brain-of-reese/tests/unit/test_sources_midstream.py

"""Unit: the navigate-away partial-persistence contract (phase 20).

The browser behavior is E2E-covered (tests/e2e/test_sources_midstream_bug.py);
here we pin the source-level wiring in app.js — the single `pagehide`
listener, its in-flight guard (uiState thinking/streaming + non-empty acc),
the `rememberBrainTurn(acc, { thinking: thinkingAcc || undefined })` reuse
(no duplicated storage code), the turn-scoped `persistedOnLeave` idempotency
flag (declared at module scope, reset per turn in the turn handler), the
updated persistence-block comment — so a silent regression is caught without
a browser. The deliberate New Chat clear (clearChatStorage in header.js,
phase 14/19) must stay the only other clear: untouched.
"""
from __future__ import annotations

import re
from pathlib import Path

FRONTEND = Path(__file__).resolve().parents[2] / "frontend"
ASSETS = FRONTEND / "assets"
APP_JS = ASSETS / "app.js"
HEADER_JS = ASSETS / "header.js"


def _js() -> str:
    assert APP_JS.is_file()
    return APP_JS.read_text(encoding="utf-8")


def _pagehide_body(js: str) -> str:
    """The body of the single `pagehide` handler in app.js."""
    m = re.search(r'window\.addEventListener\("pagehide", \(\) => \{([\s\S]*?)\n\}\);', js)
    assert m, "app.js must register a window `pagehide` handler"
    return m.group(1)


def test_exactly_one_pagehide_listener_registered() -> None:
    """One and only one `pagehide` listener — the navigate-away save
    point (phase 20); no duplicated registration (bfcache churn is
    handled by the idempotency flag, not a second listener)."""
    js = _js()
    assert js.count('addEventListener("pagehide"') == 1


def test_pagehide_guard_requires_in_flight_state_and_streamed_text() -> None:
    """The handler persists ONLY when a turn is in flight (uiState is
    `thinking` or `streaming`) and answer text has already streamed
    (non-empty acc). Thinking-only turns persist nothing brain-side:
    the question is already saved on send and the user can re-ask."""
    body = _pagehide_body(_js())
    assert "UI_STATE.thinking" in body and "UI_STATE.streaming" in body, (
        "the guard must reference both in-flight states"
    )
    idx_state = body.find("uiState !== UI_STATE.thinking && uiState !== UI_STATE.streaming")
    assert idx_state != -1, "the in-flight guard is missing"
    idx_acc = body.find("if (!acc) return;")
    assert idx_acc != -1, "the non-empty-acc guard is missing"
    assert idx_state < idx_acc, "state guard must run before the acc guard"


def test_pagehide_reuses_remember_brain_turn_with_thinking() -> None:
    """The partial is persisted through the EXISTING save-point helper —
    raw text, optional thinking field (`undefined` drops the key from the
    JSON) + phase 158 (D4): the optional timeline (non-empty only), no
    sources/deflection: a plain brain message that restore re-renders
    exactly like a completed answer (no '(partial)' marker)."""
    js = _js()
    body = _pagehide_body(js)
    assert "rememberBrainTurn(acc, {" in body
    assert "thinking: thinkingAcc || undefined" in body
    assert (
        "segments: pagehideSegments.length ? pagehideSegments : undefined"
        in body
    ), "phase 158: the partial's timeline persists (non-empty only)"
    # No duplicated storage code: the handler pushes nothing itself.
    assert "conversation.push" not in body
    assert "saveConversation()" not in body
    # The partial carries no done metadata.
    assert "deflected" not in body and "sources" not in body


