phase: 109_turn_progress_loader
Build and Push Containers / build-and-push-app (push) Successful in 4m39s
Build and Push Containers / build-and-push-db (push) Successful in 35s

All gates green. Final report:

**Phase 109 — final verification pass (all 3 tasks were already complete; no re-implementation needed)**

- Verified D15 (thinking-handler re-open `block.open = true` after `ensureThinkingBlock`; delta keeps closing; "never reopens" narrative removed; restore path still collapses) and D16 (single static `#turn-loader` in composer row, `aria-hidden`, sole `turnLoader.hidden = !inFlight` write in `setUiState`; CSS reuses `typing` keyframes + reduced-motion override after the main rule + provenance; global `[hidden]` guard intact; mock `TURN_PROGRESS_TRIGGER` repro marker; `app/` untouched — pure UI phase).
- Unit pins: `tests/unit/test_frontend_turn_loader.py` (9 tests) — green via full run.
- `uv run pytest tests/e2e/test_turn_progress_loader.py -v --no-cov` → **3 passed** (isolation, DB up): loader visible send→tool-gap→final-thinking, hidden after done; re-opened scratchpad carries `marker-thought-42`/`marker-final-thought-42`; `#send-status` carries state text, back to empty idle; reduced-motion context stills dots (computed `animation-name: none`) with loader still visible.
- Regressions in isolation: `test_thinking_display` 5 passed, `test_stop_generation` 3 passed, `test_big_read_progress` 4 passed, `test_loading_feedback` 5 passed.
- `uv run pytest` → **2314 passed**; `uv run pytest --cov=app --cov-report=term-missing` → **TOTAL 99%** (>90%); `uv run ruff check .` clean; `uv run pyright` → 0 errors, 0 warnings.
- All terminal paths (done→idle L2554, stream error→error L2543, timeout→error L2247) funnel through `setUiState` — never-stale by construction.
- No defects found; no code changes made in this pass. Commit + phase-dir move left to the harness per pipeline rules (working tree carries all changes).
- Next pending phase: `108_history_wire_check` (still in `todo/`; out of scope here).
This commit is contained in:
2026-09-14 01:27:16 -04:00
parent fbbd98d734
commit 3a81793565
22 changed files with 1459 additions and 7 deletions
+256
View File
@@ -357,6 +357,44 @@ Implements just enough of the aipi surface:
client never sends tool calls/results) — so every existing marker
flow (which classifies statelessly from TOOL results and the LAST
user message) is unaffected by the now-always-present history.
- user message containing ``answer first, then list, then think``
(``TURN_PROGRESS_TRIGGER``, phase 109 task 03 — the never-frozen-
turn story's dedicated suite
``tests/e2e/test_turn_progress_loader.py``) **and** the system
prompt carries the ``<tools>`` section -> the deterministic
REPORTED-REPRO turn (TODO.md L3: "the model responds, calls a
tool, then continues thinking without re-expanding the thinking
block"): the owner's exact sequence, deterministic, with baked-in
delays (mid-turn windows of ≥1 s each — the ``slow_llm.py``
pacing precedent) so the story E2E's window assertions are
race-free. Discriminated statelessly from the messages (streaming
only):
* request 1 (``tools`` offered, no tool results yet): ~2 s pre-
delay (model latency — the loader's start-state window), a
short ``delta.content`` stream (3 chunks, NO reasoning — the
answer starts FIRST), then the ``ls`` tool call (synthetic id
``call_0``, no arguments — the phase-37 pattern),
``finish_reason: "tool_calls"``;
* request 2 (a ``tool``-role result in the messages — the server
ran the ``ls``): a ~7 s frameless gap (long enough that the
phase-87 tool-line ``(Ns)`` counter — 5 s+ after the line's
own arm — appears and ticks BEFORE the first ``thinking``
frame settles the line), then the ``reasoning_content`` stream
(10 × 0.3 s), a ``delta.content`` stream (3 chunks ending in
the distinctive final sentence carrying
``marker-progress-42``), and the FINAL ``reasoning_content``
chunks (3 × 0.3 s — thinking AFTER the answer, the reported
repro), ``finish_reason: "stop"``.
The server is position-independent over the wire (each
``reasoning_content`` chunk → a ``thinking`` SSE frame, each
``content`` chunk → a ``delta`` frame — ``app/rag/llm.py``; the
tool call materializes after its request's stream — the
content-before-tools convention), so the turn's SSE is exactly
``delta → tool → thinking → delta → thinking → done``.
Checked BEFORE the plain ``TOOLS_TRIGGER`` flow (disjoint trigger
phrases — the phase-71/72/94 ordering convention); verified
2026-09-16: no existing E2E question or fixture file contains the
phrase, so every other suite is unaffected.
