fix(chat): keep in-flight answers alive across in-app view switches

Root cause (owner repro, verified in a real browser 2026-09-06): the
five navbar views (Chat, RAG, Sources, Tuning, History) were separate
HTML documents, so a navbar click was a REAL cross-document navigation
— the chat page unloaded, the in-flight SSE fetch was aborted, and the
phase-48 teardown (app/api/chat.py `finally`, "chat: turn cancelled")
stopped the model. Observed: send question -> click RAG mid-stream ->
click Chat -> the answer never finished: no `query_log` row, and on
return a dangling question with no brain record (the pre-token pagehide
partial persist skips because `acc` is empty).

Phase-48 LOCKED-DECISION REFINEMENT (owner-confirmed 2026-09-06,
flagged per AGENTS.md rule 3, not silently deviated): "real navigation
cancels the fetch" now means LEAVING THE APP — tab close,
external/other-document navigation, the Stop button. In-app navbar
switches are client-side view switches and no longer cancel.

Fix — Option A (SPA shell), chosen over B (Service Worker owns the
stream) and C (server-side turn registry + resume):
- frontend/index.html is the shell: ONE `<main id="main">` holds the
  five `<section class="view">` blocks; hidden views carry BOTH
  `hidden` and `inert` (WCAG — no focus/keyboard traversal). The
  shared header, the single `doc-modal-*` skeleton, and the
  `#app-version` footer each exist exactly once; the per-view copies
  from the four folded pages are dropped.
- New frontend/assets/router.js (vanilla module — no framework, no
  bundler, No-CDN rule intact): lazy-imports a view module on FIRST
  show only (mount-once, hide-forever — the chat view's in-flight SSE
  reader persists across switches; that persistence IS the fix);
  intercepts same-shell navbar links with preventDefault +
  history.pushState (never a document load); handles popstate; single
  writer of `.nav-link` active state (is-active + aria-current),
  document.title, and the per-view meta description (values carried
  over from the old pages' heads, brand-resolved at write time).
- Each folded page's JS becomes `export async function mount(root)` —
  root-scoped queries; `initSharedHeader()` dropped (the header boots
  once in the shell via the chat module; the admin flag comes from the
  same cached `fetchIsAdmin()` promise — zero extra requests).
- app/main.py: a small list-driven route factory serves the shell for
  /tuning.html, /sources.html, /git-sources.html, /history.html —
  registered AFTER the API routers and BEFORE the static catch-all
  (routes-first). The phase-33 caching middleware applies no-cache +
  `?v=` rewriting unchanged; app/core/caching.py needed NO change
  (the view paths did not change — pinned by the integration tests).
- The four old view .html files are DELETED (one source of truth);
  deep links to the old URLs keep working (the router picks the view
  from the pathname); `/?chat=<id>` is unaffected; the Containerfile
  bundles router.js (inlining the lazy view modules) and drops the
  folded page files.
- app/schemas.py: HistoryTurn.text cap 4000 -> 32000 — the shell
  keeps long saved answers in the chat, and the old cap (stricter than
  the 24_000-char total history budget) 422-rejected any second turn
  in such a chat (found by the phase-42 E2E suite on the shell).

Boundaries: login.html, shared.html, doc-edit.html, document.html
REMAIN separate documents (flow pages, not navbar tabs); a mid-stream
navigation to doc-edit/document.html still cancels per phase 48
(follow-up candidate, out of scope). The SSE API is unchanged. Real
departures still cancel the turn — phase 48 intact (pinned by
tests/e2e/test_stop_generation.py, unchanged, and by the new suite's
real-departure control).

Tests:
- Phase-20 suite REWRITTEN to the new semantics
  (tests/e2e/test_sources_midstream_bug.py): a navbar switch no longer
  cancels — the stream survives the switch and the FULL answer
  settles; the pagehide partial persist REMAINS for real departures
  (the partial's exact shape — first streamed chunk prefix, no done
  metadata — is still pinned there).
- NEW story suite tests/e2e/test_nav_switch_keeps_stream.py (mock
  LLM): the owner repro (send -> RAG mid-stream -> Chat: window
  sentinel survives = same document, FULL answer, exactly one brain
  turn in bor.chat.v1, exactly one settled query_log row, auto-saved
  row matches) + the same mid-stream switch against the other three
  views + the real-departure-still-cancels control + the no-switch
  baseline.
