#23422 fix(next perf gate): an unlit window skips the timed tab switch and new workspace counts

Summary

The perf gate's tabs and new_workspaces scenarios fail debug.tab_switch timed 0 of 40 switches and debug.new_workspace timed 0 of 50 new workspace rows on the PR scheme lane (#23308 run 39584287796, "Perf counter gate"). That runner's window draws no frames (the frame probe saw 8 in 1 s, refresh_ms 0), and the app times each switch and row against the display link: SwitchFrameTimer records nothing without a refresh interval. Every other frame-dependent check is gated by frames_ok and takes the probe's skip there; these two counts were not (f06e1d7 for tabs, 5809c6c / #21825 for new workspaces). Both were already red on 10-19; the gate did not really run until f66b984e5217 stopped a launch-time red from skipping it.

Gate.timed_count makes "no rows on a window that draws no frames" that same skip. Nothing else changes:

• With --frames required (the GUI host), no frames is still a failure.
• A lit window that times fewer rows than it should still fails.
• No budget changes.

Testing

python3 tests/test_cmux_next_perf_gate_unlit.py (new, registered in linux-guard) drives both scenarios with a fake app whose timers record nothing. On huge-blue:

• Red commit c2667e665594: 2 of 3 fail. With --frames auto, the failures were [tabs] debug.tab_switch timed 0 of 40 switches and [new_workspaces] debug.new_workspace timed 0 of 50 new workspace rows rather than the skip.
• Fix commit 76aa0d2affce: 3 of 3 pass.

Observed / Not run

• Observed: the red/green unit runs above, and the two failures with frames_probe 8 frames in #23308's perf-gate.json.
• Not run: the gate against a real app. CI's scheme lane runs it.

Changelog

none

Agent session: cosmic-gray

🤖 Generated with Claude Code

---

^{Need help on this PR? Tag @codesmith-bot with what you need.}

---

Summary by cubic

Makes the perf gate skip debug.tab_switch and debug.new_workspace timed counts when the window draws no frames, instead of failing with "timed 0 of N". Those two scenarios previously failed on the unlit CI runner while every other frame-dependent check already took the frame probe's skip.

• Adds Gate.timed_count to treat no rows on a window that draws no frames as the skip; --frames required still fails on no frames, and a lit window that times fewer counts still fails.
• Adds tests/test_cmux_next_perf_gate_unlit.py covering the skip, the required-frames failure, and the lit-window failure, and registers it in the linux-guard lane.