http://git.cashmere.rs/kitty-graphics/atom/?h=v0.4.0 2026-04-04T01:57:52Z 2026-04-04T01:57:52Z cashmere cashmere@autistici.org 2026-04-04T01:57:52Z urn:sha1:7e4faffa32d6f0ef885fceb0a9ecd93b3fe5c43f Add CHANGELOG.org with full 0.4.0 release notes covering sixel support, text sizing protocol, agent-shell, and bug fixes. Replace demo GIFs with backend-specific showcases: kitty.gif (truecolor) and foot.gif (sixel). 2026-04-03T16:13:19Z cashmere cashmere@autistici.org 2026-04-03T16:13:19Z urn:sha1:1d477511816c63e9d2ddb41a68b0f23836c9d07b Spawn mpv with --vo=kitty to paint video frames into a terminal region that Emacs controls. Emacs reserves overlay space and coordinates mpv's screen position via JSON IPC on scroll/resize. New public API: - kitty-gfx-play-video: play video file inline at point - kitty-gfx-stop-video: stop playback and cleanup - kitty-gfx-toggle-video: pause/resume Requires: mpv with --vo=kitty support, Kitty terminal. Gated behind kitty-gfx-enable-video defcustom (nil by default). 2026-04-02T21:37:13Z cashmere cashmere@autistici.org 2026-04-02T21:37:13Z urn:sha1:dd1e18c9b1a06e6eb0f1c8aef94871f9c9cf2889 Remove hardcoded column-1 guard that broke heading placement with display-line-numbers-mode and org-indent-mode. Headings were never placed (placed=0 hidden=37) because posn-at-point returns terminal columns offset by body-left, not buffer column 1. CPU optimizations reduce position queries by 86% and heading placements by 76%: - Early window-range exit skips posn-at-point for off-screen overlays - Emitted flag prevents redundant OSC 66 re-emission - Single fold-check instead of duplicate calls - Strip text properties to prevent org-modern payload corruption Add preview mode that temporarily disables conflicting minor modes (org-modern, org-appear, org-indent, visual-line, olivetti) and unfolds all headings for clean OSC 66 rendering. Modes are restored when heading sizes are toggled off. Add row-based collision detection to prevent overlapping multicell blocks when adjacent headings occupy the same terminal rows. Deferred display property shows heading text as fallback until first successful OSC 66 emission, preventing invisible headings. 2026-03-29T07:07:06Z cashmere cashmere@autistici.org 2026-03-29T07:07:06Z urn:sha1:7691a94d26190324d3f7b6ff22068fa4fc57c121 Close gaps identified in Sixel vs Kitty audit: - image-mode: mode-name now shows Image[Sixel] or Image[Kitty] based on active backend (was hardcoded to Image[Kitty]) - LaTeX preview: docstring updated to reference kitty-graphics package name, not Kitty protocol specifically - sixel-cleanup-all: now erases visible Sixel images from terminal cells before cleaning disk/memory state (prevents pixel artifacts on mode disable) - sixel-place: LRU eviction for .six temp files, capped at kitty-gfx-cache-size (default 64), oldest evicted on overflow Remaining by-design differences (protocol limitations, not bugs): - Sixel re-emits raw pixel data on scroll (no placement IDs) - Sixel erases by overwriting with spaces (no protocol-level delete) - Sixel detection is env-var based (no DA1 query yet) 2026-03-29T06:39:08Z cashmere cashmere@autistici.org 2026-03-29T06:39:08Z urn:sha1:97f0dbc31d9e7eed6ac3ae3300968a3d231270fe When org folds headings (S-TAB overview mode), the newline after each heading becomes invisible, causing multiple headings to appear on the same visual line. posn-at-point then reports col > 1 for headings that follow the fold ellipsis of a previous heading. Placing scaled text at col > 1 causes horizontal drift — headings rendered mid-line overlapping with fold ellipsis and each other. Fix: org headings always start at buffer column 0. If the computed terminal column is not 1, the heading is in a folded context where rendering at scale would be wrong. Skip it (treat as not visible), erase any previous placement, and let the