Placeholder tiles are content: printing over the cell removes the tile

[tomlm]

A Unicode-placeholder tile exists because its cell holds the placeholder character; the protocol has no command to remove one tile, because overwriting the cell IS the removal. Tiles were stored as classic Kitty overlay runs, which text never modifies, so anything drawn across a placeholder picture (a dialog, a new screen) kept the picture on top of it permanently.

• New PlacementKind.Placeholder with content semantics: the split that printing performs on Sixel runs now takes placeholder runs too (a one-column run simply goes). Classic placements keep overlay behaviour and z-ordering untouched, pinned by a new test.
• Writing a placeholder cell also clears any stale tile another picture left at that cell, since the cell write itself now splits.
• The test asserting overlay behaviour for placeholders encoded the implementation rather than the protocol (kitty itself removes the tile; image.nvim depends on it) and now asserts the removal, with the reasoning in its remarks.

Perf: the only hot-path-adjacent change is the kind test inside SplitPlacementsAt, which flipped from != Sixel to == Kitty — same single compare, and the method only runs on lines that hold placements.

Found from the consumer side: Consolonia's kitty renderer overwrites occluded placeholder cells exactly as the spec prescribes, and the pictures stayed on top of its dialogs.

🤖 Generated with Claude Code

https://claude.ai/code/session_01CxS1BnzZ6BhuGovMLNtRQH

[github-actions]

Perf comparison — this change, against its base

3 run(s) of each side, alternating on one machine. Allocation is a count and is gated exactly. Time is a measurement, so its gate is derived from the spread this job just observed in itself rather than fixed in advance.

corpus	bytes/char	gen0/Mchar	ns/char	Δ time	noise	gate
scroll-ascii	0.00 → 0.00	0.00 → 0.00	3.50 → 3.49	-0.5%	±1%	4%
sgr-churn	0.00 → 0.00	0.00 → 0.00	9.35 → 9.29	-0.6%	±1%	4%
truecolor	0.00 → 0.00	0.00 → 0.00	9.87 → 9.99	+1.2%	±2%	5%
alt-redraw	0.00 → 0.00	0.00 → 0.00	14.75 → 14.72	-0.2%	±5%	15%
unicode	7.66 → 7.66	0.45 → 0.45	35.80 → 35.49	-0.9%	±4%	11%
flood	0.00 → 0.00	0.00 → 0.00	105.14 → 105.01	-0.1%	±0%	4%

Each corpus is gated at max(4%, 3 × its own noise). A wide noise column means this runner was busy and the timing half of the table should be read as advisory; the allocation half is exact either way.

assemblies measured

• base: XTerm.NET 2.0.1.0 mvid:66758d33-46a9-4d79-a288-5482acbfe77f
• head: XTerm.NET 2.0.1.0 mvid:dc29f018-2613-4cb5-9e00-4a1b11a3e67e

---

Perf comparison — cumulative, everything since 2.0.0

3 run(s) of each side, alternating on one machine. Allocation is a count and is gated exactly. Time is a measurement, so its gate is derived from the spread this job just observed in itself rather than fixed in advance.

corpus	bytes/char	gen0/Mchar	ns/char	Δ time	noise	gate
scroll-ascii	0.00 → 0.00	0.00 → 0.00	3.49 → 3.49	-0.0%	±1%	5%
sgr-churn	0.00 → 0.00	0.00 → 0.00	9.28 → 9.29	+0.1%	±1%	5%
truecolor	0.00 → 0.00	0.00 → 0.00	9.89 → 9.99	+1.1%	±2%	5%
alt-redraw	0.00 → 0.00	0.00 → 0.00	14.73 → 14.72	-0.1%	±1%	5%
unicode	7.66 → 7.66	0.45 → 0.45	35.60 → 35.49	-0.3%	±4%	11%
flood	0.00 → 0.00	0.00 → 0.00	105.11 → 105.01	-0.1%	±0%	5%

Each corpus is gated at max(5%, 3 × its own noise). A wide noise column means this runner was busy and the timing half of the table should be read as advisory; the allocation half is exact either way.

assemblies measured

• base: XTerm.NET 2.0.0.0 mvid:a26a50e5-3343-4b31-bada-ce2a03923fe4
• head: XTerm.NET 2.0.1.0 mvid:dc29f018-2613-4cb5-9e00-4a1b11a3e67e