- tests/unit/test_frontend_router.py: source-level pins of the router
  invariants (click interceptor targets ONLY same-shell view paths,
  pushState-only switches, mount-once guard, hidden+inert pair,
  single-writer active state/title); shell-route integration tests
  (each folded path serves the shell with no-cache + `?v=` body; a
  non-view path still 404s); the file-reading unit pins re-pointed at
  the shell (the four view files are gone — the shell is the source
  of truth).

Verification (this commit): full suite green — 1565 unit+integration
tests, app/ coverage 99% (>90% floor); ruff + pyright clean; the
phase's E2E suites green in isolation (house protocol, AGENTS.md rule
9). Owner repro verified in a real browser against the real LLM
(dev server :8010, headful Chromium): "tell me about everquest" ->
RAG mid-stream -> Chat — the answer completed with one brain bubble
and no error banner, `query_log` gained exactly one settled row
(deflected=True: the dev KB holds no EverQuest docs — the settle, not
the topic, is the proof), zero "chat: turn cancelled" lines for that
turn; the control (real navigation to /shared.html mid-stream) still
cancelled (no settled row, the cancel line logged, the partial
persisted on return). Screenshots: .agents/screenshots/76_manual_*.

Phase 76 (76_spa_nav_shell) complete — moved to
.agents/phases/complete/.
This commit is contained in:
2026-09-06 06:31:31 -04:00
parent 7e567bddf3
commit ffa919b8bf
78 changed files with 5548 additions and 3244 deletions
+171 -56
View File
@@ -22,14 +22,14 @@ SOURCES_JS = ASSETS / "sources.js"
DOCUMENT_JS = ASSETS / "document.js"
LOGIN_JS = ASSETS / "login.js"
TUNING_JS = ASSETS / "tuning.js"
HISTORY_JS = ASSETS / "history.js"
STYLES_CSS = ASSETS / "styles.css"
# Phase 76 (task 02): SOURCES_HTML + GIT_SOURCES_HTML dropped — both
# views are folded into the shell (index.html); both files are deleted.
INDEX_HTML = FRONTEND / "index.html"
SOURCES_HTML = FRONTEND / "sources.html"
GIT_SOURCES_HTML = FRONTEND / "git-sources.html"
DOCUMENT_HTML = FRONTEND / "document.html"
LOGIN_HTML = FRONTEND / "login.html"
TUNING_HTML = FRONTEND / "tuning.html"
def _text(path: Path) -> str:
@@ -126,27 +126,37 @@ def test_nav_sources_ships_hidden_on_every_nav_page() -> None:
all five pages by phase 34 task 03 (owner confirmation 2026-08-26):
the RAG nav link is hidden for anonymous — so it SHIPS with the
hidden attribute (anonymous-safe default) on every page (they all
carry the nav now, viewer included)."""
for html in (INDEX_HTML, SOURCES_HTML, DOCUMENT_HTML, LOGIN_HTML, TUNING_HTML):
carry the nav now, viewer included).
Phase 76 (task 01): the page list is the post-shell set — the
folded Tuning view's header copy is gone with tuning.html (the
shell's ONE header is INDEX_HTML's). Phase 76 (task 02): the RAG
+ Sources view files drop out too (the shell's ONE header covers
all its views); task 03 drops History."""
for html in (INDEX_HTML, DOCUMENT_HTML, LOGIN_HTML):
text = _text(html)
assert re.search(r'id="nav-sources"[^>]*\bhidden\b', text), (
f"{html.name}: #nav-sources must ship hidden"
)
def test_all_six_pages_share_the_header_control_order() -> None:
def test_shell_and_standalone_pages_share_the_header_control_order() -> None:
"""Phase 34 task 03 (owner confirmation 2026-08-26) + phase 35/46
(the sixth page, git-sources): every page ships the IDENTICAL
header control inventory in the IDENTICAL order — brand, the mobile
hamburger, nav [Chat, #nav-sources, #nav-git-sources, #nav-tuning],
Sign in, Sign out (the #steering-toggle was removed from the navbar
at owner request, 2026-08-28) — inside the shared .header-inner row
(the document viewer's row 1). The #sync-btn (Sources page only)
and the #new-chat-btn (chat page only, moved from the navbar at
(the document viewer's row 1). The #sync-btn (RAG view only)
and the #new-chat-btn (chat view only, moved from the navbar at
owner request 2026-08-28) are NOT part of the shared bar anymore.
