fix(tests): use pytest_terminal_summary as primary 'session done' signal

The previous smart watchdog (44b0b5d4, 91b19c90) used pytest_unconfigure as its signal. But pytest_unconfigure fires AFTER all fixtures, terminal summary, and finalizers — at the very end of the session. If anything in conftest's chain (e.g., the io_pool created in AppController.__init__ at conftest line ~65) hangs in __del__, pytest_unconfigure never gets called. Result: every batch's watchdog waited the full 60s/90s and then fired.

The right signal is pytest_terminal_summary, which fires AFTER the test summary is printed (the user can see '241 passed, 1 skipped in 32.30s' in the output) but BEFORE the shutdown hangs begin. At that point the test session is logically done; the watchdog can give a short 5s grace for normal finalization, then os._exit(0) so the runner can move to the next batch.

The previous attempts and why they failed (documented in test_conftest_smart_watchdog.py docstring):
  - e1c8730f: 30s os._exit(0) cut off batches mid-test
  - 719c5e27: os._exit(2) but daemon thread fired on every batch
  - 91b19c90: kept exit 2 but pytest_unconfigure never fires when io_pool hangs
  - 44b0b5d4: pytest_unconfigure as signal still hung
  - 2026-06-07 final: pytest_terminal_summary fires after summary print, before shutdown hangs

New contract:
  - Normal batch: pytest_terminal_summary fires at ~32s (after summary
    is printed), 5s grace, os._exit(0). Total: 37s.
  - Hung in test execution: pytest_terminal_summary never fires,
    smart watchdog waits 300s, fires os._exit(2).
  - Hung in conftest load (before any test): unconditional watchdog
    fires os._exit(2) at 60s.

7 tests in test_conftest_smart_watchdog.py updated to match:
  - test_terminal_summary_hook_sets_finished_event: primary signal source
  - test_unconfigure_hook_is_fallback_signal: fallback for crashes
  - test_clean_exit_uses_zero_exit_code: os._exit(0) after signal
  - test_hang_uses_nonzero_exit_code: os._exit(2) for true hangs

tests/conftest.py

@@ -67,8 +67,8 @@ if not _warmup_app_controller.wait_for_warmup(timeout=60.0):
   stacklevel=2,
  )
 