def test_persisted_on_leave_flag_is_module_scoped_and_turn_reset() -> None:
    """`persistedOnLeave` makes the save point idempotent (a second
    pagehide / bfcache store+restore never appends the same partial
    twice): checked first, set true immediately before the persist
    call. It is declared at module scope (the handler reads it) and
    reset to false at the top of the turn handler — turn-scoped like
    the other turn locals. acc / thinkingAcc were hoisted the same way
    (no behavior change: same reset point, same names)."""
    js = _js()
    # Module-scope declarations (column 0).
    assert re.search(r"^let persistedOnLeave = false", js, re.M)
    assert re.search(r"^let acc = \"\"", js, re.M)
    assert re.search(r"^let thinkingAcc = \"\"", js, re.M)
    # Each hoisted local is declared exactly once (module scope only —
    # the turn handler assigns, never re-declares).
    assert js.count("let acc = \"\"") == 1
    assert js.count("let thinkingAcc = \"\"") == 1
    assert js.count("let persistedOnLeave") == 1

    body = _pagehide_body(js)
    idx_check = body.find("if (persistedOnLeave) return;")
    idx_set = body.find("persistedOnLeave = true;")
    idx_call = body.find("rememberBrainTurn(acc,")
    assert -1 < idx_check < idx_set < idx_call, (
        "flag check first, set immediately before the persist call"
    )

    # Reset at the top of the turn handler (runTurn — phase 49 extracted
    # the turn from handleSend) — before the turn's fetch, where the
    # other turn locals are initialized. The pin is the ORDER (resets
    # before the fetch), not a char window: phase 123 task 02 grew the
    # save-point-1 region above the resets (the attached image's bubble
    # + record + strip clear) without moving the resets.
    turn = js.find("async function runTurn")
    assert turn != -1
    fetch_idx = js.find('fetch("/api/chat"', turn)
    assert fetch_idx != -1
    top = js[turn:fetch_idx]
    assert "persistedOnLeave = false;" in top, (
        "persistedOnLeave must be reset per turn, at the top of the turn handler"
    )
    assert "acc = \"\";" in top
    assert "thinkingAcc = \"\";" in top
    assert "let acc" not in top and "let thinkingAcc" not in top, (
        "the turn handler must assign the hoisted locals, not re-declare them"
    )
    assert "let persistedOnLeave" not in top


def test_persistence_comment_lists_the_pagehide_save_point() -> None:
    """The phase-14 persistence block comment now lists THREE save
    points: user message on send, brain message on `done`, and the
    PARTIAL brain message on navigate-away (`pagehide`, phase 20)."""
    js = _js()
    comment_start = js.find("conversation persistence (phase 14)")
    key_idx = js.find('const STORAGE_KEY = "bor.chat.v1"')
    assert -1 < comment_start < key_idx
    comment = js[comment_start:key_idx]
    assert re.search(r"Save points:.*?pagehide", comment, re.S), (
        "the Save points sentence must list the pagehide save point"
    )
    assert "PARTIAL brain" in comment


def test_new_chat_clear_is_untouched_and_still_the_only_deliberate_clear() -> None:
    """clearChatStorage (header.js) — the deliberate New Chat clear from
    the non-chat pages — is untouched and remains the only place that
    removes the literal key string; app.js's own clear goes through
    STORAGE_KEY (#new-chat-btn, phase 14/19), and nothing new was added."""
    header = HEADER_JS.read_text(encoding="utf-8")
    fn = header.find("function clearChatStorage")
    assert fn != -1
    body = header[fn : header.find("\n}", fn)]
    assert 'localStorage.removeItem("bor.chat.v1")' in body
    assert "try" in body and "catch" in body
    # No page script clears the key via the literal string.
    for name in ("app.js", "sources.js", "document.js", "login.js"):
        text = (ASSETS / name).read_text(encoding="utf-8")
        assert 'removeItem("bor.chat.v1")' not in text, (
            f"{name}: clearChatStorage in header.js is the only literal-key clear"
        )
    # app.js still has exactly one clear (clearStoredConversation, STORAGE_KEY).
    assert _js().count("removeItem(STORAGE_KEY)") == 1