Only the current-page is-active nav marker and the static ?next=
fallback may differ per page (task 05's story E2E pins the
rendered result)."""
rendered result).
Phase 76 (task 02): the post-shell set — the folded views' header
copies are gone (the shell's ONE header covers all its views); the
RAG/Sources files are deleted, task 03 drops History."""
markers = (
'class="brand"',
'<nav class="app-nav"',
@@ -159,12 +169,9 @@ def test_all_six_pages_share_the_header_control_order() -> None:
)
for html in (
INDEX_HTML,
SOURCES_HTML,
GIT_SOURCES_HTML,
DOCUMENT_HTML,
TUNING_HTML,
LOGIN_HTML,
):
): # phase 76 (tasks 01/02): the folded view files are gone — the shell's ONE header stands in
text = _text(html)
start = text.find('<div class="container header-inner">')
assert start != -1, f"{html.name}: missing the shared .header-inner row"
@@ -185,8 +192,14 @@ def test_all_five_pages_carry_the_steering_panel() -> None:
"""Phase 34 task 03: the #steering-panel section (+ the
#steering-announcer live region) ships on every page — after
#kb-banner in the chat shell, first child of <main> on the other
four pages — ship hidden, driven by assets/header.js."""
for html in (INDEX_HTML, SOURCES_HTML, DOCUMENT_HTML, TUNING_HTML, LOGIN_HTML):
four pages — ship hidden, driven by assets/header.js.
Phase 76 (task 01): the folded Tuning view's panel COPY is dropped
(the shell keeps the chat's ONE instance — duplicate ids are not
legal). Phase 76 (task 02): the folded RAG + Sources view copies
are dropped with their files (the shell's ONE panel stands in);
task 03 drops History. The page list is the post-shell set."""
for html in (INDEX_HTML, DOCUMENT_HTML, LOGIN_HTML):
text = _text(html)
tag = re.search(r'<section[^>]*id="steering-panel"[^>]*>', text)
assert tag, f"{html.name}: missing the #steering-panel section"
@@ -200,23 +213,32 @@ def test_all_five_pages_carry_the_steering_panel() -> None:
assert text.find('id="steering-panel"') < text.find('id="steering-announcer"')
def test_nav_tuning_ships_hidden_on_the_tuning_page() -> None:
"""Phase 27: the Global Tuning page reuses the shared header — the
"Tuning" nav link is admin-only, so it SHIPS hidden (revealed by
initSharedHeader once whoami says admin), is the page's active link
(is-active + aria-current), and the page loads markdown.js (classic)
+ the tuning.js module with NO direct header.js <script> tag
(single-evaluation design)."""
text = _text(TUNING_HTML)
def test_nav_tuning_ships_hidden_and_unstamped_on_the_shell() -> None:
"""Phase 27 + phase 76 (task 01) — the shell form of this pin: the
"Tuning" nav link is admin-only, so it SHIPS hidden in the shell's
ONE header (revealed by initSharedHeader once whoami says admin).
The old page-level active stamp (is-active + aria-current on
tuning.html's own link) is GONE — the router is the SINGLE WRITER
of the active state (client-side, per view); only the Chat link
may carry a static stamp. The shell loads markdown.js (classic) +
app.js + router.js (modules) with NO direct header.js <script> tag
(single-evaluation design), and no direct tuning.js tag (the router
lazy-imports it on first show — mount-once)."""
text = _text(INDEX_HTML)
tag = re.search(r'<a[^>]*id="nav-tuning"[^>]*>', text)
assert tag, "tuning.html must carry the #nav-tuning nav link"
assert 'class="nav-link is-active"' in tag.group(0), "the Tuning link is the active one"
assert 'aria-current="page"' in tag.group(0)
assert tag, "the shell must carry the #nav-tuning nav link"
assert "hidden" in tag.group(0), "#nav-tuning must ship hidden (admin-only)"
srcs = _script_srcs(TUNING_HTML)
assert "is-active" not in tag.group(0), (
"no static active stamp — the router is the single writer"
)
assert 'aria-current="page"' not in tag.group(0)
srcs = _script_srcs(INDEX_HTML)
assert [s for s in srcs if "header.js" in s] == [], "no direct header.js <script> tag"
assert [s for s in srcs if "markdown.js" in s]
assert [s for s in srcs if "tuning.js" in s]
assert [s for s in srcs if "router.js" in s], "the shell loads the router module"
assert [s for s in srcs if "tuning.js" in s] == [], (
"no direct tuning.js tag — the router lazy-imports the view module"
)
def test_viewer_carries_the_standard_nav() -> None:
@@ -253,27 +275,52 @@ def test_viewer_carries_the_standard_nav() -> None:
assert back, "the back link keeps its /sources.html no-JS fallback"
def test_sources_and_viewer_carry_the_shared_controls() -> None:
"""Sources AND the document viewer carry the Sign in / Sign out pair
(both starting hidden — initSharedHeader reveals exactly one after
whoami), and their page scripts load header.js (phase 23: via the
page script's relative import, not a direct script tag). The New
chat button is NOT on non-chat pages — it moved from the navbar to
the chat page at owner request (2026-08-28; the single binding is
pinned in test_new_chat_binding_is_single_and_module_owned)."""
