A Sixel replaces the Sixel runs it lands on

[tomlm]

Problem

PlaceImage added a run to each line a Sixel covered but never removed the Sixel runs already under it. Printing a character over a picture correctly splits and replaces that part of it; a picture over a picture did not.

An application that repaints small regions with Sixels (Avalonia.SixelBuffer sends a tiny sixel for every cursor move and hover change) therefore left every earlier picture piece stacked on the line. In the alternate screen, lines never die, so the per-line run list, the retained TerminalImages, the view's per-image bitmaps, and the per-frame draw work all grew without bound. Avalonia.Terminal got slower by the minute; Windows Terminal, which replaces covered cells, did not.

Fix

Before a Sixel run is added to a line, the Sixel runs over its columns are split with SplitPlacementsOver, exactly as a printed character does. Fully covered runs disappear; partially covered ones keep their uncovered fragments. Kitty placements are overlays ordered by z-index with their own delete commands and are left alone.

Semantic note

A Sixel sent with background select 1 (transparent unset pixels) now replaces what it covers rather than showing the older picture through its holes. The DEC behaviour would composite; this is consistent with XTerm.NET already treating Sixel as content that printing replaces.

Tests

Three new cases in ImageCellLifetimeTests: the same picture drawn twice leaves one run and one image per row; a one-cell picture over part of a wider one keeps the uncovered fragment; fifty repaints of one cell leave two runs per row rather than fifty-two. Full suite: 2301 tests pass.

🤖 Generated with Claude Code

[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	3.38 → 3.40	+0.6%	±2%	7%
sgr-churn	0.00 → 0.00	0.00 → 0.00	8.97 → 8.98	+0.1%	±2%	6%
truecolor	0.00 → 0.00	0.00 → 0.00	9.94 → 9.97	+0.4%	±1%	4%
alt-redraw	0.00 → 0.00	0.00 → 0.00	14.34 → 14.46	+0.9%	±2%	5%
unicode	7.66 → 7.66	0.45 → 0.45	35.11 → 35.22	+0.3%	±9%	28%
flood	0.00 → 0.00	0.00 → 0.00	95.41 → 96.46	+1.1%	±1%	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 2.0.7.0 mvid:41656e21-75ec-4936-b683-6fb2b0384560
• head: XTerm.NET 2.0.7.0 mvid:96b05f35-feca-4e32-a9a0-26f1c69dae29

---

Perf comparison — cumulative, everything since 2.0.7

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	3.39 → 3.40	+0.3%	±2%	5%
sgr-churn	0.00 → 0.00	0.00 → 0.00	8.98 → 8.98	+0.0%	±3%	9%
truecolor	0.00 → 0.00	0.00 → 0.00	9.94 → 9.97	+0.3%	±3%	10%
alt-redraw	0.00 → 0.00	0.00 → 0.00	14.51 → 14.46	-0.4%	±2%	5%
unicode	7.66 → 7.66	0.45 → 0.45	35.50 → 35.22	-0.8%	±2%	6%
flood	0.00 → 0.00	0.00 → 0.00	95.95 → 96.46	+0.5%	±1%	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 2.0.7.0 mvid:2609fe33-3671-4fb3-9e45-c3e43c231eb3
• head: XTerm.NET 2.0.7.0 mvid:96b05f35-feca-4e32-a9a0-26f1c69dae29