Failure injection (phase 67, LLM retry, TODO.md L3) — deterministic
dead-endpoint behavior for the retry E2E suite (``tests/e2e/
@@ -1597,6 +1635,207 @@ def _folder_summary_incident(body: dict[str, Any]) -> bool:
return False
# ---------------------------------------------------------------------------
# Phase 109 (task 03, the never-frozen-turn story suite): the
# deterministic REPORTED-REPRO turn (delta → tool →
# thinking-after-delta) — see the module docstring
# ---------------------------------------------------------------------------
#: A user message containing this substring (case-insensitive) —
#: combined with the ``<tools>`` section in the system prompt — drives
#: the deterministic REPORTED-REPRO turn (TODO.md L3): the scripted
#: first answer (no reasoning), the no-arg ``ls``, then the post-tool
#: thinking → answer → FINAL thinking round — the owner's exact
#: "the model responds, calls a tool, then continues thinking" sequence
#: with baked-in delays (mid-turn windows of ≥1 s each — the
#: ``slow_llm.py`` pacing precedent) so the story E2E's window
#: assertions are race-free. Checked BEFORE the plain ``TOOLS_TRIGGER``
#: flow (disjoint trigger phrases — the phase-71/72/94 ordering
#: convention); verified 2026-09-16: no existing E2E question or
#: fixture file contains the phrase, so every other suite is
#: unaffected.
TURN_PROGRESS_TRIGGER = "answer first, then list, then think"
#: Request 1's pre-delay (s) — model latency: the loader's start-state
#: window (the turn is in flight, NO frame has arrived yet).
TURN_PROGRESS_PREDELAY_S = 2.0
#: Request 2's frameless gap (s) — the silence after the ``tool`` frame
#: before the first ``thinking`` frame. Deliberately past the phase-87
#: tool-line counter's 5 s gate (``TOOL_LINE_ELAPSED_AFTER_MS`` in
#: app.js) with ≥2 s of margin: the line's ``(Ns)`` suffix appears
#: (5 s tick) and ticks (6 s) BEFORE the first thinking frame settles
#: the line, so the story E2E can pin the counter in the tool gap
#: without a race.
TURN_PROGRESS_TOOL_GAP_S = 7.0
#: Per-chunk delays (s) for the two streams of request 2.
TURN_PROGRESS_THINK_DELAY_S = 0.3 # reasoning chunks (~10 × 0.3 s, ~3 × 0.3 s)
TURN_PROGRESS_CONTENT_DELAY_S = 0.05 # the short content deltas (3 + 3 chunks)
#: Chunk sizes (chars): the content deltas stay SHORT (2-3 chunks each),
#: the reasoning streams run 10 × 30 and 3 × 30.
TURN_PROGRESS_CONTENT_CHUNK = 40
TURN_PROGRESS_THINK_CHUNK = 30
#: Request 1's answer (byte-stable): 3 × 40-char content chunks, NO
#: reasoning — the answer starts FIRST (the repro's "the model
#: responds"). ``marker-progress-41`` is the sentinel the story E2E
#: matches to prove call 1's content landed in the bubble.
TURN_PROGRESS_FIRST_ANSWER = (
"Checking the listing first — opening answer: the kubernetes setup, "
"short and sweet. marker-progress-41."
)
#: Request 2's scratchpad (byte-stable): 10 × 30-char reasoning chunks —
#: thinking AFTER the answer, the reported repro. ``marker-thought-42``
#: is the sentinel the story E2E matches in the re-opened block.
TURN_PROGRESS_THINKING = (
"The listing just landed — now I can see which documents exist, so the "
"answer can anchor to the kubernetes file first and the deployments note "
"second, citing each fact by the exact path it came from. Hosts, "
"versions, and ports stay exactly as the notes write them. "
"marker-thought-42."
)
#: Request 2's answer (byte-stable): 3 × 40-char content chunks ending in
#: the DISTINCTIVE final sentence the story E2E matches on to pin the
#: post-delta state (``marker-progress-42`` — the repro's second answer).
TURN_PROGRESS_FINAL_ANSWER = (
"Here is the plan after the listing: step one, step two, step three — "
"that is the whole of it. marker-progress-42."
)
#: Request 2's FINAL reasoning (byte-stable): 3 × 30-char chunks — the
#: thinking that arrives AFTER the answer's last delta. The story E2E
#: waits for the re-opened block after the final sentence landed and
#: matches ``marker-final-thought-42`` to prove the last round's
#: thinking rendered in the scratchpad.
TURN_PROGRESS_FINAL_THINKING = (
"Final check — hosts and ports are verbatim. marker-final-thought-42."
)
def _turn_progress_flow(body: dict[str, Any]) -> str | None:
"""Classify a phase-109 reported-repro request (see the module
docstring). Stateless over the messages, like the other marker
flows:
* ``"first"`` — ``tools`` are offered and no ``tool``-role result
is in the messages yet: request 1 — the ~2 s pre-delay (model
latency — the loader's start-state window), the short
``delta.content`` stream (3 chunks, NO reasoning — the answer
starts first), then the ``ls`` tool call (synthetic id
``call_0``, no arguments), ``finish_reason: "tool_calls"``.
* ``"second"`` — a ``tool``-role result is in the messages (the
server ran the ``ls``): request 2 — the ~7 s frameless gap
(the phase-87 window), the ``reasoning_content`` stream (10 ×
0.3 s), the ``delta.content`` stream (3 chunks ending in the
distinctive final sentence), and the FINAL
``reasoning_content`` chunks (3 × 0.3 s) — thinking AFTER the
answer — then ``finish_reason: "stop"``.
* ``None`` — not the flow: the trigger is absent, the ``<tools>``
section is missing (deflected turns never carry it), or
``tools`` are not offered and no tool result is in the messages
yet (e.g. ``agent_max_rounds=0``).