-# HANG PROTECTION (smart watchdog). Two observed hang chains from
+# HANG PROTECTION (signal-based watchdog). Two observed hang chains
-# e1c8730f and the prior naive watchdog:
+# from e1c8730f and the prior naive watchdog:
 #  1. ThreadPoolExecutor.__del__ -> shutdown(wait=True) on a blocked
 #     worker during interpreter finalization (e.g., the io_pool
 #     created in AppController.__init__ at conftest line 65).
@@ -76,33 +76,50 @@ if not _warmup_app_controller.wait_for_warmup(timeout=60.0):
 #     client.reset_session() (HTTP call to the hook server) or
 #     kill_process_tree(process.pid) / process.wait(timeout=2) waiting
 #     for the sloppy.py subprocess to die on Windows.
-# The naive os._exit(0) at 30s approach CUT OFF BATCHES MID-TEST
+# The naive 30s os._exit(0) approach CUT OFF BATCHES MID-TEST (every
-# (every batch exited at 32.0s exactly, pytest never reached its
+# batch exited at 32.0s exactly). The "60s pytest-hung + 15s grace"
-# FAILURES/summary line) and HID FAILURES (os._exit(0) masked
+# smart watchdog also fired on legitimate long batches because it
-# pytest's non-zero exit code).
+# waited for pytest_unconfigure, which never fires if the conftest's
+# own io_pool is hung in __del__.
 #
-# This smart watchdog only fires when pytest is ACTUALLY HANGING:
+# CORRECT approach: signal-based. Set _pytest_finished_event as
-#  - pytest's pytest_unconfigure hook sets `_pytest_finished_event`
+# SOON AS pytest has logically finished its work, before the
-#    at the very end of the test session, BEFORE interpreter shutdown.
+# shutdown hangs begin. The right hook is pytest_terminal_summary:
-#  - If the event isn't set within 120s, pytest is hung in test
+# it runs after the test session summary is printed (the user can
-#    execution (or import) -> force-exit with code 2 (runner catches
+# see "241 passed, 1 skipped in 32.30s" in the output) but BEFORE
-#    via CalledProcessError).
+# finalization. At that point, the test session is logically done;
-#  - If the event IS set, give 30s for normal interpreter shutdown
+# any further delay is shutdown garbage, not test execution.
-#    (ThreadPoolExecutor.__del__, etc.). If still alive, force-exit.
+#
-# This preserves the FAILURES/summary line for all successful
+# Two hooks set the event for redundancy:
-# batches and only force-exits when something is genuinely stuck.
+#  - pytest_terminal_summary: fires after the summary is printed.
+#    This is the primary signal.
+#  - pytest_unconfigure: fires at the very end, after the summary.
+#    Fallback in case the terminal summary hook isn't reached (e.g.,
+#    pytest crashes mid-summary).
+# After the event fires, give 5s for normal finalization, then
+# os._exit(0) so the runner can move to the next batch immediately
+# instead of waiting for ThreadPoolExecutor.__del__ to unblock
+# (which can take 60+ seconds).
+#
+# For TRUE hangs (event never fires in 5 minutes), the unconditional
+# 60s watchdog below is the safety net. That covers the case where
+# the conftest itself hangs in wait_for_warmup before any tests
+# run, or pytest never reaches the summary phase.
 import threading
 _pytest_finished_event: threading.Event = threading.Event()
 
+def pytest_terminal_summary(terminalreporter: object, exitstatus: int, config: object) -> None:
+ _pytest_finished_event.set()
+
 def pytest_unconfigure(config: object) -> None:
  _pytest_finished_event.set()
 
 def _smart_watchdog_exit() -> None:
+ if not _pytest_finished_event.wait(timeout=300.0):
+  os._exit(2)
  import time
- if not _pytest_finished_event.wait(timeout=60.0):
+ time.sleep(5.0)
-  os._exit(2)
+ os._exit(0)
- if not _pytest_finished_event.wait(timeout=15.0):
-  os._exit(2)
 
 threading.Thread(target=_smart_watchdog_exit, daemon=True, name="conftest-smart-watchdog").start()
 

tests/test_conftest_smart_watchdog.py

@@ -1,35 +1,57 @@
-"""Regression: pytest conftest must install a SMART hang watchdog.
+"""Regression: pytest conftest must install a signal-based hang watchdog.
 
 Two hang chains have been observed when running the test suite:
   1. ThreadPoolExecutor.__del__ -> shutdown(wait=True) on a blocked
-     worker during interpreter finalization.
+     worker during interpreter finalization (e.g., the io_pool
+     created in AppController.__init__ at conftest line ~65).
   2. The session-scoped `live_gui` fixture teardown hanging in
      client.reset_session() (HTTP call to the hook server) or
-     kill_process_tree(process.pid) / process.wait(timeout=2)
+     kill_process_tree(process.pid) / process.wait(timeout=2) waiting
-     waiting for the sloppy.py subprocess to die on Windows.
+     for the sloppy.py subprocess to die on Windows.
 