for html in (SOURCES_HTML, DOCUMENT_HTML):
def test_sources_view_and_viewer_carry_the_shared_controls() -> None:
"""Sources (the shell's RAG view, phase 76 task 02) AND the document
viewer carry the Sign in / Sign out pair (both starting hidden —
initSharedHeader reveals exactly one after whoami), and their
scripts load header.js (phase 23: via the relative import, not a
direct script tag). The New chat button is chat-view only — in the
shell its ONE instance lives in the chat view (hidden + inert on
every other view); the viewer carries none (the single binding is
pinned in test_new_chat_binding_is_single_and_module_owned).
The RAG view module does NOT boot the shell's header (it boots once
via app.js) — it keeps fetchIsAdmin() for its admin gate (zero
extra requests)."""
for html in (INDEX_HTML, DOCUMENT_HTML):
text = _text(html)
assert 'id="new-chat-btn"' not in text, (
f"{html.name}: the New chat button is chat-page only (owner request)"
)
assert re.search(r'id="sign-in-link"[^>]*\bhidden\b', text)
assert re.search(r'id="sign-out-btn"[^>]*\bhidden\b', text)
for js_file in (SOURCES_JS, DOCUMENT_JS):
assert 'from "./header.js"' in _text(js_file), (
f"{js_file.name}: page script must load the shared header module"
)
# The New chat button: exactly ONE instance in the shell (the chat
# view's) — none on the viewer.
assert _text(INDEX_HTML).count('id="new-chat-btn"') == 1
assert 'id="new-chat-btn"' not in _text(DOCUMENT_HTML), (
"the New chat button is chat-view only (owner request)"
)
assert 'from "./header.js"' in _text(DOCUMENT_JS), (
"document.js: page script must load the shared header module"
)
# The RAG view module (the phase-76 view-module branch):
# relative import, NO header boot, the fetchIsAdmin gate.
rag_js = _text(SOURCES_JS)
assert 'from "./header.js"' in rag_js
import_lines = [
line for line in rag_js.splitlines() if line.strip().startswith("import")
]
assert all("initSharedHeader" not in line for line in import_lines), (
"the view must not import the header boot — the shell's header boots "
"via the chat module (app.js) at shell boot"
)
assert "await initSharedHeader()" not in rag_js, (
"the view must not re-boot the shell's header"
)
assert "fetchIsAdmin()" in rag_js, "the admin gate keeps the shared cached promise"
assert "new-chat-btn" not in rag_js, (
"no #new-chat-btn binding (the module owns the single instance)"
)
assert 'fetch("/api/whoami")' not in rag_js
# Each page's Sign in link returns to ITS OWN page after login.
assert 'href="/login.html?next=/sources.html"' in _text(SOURCES_HTML)
assert 'href="/login.html?next=/sources.html"' in _text(INDEX_HTML)
assert 'href="/login.html?next=/document.html"' in _text(DOCUMENT_HTML)
@@ -288,7 +335,6 @@ def test_header_module_loads_before_the_page_script() -> None:
would double-bind the sign-out listener)."""