"""
if TURN_PROGRESS_TRIGGER not in _user(body).lower():
return None
if "<tools>" not in _system(body):
return None
if _tool_results(body):
return "second"
if not body.get("tools"):
return None
return "first"
def _turn_progress_stream(step: str) -> Any:
"""SSE frames for one phase-109 reported-repro request.
``"first"`` (model call 1): the ~2 s pre-delay, the short
``delta.content`` stream (3 chunks — the answer starts first, NO
reasoning), the ``ls`` tool-call partial (synthetic id ``call_0``,
no arguments — the phase-37 ``_tool_call_stream`` first-partial
shape), and the ``finish_reason: "tool_calls"`` frame. The app is
position-independent over the wire (``app/rag/llm.py``): the
content chunks stream as ``delta`` SSE frames as they arrive and
the tool call materializes AFTER the stream (the content-before-
tools convention), so the request's SSE is exactly ``delta →
tool``.
``"second"`` (model call 2 — after the server ran the ``ls``): the
~7 s frameless gap (the phase-87 window — the tool line's
``(Ns)`` counter appears and ticks before the first frame settles
it), the ``reasoning_content`` stream (10 × 30 chars, 0.3 s each),
the ``delta.content`` stream (3 chunks ending in the distinctive
final sentence), the FINAL ``reasoning_content`` chunks (3 × 30
chars, 0.3 s each — thinking AFTER the answer, the reported
repro), and the ``finish_reason: "stop"`` frame.
"""
model = "turbo"
chunk_id = f"chatcmpl-{uuid.uuid4()}"
def frame(delta: dict[str, Any], finish: str | None = None) -> str:
return (
"data: "
+ json_dumps(
{
"id": chunk_id,
"object": "chat.completion.chunk",
"created": int(time.time()),
"model": model,
"choices": [{"index": 0, "delta": delta, "finish_reason": finish}],
}
)
+ "\n\n"
)
if step == "first":
time.sleep(TURN_PROGRESS_PREDELAY_S)
for piece in re.findall(
rf".{{1,{TURN_PROGRESS_CONTENT_CHUNK}}}", TURN_PROGRESS_FIRST_ANSWER, re.S
):
yield frame({"content": piece})
time.sleep(TURN_PROGRESS_CONTENT_DELAY_S)
yield frame(
{
"role": "assistant",
"tool_calls": [
{
"index": 0,
"id": "call_0",
"type": "function",
"function": {"name": "ls", "arguments": "{}"},
}
],
}
)
time.sleep(0.1)
yield frame({}, "tool_calls")
yield "data: [DONE]\n\n"
return
time.sleep(TURN_PROGRESS_TOOL_GAP_S)
for piece in re.findall(
rf".{{1,{TURN_PROGRESS_THINK_CHUNK}}}", TURN_PROGRESS_THINKING, re.S
):
yield frame({"reasoning_content": piece})
time.sleep(TURN_PROGRESS_THINK_DELAY_S)
for piece in re.findall(
rf".{{1,{TURN_PROGRESS_CONTENT_CHUNK}}}", TURN_PROGRESS_FINAL_ANSWER, re.S
):
yield frame({"content": piece})
time.sleep(TURN_PROGRESS_CONTENT_DELAY_S)
for piece in re.findall(
rf".{{1,{TURN_PROGRESS_THINK_CHUNK}}}", TURN_PROGRESS_FINAL_THINKING, re.S
):
yield frame({"reasoning_content": piece})
time.sleep(TURN_PROGRESS_THINK_DELAY_S)
yield frame({}, "stop")
yield "data: [DONE]\n\n"
def compose_answer(body: dict[str, Any]) -> str:
system = _system(body)
user = _user(body)
@@ -2222,6 +2461,23 @@ def chat_completions(body: dict[str, Any]) -> Any:
media_type="text/event-stream",
headers={"Cache-Control": "no-cache", "X-Accel-Buffering": "no"},
)
# Phase 109 (task 03, never-frozen-turn story suite): the
# deterministic REPORTED-REPRO turn (delta → tool →
# thinking-after-delta — TODO.md L3): the scripted first answer,
# the no-arg ls, then the post-tool thinking → answer → FINAL
# thinking round — the owner's exact repro with baked-in delays
# (see the module docstring). Checked BEFORE the plain
# TOOLS_TRIGGER flow (disjoint trigger phrases — the
# phase-71/72/94 ordering convention; the trigger needs the
# ``<tools>`` section, so deflected turns never hit it).
turn_progress = _turn_progress_flow(body)
if turn_progress is not None:
stream = _turn_progress_stream(turn_progress)
return StreamingResponse(
stream,
media_type="text/event-stream",
headers={"Cache-Control": "no-cache", "X-Accel-Buffering": "no"},
)
flow = _tool_flow(body)
if flow is not None:
if flow[0] == "list":
+321
View File
@@ -0,0 +1,321 @@
"""Phase 109 E2E (Playwright): the never-frozen turn — the reported
repro (delta → tool → thinking-after-delta) replayed deterministically.
Story: n/a — owner request 2026-09-16 (TODO.md L3): "Thinking can
happen after the model starts responding. This sometimes results in a
the chat appearing 'frozen' because the model responds, calls a tool,
then continues thinking without re-expanding the thinking block. There
should be a visual that the chat is still progressing regardless of
what state it's in (some kind of loader will do)."