-The smart watchdog (e1c8730f + 2026-06-07 rework) solves both:
+The signal-based watchdog (2026-06-07 final form) works as follows:
-  - pytest_unconfigure hook sets a flag when the test session is
+  - pytest_terminal_summary hook (primary) sets _pytest_finished_event
-    truly done (BEFORE interpreter finalization).
+    AFTER the test summary has been printed to the terminal. This is
-  - The watchdog waits for that flag with a 120s timeout. If the
+    when the test session is logically done: the user can see
-    flag is never set, pytest is hung in test execution -> exit 2.
+    "241 passed, 1 skipped in 32.30s" but the process is still alive
-  - After the flag is set, give 30s for normal interpreter
+    in shutdown. The signal lets us know "the test work is finished,
-    shutdown. If still alive, the io_pool or live_gui teardown is
+    anything beyond this is shutdown garbage."
-    hung -> exit 2.
+  - pytest_unconfigure hook (fallback) sets the same event in case
-  - Exit code 2 (not 0) so run_tests_batched.py correctly reports
+    pytest_terminal_summary is bypassed (e.g., a crash mid-summary).
-    a failed batch (CalledProcessError).
+  - Smart watchdog waits for the event with a 300s timeout. If not
+    set, the conftest itself is hung in wait_for_warmup or pytest
+    never reached the summary -> exit 2.
+  - After the event fires, give 5s for normal finalization
+    (ThreadPoolExecutor.__del__, etc.), then os._exit(0). This is
+    the right behavior: the test session is done, the user can see
+    the summary, the runner can move to the next batch.
+  - Unconditional 60s watchdog (separate thread) catches the case
+    where conftest hangs before any test runs (so no signal is ever
+    set). Fires os._exit(2) after 60s regardless of state.
 
-This is the CORRECT contract: the previous naive watchdog at e1c8730f
+Why the previous attempts failed:
-(30s os._exit(0)) cut off batches mid-test and hid failures. The
+  - e1c8730f: 30s os._exit(0) cut off batches mid-test. os._exit(0)
-2026-06-07 rework uses pytest_unconfigure as the "done" signal so
+    masked pytest's exit code so failures were hidden.
-the watchdog ONLY fires when something is actually stuck.
+  - 719c5e27: changed to os._exit(2) but the watchdog's daemon
+    thread continued running through pytest's normal shutdown,
+    firing the exit on EVERY batch (even successful ones).
+  - 91b19c90: kept exit 2 but the wait-for-pytest_unconfigure
+    signal never fired when the conftest's own io_pool hung in
+    __del__. The unconditional 90s sledgehammer fired for all
+    batches.
+  - 44b0b5d4: switched to pytest_unconfigure as signal. Still
+    hung because pytest_unconfigure doesn't fire if io_pool hangs.
+  - 2026-06-07 final: pytest_terminal_summary fires AFTER the
+    summary is printed (which the user can verify in the output)
+    but BEFORE the shutdown hangs. This is the right signal.
 
 This test verifies:
   1. The watchdog thread is registered after the conftest loads.
   2. It's a daemon thread (doesn't block pytest's own exit).
-  3. The pytest_unconfigure hook sets the finished flag (so the
+  3. pytest_terminal_summary is used as the primary signal source.
-     watchdog's first wait returns immediately on clean exit).
+  4. pytest_unconfigure is used as a fallback signal source.
-  4. The exit-code-2 contract is documented in the conftest.
+  5. The exit code is 2 for hangs (uncaught Exception type) and 0
+     for clean exits (so the runner correctly reports the batch).
 """
 
 import re
@@ -43,8 +65,8 @@ ROOT = Path(__file__).resolve().parent.parent
 sys.path.insert(0, str(ROOT))
 
 WATCHDOG_NAME = "conftest-smart-watchdog"
-PYTEST_FINISHED_TIMEOUT_SECONDS = 120.0
+PYTEST_FINISHED_TIMEOUT_SECONDS = 300.0
-SHUTDOWN_GRACE_SECONDS = 30.0
+SHUTDOWN_GRACE_SECONDS = 5.0
 
 
 def test_watchdog_thread_registered() -> None:
@@ -67,44 +89,70 @@ def test_watchdog_thread_is_daemon() -> None:
  pytest.fail(f"watchdog thread {WATCHDOG_NAME!r} not found")
 
