[rumours86]

UiActor.vfunc_paint() calls child.paint() from JS for every child of uiGroup, including hidden tooltips and menus. Each call marshals a ClutterPaintContext; a profile with a continuously redrawing window put about 9% of Cinnamon's CPU time in gjs_value_from_g_argument alone.

Chain up to StWidget so that Clutter traverses the children natively whenever no mapped actor needs to be skipped. Unmapped actors are already ignored by ClutterActor.paint(), so they do not need to disable this path. Keep the original JS loop when a mapped skipped actor exists. uiGroup has no style, so the background part of StWidget.paint() draws nothing. Picking is unchanged.

The first version checked only whether skip_paint_actors was empty. In a longer desktop session I found 278 children, only six mapped, with two unmapped actors in the skip set; those two actors prevented the fast path on every frame. Also clear skip membership in Chrome._untrackActor(), while the actor is still alive, so subsequent paints do not call is_mapped() on a disposed wrapper.

Initial fast-path measurements, before the mapped-skip extension: muffin 192cd42, Cinnamon c3e2d34, cinnamon-desktop 6c6c48d, nested Xephyr session and a GTK window redrawing at 60 fps. Main-thread userspace instructions (perf stat -e instructions:u -t), divided by 60 frames/s, in A/B/A/B order:

	run 1	run 2
baseline	1.66M	1.68M
native paint fast path	0.77M	0.77M

Validation of the extension:

• All 64 combinations of mapped/skipped state for three children preserve the visible child sequence and choose the expected fast/fallback path in an isolated JS model.
• A CJS/Muffin lifecycle harness using the actual UiActor and Chrome methods covers destroy, explicit untrack followed by destroy, remove followed by destroy, and reparent followed by destroy. Without skip cleanup, each case emits disposed-object criticals on subsequent paints; with cleanup, all four cases leave an empty skip set and take the native path without those criticals. The parent paint is a counter in this harness, so this tests lifetime and branch selection rather than framebuffer output.
• The backport is exercised on Cinnamon 6.6.9 (Mint 22.3); panels, menus, tooltips, OSDs and fullscreen behavior were checked locally. The mapped-skip extension has no separate controlled performance benchmark; the table above measures the original native path.

For the original nested test build only, the background daemon was made optional because this Mint 22.3 installation lacks master's GTK >= 4.18 dependency, and the files/ installation destination was adjusted. Those build changes are not part of this PR. On the updated head (7c0c775), LMDE 7 and Mint 27 builds, Pattern Check and CodeFactor pass. Mint 22 repeats the existing failure in mk-build-deps, during build-dependency installation and before Cinnamon compilation; its log does not identify the individual unmet dependency.

[rumours86]

The Mint 22 job fails before compilation, while mk-build-deps installs the dependencies from debian/control. This PR inherits master's GTK >= 4.18 requirement for the background daemon, but the Mint 22.3 image provides GTK 4.14.5. Unmodified master at c3e2d34 fails at the same step. The LMDE 7 and Mint 27 builds pass.

Correction to my earlier explanation: Codespell runs with only_warn: 1 and succeeds; its typo annotations are not the cause of this failure.