Run in isolation (DB must be up: ``podman compose up -d db``):
uv run pytest tests/e2e/test_turn_progress_loader.py -v --no-cov
Marker contract: the mock's ``answer first, then list, then think``
marker (``TURN_PROGRESS_TRIGGER`` — see the ``mock_llm.py`` module
docstring) drives the deterministic reported repro. Model call 1
streams a short content delta (3 chunks, NO reasoning) after a ~2 s
pre-delay (model latency — the loader's start-state window), then the
no-arg ``ls`` tool call; model call 2 (after the server ran the ``ls``)
streams a ~7 s frameless gap (past the phase-87 tool-line ``(Ns)``
counter's 5 s gate), then a ``reasoning_content`` stream (10 × 0.3 s),
a content delta (3 chunks ending in the distinctive final sentence
carrying ``marker-progress-42``), and a FINAL ``reasoning_content``
chunk group (3 × 0.3 s). The server is position-independent over the
wire (``app/rag/llm.py`` — each ``reasoning_content`` chunk → a
``thinking`` SSE frame, each ``content`` chunk → a ``delta`` frame;
the tool call materializes after its request's stream), so the turn's
SSE is exactly ``delta → tool → thinking → delta → thinking → done`` —
the owner's repro, and every window is deterministic (the mock's
baked delays, the ``slow_llm.py`` pacing precedent).
Each test sends the marker question in its OWN fresh conversation
(one full turn per test — the ``page`` fixture is a fresh browser
context, ``_reset_db`` reseeds the KB deterministically):
* ``test_loader_visible_from_send_through_the_tool_gap`` — the loader
is visible right after the send (the thinking state, no frame yet —
the start-state window), is STILL visible when the ``.tool-call``
line lands (the turn is provably in flight), the tool line carries
the phase-87 elapsed counter in the tool gap, and ``#send-status``
carries a state text (not empty) throughout.
* ``test_thinking_block_reopens_after_delta_with_visible_loader`` —
THE reported symptom's state: the thinking block re-opens after the
answer has started (open + non-empty ``.thinking-text`` while the
bubble already carries call 1's content) with the loader STILL
visible (the frozen window is gone); the terminal state is clean
(loader hidden, the scratchpad carries both thinking rounds, the
bubble carries BOTH answers, the send button reads "Send" not
"Stop", ``#send-status`` back to the idle shape).
* ``test_loader_a11y_and_reduced_motion`` — after a full turn: the
loader element is ``aria-hidden="true"`` in the DOM and
``#send-status`` (the sole ``aria-live`` announcer) is back to the
``SEND_STATUS`` idle shape (empty — not stuck on a mid-turn label);
then, in a context with ``reducedMotion: "reduce"`` (the Playwright
context option), a second turn shows the loader STILL visible
mid-turn — the reduced-motion variant stills the dots (computed
``animation-name: none``), it does not HIDE the cue: visibility is
owned by the JS ``hidden`` attribute in ``setUiState`` (D16), not by
CSS.
"""
from __future__ import annotations
import asyncio
import re
from pathlib import Path
from threading import Thread
from typing import Any
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"
#: The marker question — the mock's ``TURN_PROGRESS_TRIGGER`` phrase +
#: an on-topic tail (HIGH gate → the ``<tools>`` section, so the
#: scripted ``ls`` round is reachable — the phase-37/70 convention).
QUESTION = "answer first, then list, then think — how is my Kubernetes cluster set up?"
#: The mock's byte-stable sentinels (``mock_llm.py``): call 1's
#: content, call 2's final sentence, call 2's scratchpad, call 2's
#: FINAL scratchpad (the thinking after the answer's last delta).
SENTINEL_FIRST = "marker-progress-41"
SENTINEL_FINAL = "marker-progress-42"
SENTINEL_THOUGHT = "marker-thought-42"
SENTINEL_FINAL_THOUGHT = "marker-final-thought-42"
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'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 _reset_db(mock_port: int, seed: bool) -> ImportSummary | None:
"""Truncate the KB (+ query log + steering notes — deterministic
mock answers), then optionally re-import fixtures."""
with SessionLocal() as db:
db.execute(text("TRUNCATE chunks, documents, query_log, steering_notes"))
db.commit()
if not seed:
return None
return _run_in_thread(_import_fixtures(mock_port))
def _send(page: Page) -> None:
"""Send the marker question (one full reported-repro turn)."""
page.fill("#message-input", QUESTION)
page.click("#send-btn")
def _answer_bubble(page: Page) -> Any:
"""The answer bubble (the typing indicator's bubble is ``.typing``
— scoped out)."""