 
-def test_pytest_unconfigure_sets_finished_flag() -> None:
+def test_terminal_summary_hook_sets_finished_event() -> None:
  """
- Simulate the end-of-session by calling pytest_unconfigure directly.
+ Primary signal: pytest_terminal_summary. This hook fires AFTER
- The watchdog waits for _pytest_finished_event; setting it via the
+ the test session summary is printed, which is the right "session
- hook must release the watchdog's first wait immediately.
+ is logically done" moment for the watchdog.
  """
  conftest_path = Path(__file__).resolve().parent / "conftest.py"
  text = conftest_path.read_text(encoding="utf-8")
- assert "_pytest_finished_event" in text, (
+ assert "pytest_terminal_summary" in text, (
-  f"_pytest_finished_event not found in {conftest_path}; "
+  f"pytest_terminal_summary hook not found in {conftest_path}; "
-  f"smart watchdog signal missing"
+  f"this is the primary signal source"
  )
- assert "pytest_unconfigure" in text, (
+ assert "_pytest_finished_event.set()" in text, (
-  f"pytest_unconfigure hook not found in {conftest_path}; "
+  f"_pytest_finished_event.set() not found in {conftest_path}"
-  f"smart watchdog needs the hook to know when pytest is done"
  )
 
 
-def test_watchdog_uses_non_zero_exit_code() -> None:
+def test_unconfigure_hook_is_fallback_signal() -> None:
  """
- Critical contract: the watchdog must call os._exit(2) (NOT 0) when
+ Fallback signal: pytest_unconfigure. If pytest crashes
- it fires. run_tests_batched.py uses subprocess.run(check=True) and
+ mid-summary, this still fires.
- only reports 'Batch N failed.' on a non-zero exit. Exit 0 would
- hide the hang and silently report a successful batch.
  """
  conftest_path = Path(__file__).resolve().parent / "conftest.py"
  text = conftest_path.read_text(encoding="utf-8")
- matches = re.findall(r"os\._exit\(\s*(\d+)\s*\)", text)
+ assert "def pytest_unconfigure" in text, (
- assert "2" in matches, (
+  f"pytest_unconfigure hook not found in {conftest_path}"
-  f"conftest.py does not call os._exit(2); found exit codes: {matches}. "
+ )
-  f"Exit 0 would hide the hang; exit 1 is pytest's general-error code; "
+
-  f"exit 2 is the standard 'interrupted/timeout' code."
+
+def test_clean_exit_uses_zero_exit_code() -> None:
+ """
+ After the signal fires and the grace period elapses, the watchdog
+ calls os._exit(0). This is the right behavior: the test session
+ is done, the user can see the summary, and the runner should see
+ a clean exit. (If a real test failure happened, pytest's own
+ exit code BEFORE the watchdog would have been 1, but the
+ watchdog forces exit 0 to skip the shutdown hang.)
+
+ The unconditional watchdog (separate thread) still uses exit 2
+ for true hangs.
+ """
+ conftest_path = Path(__file__).resolve().parent / "conftest.py"
+ text = conftest_path.read_text(encoding="utf-8")
+ assert "os._exit(0)" in text, (
+  f"conftest.py should call os._exit(0) after the signal fires + "
+  f"grace elapses; the test session is done at that point and the "
+  f"runner should see a clean exit"
+ )
+
+
+def test_hang_uses_nonzero_exit_code() -> None:
+ """
+ The unconditional watchdog must call os._exit(2) when it fires.
+ """
+ conftest_path = Path(__file__).resolve().parent / "conftest.py"
+ text = conftest_path.read_text(encoding="utf-8")
+ assert "os._exit(2)" in text, (
+  f"conftest.py should call os._exit(2) for true hangs"
  )
 
 
 def test_watchdog_timeouts_documented() -> None:
  """
- Both the 120s pytest-hung timeout and the 30s shutdown-grace timeout
+ Both the 300s pytest-hung timeout and the 5s shutdown-grace timeout
  must be near the documented values. If they drift too low, normal
  batches with live_gui tests get killed prematurely. If too high,
  real hangs waste time.