[skbarber]

Stacked PR, step 2a of the panels arc. Base: feat/geecs-panels (#1131), which is based on #1130. The maintainer asked to finish the arc before merging, so the stack merges in order at the end: #1130 → #1131 → this → the next steps. Don't delete a base branch while a child is open; GitHub auto-closes the child.

Feedback from looking at #1131: make Live collapsible, give the image the panel, add zoom, drag and log scale.

What

• Viewer. The image fills the panel, taking most of the window's height:
  • Wheel zooms at the cursor, drag pans, double-click or Fit fits, 1:1 shows camera pixels, unsmoothed when magnified.
  • The colour-mapped frame is kept separate from the view (toView/toImage), so 2b's crosshairs and ROI box zoom and pan with the image.
• Scale: linear, square root or log, in the same lookup table as black/white. It stretches the 8-bit stream and adds no dynamic range (said on the page).
• Live strip: a native <details class="disc"> (rung 0). Folded, its summary keeps centroid and max; open/closed is remembered per camera.
• Frames only when the camera makes one. FrameSource reuses a frame exactly while MJPG1:ArrayCounter_RBV stands still and sends it as X-Frame-Counter; the page skips decoding a frame it already shows. A 1 Hz camera polled at 10 Hz is now one upstream fetch and one draw a second (maintainer question on host load).
• Fixed while measuring:
  • The reuse window was timed from the fetch's reply. A poll at the cap's period then landed inside it behind a ~55 ms fetch and got the old frame back about a third of the time: 5.5 fps shown at a 10 fps cap.
  • The window is now timed from the request, and the counter decides whenever it's known. The page also fetches the next frame while drawing the last.
  • Measured in-process: 0 repeats in 24 polls (was 10). In the browser: 9.5–10 fps.

398 lines added, 89 removed, mostly camera.js.

Tests

• GeecsPanels: 53 passed. New tests: an unchanged counter means no new fetch (proven to bite); no counter falls back to the window; the window is timed from the request; a changed counter refetches inside the window; the frame route sends X-Frame-Counter.
• GeecsWebTheme: 198 passed. Lint OK.

Hardware verification (done)

Against ADTEST_CAM1 from a Mac over the VPN:

• 9.5–10 fps at the 10 fps cap.
• 467% zoom on a log scale, and the folded Live summary. Screenshots were reviewed; both rooms are dark.
• OWED: a lit scene, and a camera at 1 Hz, to see the fetch rate follow the camera.

🤖 Generated with Claude Code

[skbarber]

Adversarial review (fresh-context reviewer, two rounds)

#	Finding	Disposition
F1	With the counter known there was no fetch floor: a fast camera with N out-of-step viewers cost the IOC N× fetches, which broke "N viewers cost one stream".	Fixed in 051a5b4a0: the min_interval floor applies first, whatever the counter says. test_every_viewer_shares_one_fetch_per_floor_interval (5 viewers, a 100 Hz camera: at most 21 fetches/s) fails with the floor removed. Confirmed.
F2	A frame tagged by a counter that arrived before its JPEG would be served stale for a whole camera period.	Fixed (safety net) in 051a5b4a0. ffmpegServer stores the JPEG (ffmpegServer.cpp:569) before callParamCallbacks (:586), so the counter should follow its JPEG. Even so, a frame is now reused at most max_reuse (1 s) under an unchanged counter, and each fetch gets its own X-Frame-Tag, so the page redraws a refetch. Pinned. Confirmed.
F3	A change of devicePixelRatio (window moved between screens) left the image mis-scaled and the cursor readout wrong.	Fixed: a matchMedia resolution listener, re-armed after each change. Confirmed.
F4	A frame that failed to decode was an unhandled rejection: no error shown, "live" chip stayed up.	Fixed: the error is shown and the frame retried. Confirmed.
F5	1:1 couldn't reach 100% on a frame smaller than half the view.	Fixed: the floor is min(1, fit/2). Confirmed.
R2-1	(round 2) The fps counted safety refetches, so a stopped camera read about 0.5 fps.	Fixed in the head commit: only a new camera counter counts.
R2-2	(round 2) The decode-error box flickered when every frame failed.	Fixed: it hides only after a frame draws.
R2-3	(round 2) Two comments (__main__, the feed.py docstring) described the old reuse rule.	Fixed.

Round-2 items were low severity and fixed without a third round (review stopping rule). Noted: for a camera at 1 Hz or faster, the 1 s net doesn't shorten a stale frame below one camera period; it's a backstop, not a guarantee.

After the fixes: GeecsPanels 54 passed, GeecsWebTheme 198 passed, lint OK. Live: 9.5–10 fps at the 10 fps cap.

[skbarber]

Reviewed this PR in the #1130–#1135 stack, including the existing review fixes. No additional actionable findings in this PR. Validation at stack head dc8ac07: panel suite plus Core EPICS bootstrap/light-import tests, 143 passed (including JavaScript syntax checks). Targeted offline probes reproduced the findings on #1134/#1135. No hardware writes or live camera testing performed.

[skbarber]

Closed by GitHub when its base branch was deleted; its base cannot be changed while closed. Superseded by a new PR from the same head commit (feat/panels-viewer → master). The review record stays here.