cases = [
(INDEX_HTML, "app.js"),
(SOURCES_HTML, "sources.js"),
(DOCUMENT_HTML, "document.js"),
(LOGIN_HTML, "login.js"),
]
@@ -307,6 +353,20 @@ def test_header_module_loads_before_the_page_script() -> None:
assert 'from "/assets/header.js"' not in js, (
f"{page_script}: absolute header import would break the esbuild bundle"
)
# Phase 76 (task 02): the folded RAG view module is NOT directly
# loaded by the shell — the router lazy-imports it on first show
# (mount-once, like the Tuning view module); it still imports
# header.js relatively (single-evaluation design).
shell_srcs = _script_srcs(INDEX_HTML)
assert [s for s in shell_srcs if "sources.js" in s] == [], (
"no direct sources.js <script> tag — the router lazy-imports the view"
)
assert [s for s in shell_srcs if "router.js" in s], (
"the router is the loader of the folded view modules"
)
js = _text(SOURCES_JS)
assert 'from "./header.js"' in js
assert 'from "/assets/header.js"' not in js
def test_login_page_carries_the_full_header() -> None:
@@ -403,16 +463,71 @@ def test_new_chat_binding_is_single_and_module_owned() -> None:
assert 'window.location.href = "/"' not in js, (
"no navigate-to-chat branch left (the button left the navbar)"
)
for js_file in (SOURCES_JS, DOCUMENT_JS, TUNING_JS):
page_js = _text(js_file)
assert 'from "./header.js"' in page_js
assert "initSharedHeader()" in page_js
assert "new-chat-btn" not in page_js, (
f"{js_file.name}: no #new-chat-btn binding (the module owns it)"
)
assert 'fetch("/api/whoami")' not in page_js, (
f"{js_file.name}: whoami goes through the shared cached promise"
)
doc_js = _text(DOCUMENT_JS)
assert 'from "./header.js"' in doc_js
assert "initSharedHeader()" in doc_js
assert "new-chat-btn" not in doc_js, (
"document.js: no #new-chat-btn binding (the module owns it)"
)
assert 'fetch("/api/whoami")' not in doc_js, (
"document.js: whoami goes through the shared cached promise"
)
# Phase 76 (task 02): the RAG view module is the VIEW-MODULE branch
# (like the Tuning view module below) — it does not boot the shell's
# header, binds no #new-chat-btn, and gates on the shared cached
# promise.
rag_js = _text(SOURCES_JS)
assert 'from "./header.js"' in rag_js
assert "await initSharedHeader()" not in rag_js
assert "fetchIsAdmin()" in rag_js, "the admin gate keeps the shared cached promise"
assert "new-chat-btn" not in rag_js, (
"sources.js: no #new-chat-btn binding (the module owns it)"
)
assert 'fetch("/api/whoami")' not in rag_js, (
"sources.js: whoami goes through the shared cached promise"
)
# Phase 76 (task 01): the folded Tuning VIEW module no longer boots
# the header — in the shell it runs exactly once, via the chat
# module (app.js) at shell boot. The view keeps fetchIsAdmin() for
# its admin gate (the SAME cached whoami promise — zero extra
# requests) and imports the module relatively (single-evaluation).
tuning_js = _text(TUNING_JS)
assert 'from "./header.js"' in tuning_js
# No CALL to the header boot — the import line carries no
# initSharedHeader and no `await initSharedHeader()` call site
# exists (a docstring may name it; a call may not).
import_lines = [
line for line in tuning_js.splitlines() if line.strip().startswith("import")
]
assert all("initSharedHeader" not in line for line in import_lines), (
"the view must not import the header boot — the shell's header boots "
"via the chat module (app.js) at shell boot"
)
assert "await initSharedHeader()" not in tuning_js, (
"the view must not re-boot the shell's header"
)
assert "fetchIsAdmin()" in tuning_js, "the admin gate keeps the shared cached promise"
assert "new-chat-btn" not in tuning_js
assert 'fetch("/api/whoami")' not in tuning_js
# Phase 76 (task 03): the folded History VIEW module — the same
# view-module contract: no header boot, no #new-chat-btn binding,
# the admin gate on the shared cached whoami promise (zero extra
# requests), the relative import for single evaluation.
history_js = _text(HISTORY_JS)
assert 'from "./header.js"' in history_js
import_lines = [
line for line in history_js.splitlines() if line.strip().startswith("import")
]
assert all("initSharedHeader" not in line for line in import_lines), (
"the view must not import the header boot — the shell's header boots "
"via the chat module (app.js) at shell boot"
)
assert "await initSharedHeader()" not in history_js, (
"the view must not re-boot the shell's header"
)
assert "fetchIsAdmin()" in history_js, "the admin gate keeps the shared cached promise"
assert "new-chat-btn" not in history_js
assert 'fetch("/api/whoami")' not in history_js
app_js = _text(APP_JS)
assert 'window.addEventListener("bor:new-chat", startNewChat)' in app_js
assert "newChatBtn.addEventListener" not in app_js, (