return page.locator(".msg.brain .bubble:not(.typing)")
def test_loader_visible_from_send_through_the_tool_gap(
page: Page, app_url: str, mock_llm: int, db_ready: None
) -> None:
summary = _reset_db(mock_llm, seed=True)
assert summary is not None and summary.added == 13 # A9 formats (phase 47 added quadlet+j2)
page.set_default_timeout(30_000)
login(page, app_url, next="/")
_send(page)
# Right after the send — inside call 1's ~2 s pre-delay window:
# the turn is in flight (the thinking state, NO frame has arrived
# yet) and the loader is ALREADY visible — the start-state cue.
# ``setUiState(thinking)`` runs synchronously in the submit path,
# so this is race-free: the button is the Stop control, the
# loader is out, and no frame-derived surface exists yet (no
# answer bubble, no thinking block, no tool line — only the
# pre-delta typing bubble).
loader = page.locator("#turn-loader")
expect(loader).to_be_visible()
expect(page.locator("#send-label")).to_have_text("Stop")
expect(page.locator("#typing-indicator")).to_be_visible()
expect(_answer_bubble(page)).to_have_count(0)
expect(page.locator(".msg.brain .thinking")).to_have_count(0)
expect(page.locator(".tool-call")).to_have_count(0)
# The ``.tool-call`` line lands (call 1's content + the ls): the
# turn is provably in flight — and the loader is STILL visible.
expect(page.locator(".tool-call")).to_be_visible(timeout=30_000)
expect(loader).to_be_visible()
expect(_answer_bubble(page)).to_contain_text(SENTINEL_FIRST)
# The tool gap (call 2's ~7 s frameless delay — deliberately past
# the phase-87 counter's 5 s gate): the tool line's elapsed
# counter appears and ticks BEFORE the first thinking frame
# settles the line ...
elapsed = page.locator(".tool-call .tool-elapsed")
expect(elapsed).to_be_visible(timeout=20_000)
expect(elapsed).to_have_text(re.compile(r"\(\d+s\)"))
# ... and the loader + the #send-status state text hold through
# the whole gap (the status line carries the streaming state's
# text — it is never empty while a frame is in flight).
expect(loader).to_be_visible()
expect(page.locator("#send-status")).not_to_be_empty()
def test_thinking_block_reopens_after_delta_with_visible_loader(
page: Page, app_url: str, mock_llm: int, db_ready: None
) -> None:
summary = _reset_db(mock_llm, seed=True)
assert summary is not None and summary.added == 13 # A9 formats (phase 47 added quadlet+j2)
page.set_default_timeout(30_000)
login(page, app_url, next="/")
_send(page)
bubble = _answer_bubble(page)
block = page.locator(".msg.brain .thinking")
text_el = page.locator(".msg.brain .thinking-text")
loader = page.locator("#turn-loader")
# THE reported symptom's state (D15): the thinking block is open
# with non-empty text WHILE the answer bubble already carries
# call 1's content — the post-delta re-open (call 2's thinking
# frames re-opened the block the first delta had closed).
page.wait_for_function(
"""() => {
const b = document.querySelector('.msg.brain .thinking');
const t = b ? b.querySelector('.thinking-text') : null;
return !!b && b.open && !!t && t.textContent.trim().length > 0;
}""",
timeout=30_000,
)
expect(block).to_have_count(1)
expect(block).to_have_attribute("open", "")
expect(text_el).to_contain_text(SENTINEL_THOUGHT)
# The re-open is genuinely AFTER a delta: call 1's content is
# already in the bubble ...
expect(bubble).to_contain_text(SENTINEL_FIRST)
# ... and the loader is STILL visible — the frozen window is gone.
expect(loader).to_be_visible()
# Terminal approach: call 2's distinctive final sentence landed in
# the bubble ...
expect(bubble).to_contain_text(SENTINEL_FINAL, timeout=30_000)
# ... and the FINAL thinking round (the last frames before done —
# thinking AFTER the answer's last delta) re-opened the block
# AGAIN: the scratchpad carries the last round's text and the
# block is open (the `thinking` handler opens it before it
# renders — D15), with the loader still visible (the turn is
# still in flight — no terminal frame yet).
expect(text_el).to_contain_text(SENTINEL_FINAL_THOUGHT, timeout=30_000)
expect(block).to_have_attribute("open", "")
expect(loader).to_be_visible()
# The turn settles (done → idle through setUiState): the loader is
# HIDDEN (its sole owner — D16, no per-handler cleanup) ...
expect(page.locator("#send-label")).to_have_text("Send", timeout=30_000)
expect(loader).to_be_hidden()
# ... the scratchpad text is intact (BOTH thinking rounds) ...
expect(text_el).to_contain_text(SENTINEL_THOUGHT)
expect(text_el).to_contain_text(SENTINEL_FINAL_THOUGHT)
# ... the bubble carries BOTH answers (call 1 + call 2) ...
expect(bubble).to_contain_text(SENTINEL_FIRST)
expect(bubble).to_contain_text(SENTINEL_FINAL)
# ... the send button is back to "Send" (asserted above via the
# wait) and NOT the Stop treatment ...
expect(page.locator("#send-btn")).not_to_have_class(re.compile(r"is-stop"))
# ... and #send-status is back to the idle shape (not stuck on a
# mid-turn label).
expect(page.locator("#send-status")).to_have_text("")
def test_loader_a11y_and_reduced_motion(
page: Page, browser: Browser, app_url: str, mock_llm: int, db_ready: None
) -> None:
summary = _reset_db(mock_llm, seed=True)
assert summary is not None and summary.added == 13 # A9 formats (phase 47 added quadlet+j2)
page.set_default_timeout(30_000)
login(page, app_url, next="/")
_send(page)
# The full turn lands (the distinctive final sentence + done →
# the Send button back).
expect(_answer_bubble(page)).to_contain_text(SENTINEL_FINAL, timeout=60_000)
expect(page.locator("#send-label")).to_have_text("Send", timeout=30_000)
