the Tokens link — a
CLIENT-SIDE view switch in the phase-76 shell, so the same-document
proof is the phase-76 canonical one: the window sentinel set
before the click is still readable after it (a real document load
would wipe ``window`` globals; the navigation-timeline count is
deliberately NOT used — a real load resets it, so it cannot
distinguish pushState from a reload). On arrival the Tokens page
keeps BOTH pins: viewport-width AND the table still scrolling
inside its card."""
page = _mobile_page(browser)
try:
login(page, app_url, next="/history.html")
_wait_settled_admin(page)
expect(page.locator("#history-table-wrap")).to_be_visible(timeout=15_000)
_assert_viewport_width(page, "/history.html (pre-switch)")
# THE SWITCH (mobile menu → Tokens link), same document:
page.evaluate("() => { window.__phase86_width = 'phase86'; }")
_open_menu(page)
page.click("#nav-tokens")
expect(page).to_have_url(app_url + "/tokens.html", timeout=15_000)
assert page.evaluate("() => window.__phase86_width") == "phase86", (
"a real document load would have wiped the window sentinel — "
"the view switch must be same-document (the phase-76 router)"
)
expect(page.locator("#tokens-table-wrap")).to_be_visible(timeout=15_000)
_assert_viewport_width(page, "/tokens.html (SPA switch)")
_assert_in_card_scroll(page, _wrap_handle(page, "tokens"), "/tokens.html (SPA)")
finally:
page.close()
# ---------------------------------------------------------------------------
# 4. Regression: the RAG view (already clean) stays clean + full-width
# ---------------------------------------------------------------------------
def test_rag_view_regression(
browser: Browser, app_url: str, mock_llm: int, db_ready: None
) -> None:
"""``/sources.html`` at 375px (admin): the RAG view measured clean
pre-fix (``scrollWidth`` 375 — the ``.docs-table`` headers are
visible text, no positioned hidden spans at the table's right
edge). Pinned so the shared ``.table-wrap`` rule change cannot
regress it — AND its table is still full-width inside the card:
the 640px ``min-width`` still engages, so the card keeps its
in-card scroll (the shared rule added a containing block, not a
width)."""
summary = _seed_kb(mock_llm)
assert summary is not None and summary.added == 13 # A9 formats (phase 47 added quadlet+j2)
page = _mobile_page(browser)
try:
login(page, app_url, next="/sources.html")
_wait_settled_admin(page)
expect(page.locator(".docs-table")).to_be_visible(timeout=15_000)
_assert_viewport_width(page, "/sources.html (direct)")
_assert_in_card_scroll(page, _wrap_handle(page, "sources"), "/sources.html")
finally:
page.close()
# ---------------------------------------------------------------------------
# 5. Desktop regression: no overflow anywhere, tables at container width
# ---------------------------------------------------------------------------
def test_desktop_unchanged(
browser: Browser, app_url: str, mock_llm: int, db_ready: None
) -> None:
"""1280×800 regression on ALL THREE table views: no document
overflow at any width, and each table renders at container width —
the cards need no internal scroll
(``wrap.scrollWidth <= wrap.clientWidth + 1``) and are far wider
than the mobile 346px card. The zero-offset
``position: relative`` changed no layout, so desktop is
byte-identical in behavior. The KB is seeded the house way for
the RAG view's table (it hides itself on an empty KB).
Data state: the desktop pin is enforced on the EMPTY-table state
the module docstring declares ("the History/Tokens contract holds
with EMPTY tables") — the History table's auto layout cannot fit
ROWS in the 1110px 72rem card (the nowrap date/share/actions
cells + the ellipsized title column at its 34rem ``max-width``
sum to ~1130px of minimum column width), and other suites on the
shared dev DB (e.g. ``test_responsive_polish``) leave saved chats
behind. So this test resets the state it pins: ``saved_chats``
TRUNCATEd (the ``test_bottom_chat_actions.py`` house precedent —
the table is rendered empty here, rows are not under test) and
the ``e2e-``-labeled tokens deleted (the ``test_api_tokens.py``
label-scoped pattern — never a TRUNCATE; the shared dev DB may
hold the owner's real tokens)."""
summary = _seed_kb(mock_llm)
assert summary is not None and summary.added == 13
_reset_history_and_e2e_tokens()
page = browser.new_page(viewport=DESKTOP)
try:
login(page, app_url, next="/history.html")
_wait_settled_admin(page)
views = (
("history", "/history.html"),
("tokens", "/tokens.html"),
("sources", "/sources.html"),
)
for i, (view, path) in enumerate(views):
if i: # the first view is the login landing
page.goto(app_url + path)
_wait_settled_admin(page)
marker = ".docs-table" if view == "sources" else f"#{view}-table-wrap"
expect(page.locator(marker)).to_be_visible(timeout=15_000)
_assert_viewport_width(page, f"{path} (desktop)")
report = _wrap_handle(page, view).evaluate(
"el => ({ scroll: el.scrollWidth, client: el.clientWidth })"
)
assert report["client"] > 346, (
f"{path} (desktop): the table card must render at container "
f"width, got {report['client']}px (the mobile card is 346px)"
)
assert report["scroll"] <= report["client"] + 1, (
f"{path} (desktop): no in-card scroll is needed at this width "
f"(table {report['scroll']}px in a {report['client']}px card)"
)
finally:
page.close()