diff options
Diffstat (limited to 'working/eat-sixel-patch')
| -rw-r--r-- | working/eat-sixel-patch/eat-xtwinops.patch | 46 | ||||
| -rw-r--r-- | working/eat-sixel-patch/rulesets-companion-note.org | 5 | ||||
| -rw-r--r-- | working/eat-sixel-patch/takuzu-original-bug-report.org | 17 |
3 files changed, 68 insertions, 0 deletions
diff --git a/working/eat-sixel-patch/eat-xtwinops.patch b/working/eat-sixel-patch/eat-xtwinops.patch new file mode 100644 index 00000000..4347d729 --- /dev/null +++ b/working/eat-sixel-patch/eat-xtwinops.patch @@ -0,0 +1,46 @@ +--- a/eat.el ++++ b/eat.el +@@ -2809,6 +2809,33 @@ + (funcall (eat--t-term-input-fn eat--t-term) eat--t-term + "\e[>0;0;0c"))))) + ++(defun eat--t-send-window-size-report (n) ++ "Respond to XTWINOPS window size report request N. ++ ++N is 14 (text area size in pixels), 16 (cell size in pixels) or 18 ++\(text area size in characters). Other operations are ignored. ++Multiplexers like tmux send \\='CSI 14 t\\=' to discover the cell ++pixel size they need before they will emit Sixel to a terminal." ++ (let ((disp (eat--t-term-display eat--t-term))) ++ (pcase n ++ (14 ++ (funcall (eat--t-term-input-fn eat--t-term) eat--t-term ++ (format "\e[4;%i;%it" ++ (* (eat--t-disp-height disp) ++ (eat--t-term-char-height eat--t-term)) ++ (* (eat--t-disp-width disp) ++ (eat--t-term-char-width eat--t-term))))) ++ (16 ++ (funcall (eat--t-term-input-fn eat--t-term) eat--t-term ++ (format "\e[6;%i;%it" ++ (eat--t-term-char-height eat--t-term) ++ (eat--t-term-char-width eat--t-term)))) ++ (18 ++ (funcall (eat--t-term-input-fn eat--t-term) eat--t-term ++ (format "\e[8;%i;%it" ++ (eat--t-disp-height disp) ++ (eat--t-disp-width disp))))))) ++ + (defun eat--t-send-graphics-attrs (attr operation) + "Send graphics attributes. + +@@ -3632,6 +3659,9 @@ + ;; CSI s. + (`((?s) nil nil) + (eat--t-save-cur)) ++ ;; CSI <n> t. ++ (`((?t) nil ((,n))) ++ (eat--t-send-window-size-report n)) + ;; CSI u. + (`((?u) nil nil) + (eat--t-restore-cur))))))) diff --git a/working/eat-sixel-patch/rulesets-companion-note.org b/working/eat-sixel-patch/rulesets-companion-note.org new file mode 100644 index 00000000..2df33273 --- /dev/null +++ b/working/eat-sixel-patch/rulesets-companion-note.org @@ -0,0 +1,5 @@ +#+TITLE: EAT XTWINOPS patch companion note +#+SOURCE: from rulesets +#+DATE: 2026-07-13 10:58:54 -0500 + +EAT XTWINOPS patch (companion file eat-xtwinops.patch, sent separately just now): verified live on ratio 2026-07-13, this makes sixel images render AND persist in tmux inside EAT. Root cause chain: tmux 3.7b has native sixel but refuses to transmit unless it knows the client's cell pixel size; it asks via CSI 14 t (XTWINOPS), which EAT 0.9.4 silently ignores (its CSI dispatch has no 't' case). The patch adds eat--t-send-window-size-report answering CSI 14/16/18 t from eat--t-term-char-width/height and the display dims — same fields EAT's own XTSMGRAPHICS reply already uses. With it, tmux learns the geometry at client attach, ingests raw sixel into its grid, and re-emits on every redraw — images survive window switches, scrolling, and resizing (Craig verified visually). Needs on the emacsd side: (1) a durable home for the patch — the elpa file is version-pinned, so likely a load-path shadow or an advice-free vendored fix, your call, plus wiring so both daily drivers load it (velox verified ready: same tmux build, imagemagick present); (2) written test-first per testing.md when productionizing — this was a timeboxed spike, treat the diff as evidence not deliverable; (3) strongly consider sending it upstream to EAT (codeberg akib/emacs-eat) so 0.9.5+ ships it stock. tmux side is separate (two conf lines, handoff sent to dotfiles). Ask rulesets for the full investigation log if useful. diff --git a/working/eat-sixel-patch/takuzu-original-bug-report.org b/working/eat-sixel-patch/takuzu-original-bug-report.org new file mode 100644 index 00000000..02107a88 --- /dev/null +++ b/working/eat-sixel-patch/takuzu-original-bug-report.org @@ -0,0 +1,17 @@ +#+TITLE: Bug/investigate: images from Claude Code don't display in th +#+SOURCE: from takuzu +#+DATE: 2026-07-11 14:41:39 -0500 + +Bug/investigate: images from Claude Code don't display in the EAT terminal. + +Context: running Claude Code inside EAT (eat 0.9.4 from elpa). When the agent posts an image to the terminal, nothing renders — Craig can't see it. + +Finding: EAT *does* have Sixel image support. eat.el defines eat-sixel-scale, eat-sixel-aspect-ratio, eat-sixel-render-formats, and a full Sixel decode/render path (sixel-buffer, sixel-palette, sixel-render-format, etc.). So the capability exists; this isn't a plain 'EAT can't do images' limitation. + +Likely causes to check: +1. Protocol mismatch: Claude Code CLI probably emits the kitty graphics protocol or the iTerm2 inline-image protocol (OSC 1337), NOT Sixel. EAT renders Sixel only, so a kitty/iTerm image sequence is dropped. Confirm which protocol Claude Code uses (env like TERM, or its terminal-capability detection) and whether it can be told to emit Sixel. +2. Sixel render-format availability: eat-sixel-render-formats lists preferred render backends; if none is available in this Emacs build, Sixel silently won't display. Test EAT with a known-good sixel image (e.g. img2sixel output) to confirm EAT-side rendering works at all. + +Repro: in an EAT session, have Claude Code (or any tool) post an image; observe nothing renders. Next step: identify the emitted escape sequence (capture raw bytes) to confirm protocol. + +From the takuzu session, 2026-07-11 — surfaced because the agent kept posting game screenshots the terminal couldn't show. |