# The a11y split: the loader element is ``aria-hidden="true"`` in
# the DOM (decorative — it never carries meaning) ...
expect(page.locator("#turn-loader")).to_have_attribute("aria-hidden", "true")
# ... and #send-status (the sole ``aria-live`` announcer) is back
# to the SEND_STATUS idle shape after done — EMPTY, not stuck on a
# mid-turn label ("… is thinking" / "… is answering").
status = page.locator("#send-status")
expect(status).to_have_attribute("aria-live", "polite")
expect(status).to_have_text("")
# Control (default-motion context): the loader's dots run the
# typing animation — the reduced-motion check below stills it.
assert (
page.evaluate(
"getComputedStyle(document.querySelector('#turn-loader')).animationName"
)
== "typing"
)
# Reduced motion: a FRESH context with ``reducedMotion: "reduce"``
# (the Playwright context option) — send a second turn and pin
# that the loader is STILL visible mid-turn: the reduced-motion
# variant stills the dots (the CSS rule changes — computed
# ``animation-name: none``), but visibility is owned by the JS
# ``hidden`` attribute in setUiState (D16), not by CSS.
rm_context = browser.new_context(
reduced_motion="reduce", viewport={"width": 1280, "height": 800}
)
try:
rm_page = rm_context.new_page()
rm_page.set_default_timeout(30_000)
login(rm_page, app_url, next="/")
_send(rm_page)
# Mid-turn (the tool gap — call 2's ~7 s frameless delay): the
# loader is still the visible cue ...
expect(rm_page.locator(".tool-call")).to_be_visible(timeout=30_000)
expect(rm_page.locator("#turn-loader")).to_be_visible()
# ... and the reduced-motion CSS variant IS applied: static
# dots, no pulse (the animation is gone from the computed
# style — the element itself is unchanged).
assert (
rm_page.evaluate(
"getComputedStyle(document.querySelector('#turn-loader')).animationName"
)
== "none"
)
finally:
rm_context.close()
+319
View File
@@ -0,0 +1,319 @@
"""Unit: the phase-109 never-frozen-turn frontend contract (tasks 01-02).
No new Python app logic exists for this task — the behavior lives in
``frontend/assets/app.js`` and is E2E-gated by the story suite (task 03).
Like the other frontend-adjacent unit files (the house
read-the-assets-as-text pattern, cf. ``test_frontend_tool_states.py``),
this module pins the JS markers the phase depends on: the thinking block
becomes a TOGGLE (D15) — ``thinking`` frames open it (idempotent),
``delta`` frames close it — while the phase-14 restore path and the
phase-17 follow-the-tail pin logic stay byte-for-byte untouched. Task 02
pins the persistent in-turn loader (D16): the static ``#turn-loader``
markup, ``setUiState`` as its SOLE visibility owner (the structural
never-stale guarantee), and the CSS contract (reused typing-dot
animation + reduced-motion + provenance).
"""
from __future__ import annotations
import re
from pathlib import Path
FRONTEND = Path(__file__).resolve().parents[2] / "frontend"
APP_JS = FRONTEND / "assets" / "app.js"
INDEX_HTML = FRONTEND / "index.html"
STYLES_CSS = FRONTEND / "assets" / "styles.css"
def _js() -> str:
return APP_JS.read_text(encoding="utf-8")
def _html() -> str:
return INDEX_HTML.read_text(encoding="utf-8")
def _css() -> str:
return STYLES_CSS.read_text(encoding="utf-8")
def _thinking_branch() -> str:
"""The `thinking` handler branch (between the thinking and tool
branches of the turn dispatch)."""
js = _js()
thinking_idx = js.find('ev.type === "thinking"')
tool_idx = js.find('ev.type === "tool"')
assert -1 < thinking_idx < tool_idx, (
"the turn handler must branch on thinking frames"
)
return js[thinking_idx:tool_idx]
def test_thinking_handler_reopens_the_collapsed_block() -> None:
"""Phase 109 (D15): the `thinking` handler re-opens the block —
``block.open = true`` sits in the handler, positioned AFTER the
``ensureThinkingBlock(wrap)`` line (the block must exist before it
can be opened). Idempotent: a no-op while already open (the
pre-delta live flow), a re-open after a `delta` closed it (the next
agent round — the reported freeze, TODO.md L3)."""
branch = _thinking_branch()
ensure_idx = branch.find("ensureThinkingBlock(wrap)")
assert ensure_idx != -1, "the handler must ensure the block first"
reopen_idx = branch.find("block.open = true")
assert reopen_idx != -1, (
"the `thinking` handler must open the block (D15: "
"open-while-thinking)"
)
assert ensure_idx < reopen_idx, (
"`block.open = true` must sit AFTER `ensureThinkingBlock(wrap)` "
"— the block must exist before it is opened"
)
def test_delta_handler_still_closes_the_block() -> None:
"""Phase 109 (D15): the close side of the toggle survives — the
`delta` handler still calls ``closeThinkingBlock(wrap)`` (closed
while answering); nothing else about the block's lifecycle changes."""