refresh cycle re-place it when unfolded. Root cause identified via debug log analysis showing: Before fold: all headings at col=1 (correct) After S-TAB: headings at col=21/24 on same row (wrong) 2026-03-29T06:31:03Z cashmere cashmere@autistici.org 2026-03-29T06:31:03Z urn:sha1:11da531e49fca10ac7dab149ac9790d0c5e76bb9 Replace space-writing erase with ECH escape (\e[NX) which erases characters at the cursor without moving it — more efficient and cleaner terminal state. Changes: - kitty-gfx--erase-heading-at: new low-level function using ECH, disables DECAWM during erase to prevent wrapping artifacts, erases all rows of multicell blocks (not just topmost) - kitty-gfx--erase-heading: now delegates to erase-heading-at - kitty-gfx--place-heading: pre-erases target area before emitting OSC 66, preventing ghost artifacts from partial overwrites Adapted from mdfried's "erase-character dance" pattern which erases the full multicell area using ECH before every OSC 66 emission. 2026-03-29T06:28:44Z cashmere cashmere@autistici.org 2026-03-29T06:28:44Z urn:sha1:807a7ae4fa4741733f4c1d05b0c42c723439a0cc Comprehensive analysis of 5 OSC 66 implementations: - mdfried (Rust): erase-character dance, fractional scaling with s=2 base, unicode chunking for wide chars, bug history - neovim#32539: architecture analysis confirming our overlay approach - render-markdown.nvim#560: ecosystem blocked on neovim core - opentui (TS+Zig): separator convention, ST vs BEL terminator - foot!1927 (C): width-only impl, variation selector bugs Key actionable findings: ECH erase, s=2 fractional tiers, wide-char chunking, pre-erase before emission. 2026-03-29T05:07:11Z cashmere cashmere@autistici.org 2026-03-29T05:07:11Z urn:sha1:71d5c4402248e45eb2c713830dd012d3bd4c9d13 - Version bump 0.3.0 → 0.4.0 (text sizing feature complete) - kitty-gfx-debug-state: show text-sizing-support status in header, distinguish heading vs image overlays, display heading-specific properties (text, scale, level, cell-s) instead of image properties (id, pid, file) 2026-03-29T05:06:09Z cashmere cashmere@autistici.org 2026-03-29T05:06:09Z urn:sha1:c6e4967b52eacb13360e649cddbaf9afcd6213dc Add heading-aware org-cycle handling: - kitty-gfx--nuke-headings: erases ALL heading multicell blocks in the current buffer, clearing their cached positions. This is the "nuke" phase — a clean slate for the terminal. - kitty-gfx--on-org-cycle: now handles both image and heading overlays. Images get their placements deleted (existing logic), headings get their multicell blocks erased (nuke phase). The subsequent scheduled refresh is the "repaint" phase: kitty-gfx--overlay-screen-pos checks in-folded-region-p, so folded headings return nil and won't be re-emitted, while visible headings get fresh OSC 66 emission at correct positions. The nuke-and-repaint strategy avoids incremental erase complexity (which failed in two prior prototypes) by simply erasing everything and letting the refresh cycle figure out what's still visible. 2026-03-29T04:54:09Z cashmere cashmere@autistici.org 2026-03-29T04:54:09Z urn:sha1:e5ee5bb0e498c66b473e7eecc8c31e29f503d85c Fix scroll/buffer-switch behavior for heading overlays: - on-window-change: preserve heading overlay position caches (images clear theirs for re-placement, but headings need old→new comparison to erase multicell blocks properly) - on-buffer-change: erase heading multicell blocks when their buffer becomes hidden, preventing ghost artifacts if the new buffer's content doesn't fully overwrite the heading cells - on-buffer-change: preserve heading caches for visible buffers (same rationale as on-window-change) The existing two-phase refresh already handles scroll correctly: phase 1 detects position changes and erases moved headings, phase 2 re-emits all visible headings. These fixes ensure the cache state is correct for that comparison to work.