[eric] browser: split the opener-wait tests out at the file cap

backend/tests/test_browser_send_script.py

@@ -286,29 +286,3 @@ async def test_login_wall_url_declines_before_any_fill():
                                  current_url="https://x.com/i/flow/login")
     assert r is None
     assert not calls["clicks"]
-
-
-@pytest.mark.asyncio
-async def test_opener_wait_stops_once_the_surface_settles():
-    """A budget spent waiting on a page that has stopped changing is pure cost.
-
-    Fixed budgets were wrong in both directions: 1.8s missed gmail and linkedin, 5.3s still missed
-    a cold gmail compose window, and just raising the number taxes every run that was never going
-    to work. The stop condition is now page stability, so an opener that leads nowhere gives up as
-    soon as two reads match instead of sleeping out the rest."""
-    # Opener clicked, then the same composer-less page twice: identical reads = nothing coming.
-    r, calls = await run(TASK, PROFILE, [PROFILE, PROFILE, PROFILE, PROFILE, PROFILE, PROFILE])
-    assert r is None
-    # 6 sleeps were budgeted; settling must cut the reads well short of consuming them all.
-    assert calls["list"] < 9, f"kept polling a settled page: {calls['list']} reads"
-
-
-@pytest.mark.asyncio
-async def test_opener_wait_still_catches_a_late_composer():
-    """The other direction: a surface that is still mounting must not be abandoned early, which is
-    the failure that made gmail intermittent."""
-    # Changing reads (so never 'settled'), composer only on the 4th look.
-    late = [PROFILE, PROFILE, PROFILE, PROFILE + "\n[99]<button \"x\">",
-            COMPOSER_EMPTY, COMPOSER_FILLED, COMPOSER_SENT]
-    r, calls = await run(TASK, PROFILE, late)
-    assert r is not None and r["sent"] is True

backend/tests/test_opener_wait_settles.py

@@ -0,0 +1,37 @@
+"""When to stop waiting for a compose surface that was just opened.
+
+Split out of test_browser_send_script.py, which hit the 300-line cap. The rule under test is its
+own idea and deserves its own file: a fixed wait was wrong in both directions (1.8s missed gmail
+and linkedin; 5.3s still missed a cold gmail compose window; raising it further taxes every run
+that was never going to succeed), so the stop condition is page stability instead of a clock.
+"""
+import pytest
+
+from backend.tests.test_browser_send_script import (
+    COMPOSER_EMPTY, COMPOSER_FILLED, COMPOSER_SENT, PROFILE, TASK, run,
+)
+
+@pytest.mark.asyncio
+async def test_opener_wait_stops_once_the_surface_settles():
+    """A budget spent waiting on a page that has stopped changing is pure cost.
+
+    Fixed budgets were wrong in both directions: 1.8s missed gmail and linkedin, 5.3s still missed
+    a cold gmail compose window, and just raising the number taxes every run that was never going
+    to work. The stop condition is now page stability, so an opener that leads nowhere gives up as
+    soon as two reads match instead of sleeping out the rest."""
+    # Opener clicked, then the same composer-less page twice: identical reads = nothing coming.
+    r, calls = await run(TASK, PROFILE, [PROFILE, PROFILE, PROFILE, PROFILE, PROFILE, PROFILE])
+    assert r is None
+    # 6 sleeps were budgeted; settling must cut the reads well short of consuming them all.
+    assert calls["list"] < 9, f"kept polling a settled page: {calls['list']} reads"
+
+
+@pytest.mark.asyncio
+async def test_opener_wait_still_catches_a_late_composer():
+    """The other direction: a surface that is still mounting must not be abandoned early, which is
+    the failure that made gmail intermittent."""
+    # Changing reads (so never 'settled'), composer only on the 4th look.
+    late = [PROFILE, PROFILE, PROFILE, PROFILE + "\n[99]<button \"x\">",
+            COMPOSER_EMPTY, COMPOSER_FILLED, COMPOSER_SENT]
+    r, calls = await run(TASK, PROFILE, late)
+    assert r is not None and r["sent"] is True