js = _js()
delta_idx = js.find('ev.type === "delta"')
done_idx = js.find('ev.type === "done"')
assert -1 < delta_idx < done_idx, "the turn handler must branch on delta"
branch = js[delta_idx:done_idx]
assert "closeThinkingBlock(wrap)" in branch, (
"the `delta` handler must keep closing the block (D15: "
"closed-while-answering)"
)
def test_close_thinking_block_docstring_says_toggle_not_one_way() -> None:
"""Phase 109 (D15): the narrative flips. The section comment
documenting ``closeThinkingBlock`` no longer claims "never reopens"
— nowhere in app.js does — and documents the toggle contract instead:
open-while-thinking / closed-while-answering, naming the `thinking`
handler's re-open and the `delta` handler's close. The function
body itself is unchanged (idempotent close, no-op without a
block)."""
js = _js()
fn = js.find("function closeThinkingBlock")
assert fn != -1, "closeThinkingBlock must exist"
body = js[fn : js.find("\n}\n", fn)]
assert "block.open = false" in body, "the close side is unchanged"
# The block comment that documents the helpers: walk back from the
# function to its section comment start.
comment_start = js.rfind("/*", 0, fn)
assert comment_start != -1
section = js[comment_start:fn]
assert "never reopens" not in section, (
"the one-way-door claim is gone (phase 109, D15)"
)
assert "never reopens" not in js, (
"the 'never reopens' narrative must not survive anywhere in app.js"
)
# The toggle contract is documented — and it names both sides:
# the `thinking` handler's re-open and the `delta` handler's close.
assert "TOGGLE" in section, "the contract is named: a toggle"
assert "open-while-thinking" in section
assert "closed-while-answering" in section
assert "`thinking`" in section, "the contract names the thinking handler"
assert "re-opens" in section, "the contract names the thinking handler's re-open"
assert "`delta`" in section, "the contract names the delta handler's close"
def test_restore_path_still_collapses_stored_blocks() -> None:
"""Phase-14 contract regression: ``renderStoredMessage`` restores
stored thinking blocks COLLAPSED (``block.open = false``) — the
phase-109 toggle only changes the LIVE `thinking` handler, the
restore path is untouched."""
js = _js()
fn = js.find("function renderStoredMessage")
assert fn != -1, "renderStoredMessage must exist"
body = js[fn : js.find("\n}\n", fn)]
assert "if (m.thinking)" in body
assert "ensureThinkingBlock(wrap)" in body
assert "block.open = false" in body, "stored blocks restore collapsed"
# The live re-open must NOT have leaked into the restore path:
# only the collapse assignment may appear in the function body.
assert "block.open = true" not in body, (
"the restore path must never open a stored block (phase-14)"
)
def test_follow_the_tail_pin_logic_is_untouched() -> None:
"""Cross-file regression (phase 17): the follow-the-tail pin is
unchanged — ``THINKING_NEAR_BOTTOM_PX`` (32px band) +
``isThinkingNearBottom`` and the pre-render capture
``const pinned = block.open && isThinkingNearBottom(textEl);``
(measured BEFORE the re-render) still sit in the `thinking` handler.
The capture already keys off ``block.open``, so a re-opened block
resumes pinned tail-following exactly like the live pre-delta
block — no pin logic change needed for D15."""
js = _js()
assert "export const THINKING_NEAR_BOTTOM_PX = 32;" in js, (
"the 32px window band is unchanged"
)
fn = re.search(
r"function isThinkingNearBottom\(textEl\) \{([\s\S]*?)\n\}", js
)
assert fn, "isThinkingNearBottom must still exist"
assert "THINKING_NEAR_BOTTOM_PX" in fn.group(1)
branch = _thinking_branch()
capture = "const pinned = block.open && isThinkingNearBottom(textEl);"
assert capture in branch, (
"the pre-render `block.open &&` guard line is unchanged — D15 "
"keeps the pin logic untouched (it already reads block.open "
"before the re-render)"
)
capture_idx = branch.find(capture)
render_idx = branch.find("textEl.innerHTML = renderMarkdown(thinkingAcc)")
assert -1 < capture_idx < render_idx, (
"pin state is still measured BEFORE the re-render (phase-17 "
"regression — the 2-newline-gap fix)"
)
assert "textEl.scrollTop = textEl.scrollHeight" in branch, (
"pinned tail-following is intact"
)
# ---------- task 02: the persistent in-turn loader (D16) ----------
def test_index_html_carries_exactly_one_turn_loader() -> None:
"""Phase 109 (D16): ``index.html`` carries exactly ONE static
``#turn-loader`` — an empty ``<div>`` (never constructed in JS: the
createElement/textContent house rule), ``aria-hidden="true"``
(decorative — ``#send-status`` carries the meaning), ``hidden`` by
default (idle on load), and it sits INSIDE the composer form — the
composer's status row, the visible companion of the ``#send-status``
line."""
html = _html()
assert html.count('id="turn-loader"') == 1, (
"exactly one #turn-loader — static markup, never JS-built"
)
match = re.search(r'<div\s+id="turn-loader"[^>]*>', html)
assert match, "the loader must be a static <div>"
tag = match.group(0)
assert 'class="turn-loader"' in tag
assert 'aria-hidden="true"' in tag, "decorative — aria-hidden"
# the hidden attribute (a standalone word in the tag, not the
# substring of some other attribute value): ships hidden (idle).
assert re.search(r"\shidden\s*/?>$", tag), (
"the loader ships hidden (idle on load)"
)
# empty element: the closing tag follows immediately — no children,
# no JS-built HTML anywhere near it.
assert html[match.end() : match.end() + 6] == "</div>", (
"the loader element is empty (static shell)"
)
# inside the composer form (the status row), not elsewhere in the
# shell.
composer_idx = html.find('<form class="composer" id="composer"')
assert composer_idx != -1
form_end = html.find("</form>", composer_idx)
assert composer_idx < match.start() < form_end, (
"the loader sits in the composer's status row"
)
def test_set_ui_state_is_the_sole_owner_of_the_loader() -> None:
"""Phase 109 (D16) — the never-stale guarantee, structural: in
``app.js`` the string ``turnLoader.hidden`` appears EXACTLY ONCE,
inside ``setUiState`` (the cross-file single-owner check — any
second write site fails this test). The toggle is
``turnLoader.hidden = !inFlight`` next to the existing ``is-stop``
toggle: shown iff ``uiState`` is thinking|streaming. Every terminal
path funnels through ``setUiState`` (done → idle, error → error,
stop/timeout → their landings), so the loader hides in every
terminal state BY CONSTRUCTION — no per-handler cleanup."""
js = _js()
# The element lookup joins the other module-top lookups.
assert 'const turnLoader = document.querySelector("#turn-loader");' in js
# THE invariant: exactly one write site in the whole file.
assert js.count("turnLoader.hidden") == 1, (
"setUiState is the SOLE writer of the loader's hidden attribute — "
"a second write site breaks the §7.4 never-stale guarantee"
)
# ...and it sits INSIDE setUiState, next to the is-stop toggle.
fn = js.find("export function setUiState")
assert fn != -1
body_end = js.find("\n}\n", fn)
owner_idx = js.find("turnLoader.hidden")
assert fn < owner_idx < body_end, (
"the toggle must live inside setUiState (the single entry point)"
)
stop_idx = js.find('sendBtn.classList.toggle("is-stop", inFlight);', fn)
assert -1 < stop_idx < owner_idx, (
"the loader toggle joins the existing is-stop toggle"
)
assert "turnLoader.hidden = !inFlight;" in js, (
"shown iff inFlight (thinking|streaming)"
)
def test_loader_css_reuses_the_typing_animation_and_reduced_motion() -> None:
"""Phase 109 (D16) — the CSS contract: the ``.turn-loader`` rule
sits NEXT TO the typing-dots rules (after the ``@keyframes typing``
block), reuses the SAME animation name (``typing`` — no new
animation family), a ``prefers-reduced-motion`` block covers it
(static dots, no pulse — §7.2 house law), and the provenance
comment names phase 109 + ``TODO.md``."""
css = _css()
keyframes_idx = css.find("@keyframes typing")
assert keyframes_idx != -1, "the typing-dot keyframes exist"
# The MAIN rule (".turn-loader {" — the reduced-motion block's
# selector list never starts that exact string) must sit next to
# (after) the typing-dots rules.
rule_idx = css.find(".turn-loader {")
assert rule_idx != -1, "styles.css must carry a .turn-loader rule"
assert keyframes_idx < rule_idx, (
"the .turn-loader rule sits next to (after) the typing-dots rules"
)
# The provenance comment: the /* ... */ block immediately preceding
# the rule names phase 109 and TODO.md.
comment_start = css.rfind("/*", 0, rule_idx)
assert comment_start != -1
comment = css[comment_start:rule_idx]
assert "Phase 109" in comment, "the provenance comment names phase 109"
assert "TODO.md" in comment, "the provenance comment cites the source"
# Reuses the EXISTING typing-dot animation: the same animation name
# in the rule and its pseudo-element rules (no new @keyframes
# family anywhere).
rule_block = css[rule_idx : rule_idx + 1200]
assert re.search(r"animation:\s*typing\b", rule_block), (
"the loader reuses the typing dots' animation (same name — no "
"new animation family)"
)
assert "@keyframes turn" not in css, "no new animation family for the loader"
# The prefers-reduced-motion block covers it: static dots, no pulse.
reduced_ok = False
for m in re.finditer(r"@media \(prefers-reduced-motion: reduce\) \{([\s\S]*?)\n\}", css):
if ".turn-loader" in m.group(1) and "animation: none" in m.group(1):
reduced_ok = True
break
assert reduced_ok, (
"a prefers-reduced-motion block must cover .turn-loader (static "
"dots, no pulse — §7.2)"
)
def test_loader_is_aria_hidden_and_status_untouched() -> None:
"""Phase 109 (D16) — the a11y split is untouched: the loader is
``aria-hidden`` (decoration), and ``#send-status`` keeps its exact
attributes in ``index.html`` — still the sole ``aria-live``
announcer (the house a11y split: visual cues are aria-hidden, the
live region carries the state text)."""
html = _html()
match = re.search(r'<div\s+id="turn-loader"[^>]*>', html)
assert match and 'aria-hidden="true"' in match.group(0)
# #send-status is unchanged: one element, the exact same tag —
# the visually-hidden polite live region inside the send button,
# still the chat view's state announcer.
assert html.count('id="send-status"') == 1
status = re.search(r'<span[^>]*id="send-status"[^>]*></span>', html)
assert status, "the #send-status live region is intact"
tag = status.group(0)
assert tag == (
'<span class="visually-hidden" aria-live="polite" '
'id="send-status"></span>'
), "the #send-status attributes are byte-unchanged"
# the loader added no a11y surface: it is aria-hidden and carries
# no live region of its own.
assert 'aria-live' not in match.group(0)