Suppress cursor when cell face background matches cursor color

[devsunb]

Summary

Hide the Emacs cursor when its color collides with the cursor cell's face background, so terminal apps that draw their own reverse-video cursor (Claude Code, Gemini CLI, etc.) stay visible.

Why

Emacs's GUI ports (X11, mac-port, w32, pgtk) share a set_cursor_gc distinct-swap fallback that swaps the cursor's fg/bg with the cell's fg/bg whenever they match (xterm.c:8017 and the mac-port/w32/pgtk equivalents). It's meant to keep the cursor visible against a matching cell, but it backfires for apps that draw their input cursor via SGR reverse-video: on a reverse-video cell the swap reproduces the colors of the surrounding non-inverse cells, so the Emacs box cursor becomes visually indistinguishable from normal text.

Fix

Detect the collision and force cursor-type off so the cell's own reverse-video face represents the cursor:

(defun ghostel--cursor-cell-conflicts-p ()
  (when-let* ((pos ghostel--cursor-char-pos)
              ((< pos (point-max)))
              (cell-bg (ghostel--cell-face-background pos))
              (cursor-color (frame-parameter nil 'cursor-color))
              (cell-rgb (color-values cell-bg))
              (cursor-rgb (color-values cursor-color)))
    (equal cell-rgb cursor-rgb)))

Suppression runs from a buffer-local pre-redisplay-functions hook rather than from ghostel--set-cursor-style. A one-time assignment there would not stick, since cursor-type is restored to its default between module render passes; the hook reapplies the suppression just before each redisplay. It also keeps the cursor off when the terminal sends DECTCEM hide (\e[?25l), tracked in ghostel--cursor-visible.

Screenshots

Before

After

[dakra]

@devsunb Thanks. Can you open an issue with more info and exact steps how to reproduce it?

Neither claude nor gemini have any problems here. I set my focused cursor color exactly the same as the default fg, but everything works as expected.

Also I'm not sure if a pre-redisplay-function that always runs and hides the cursor is the best solution.
So I would prefer if I can first reproduce the issue.

[devsunb]

I found the root cause. The cursor was being redrawn by evil-mode, not by anything in this package. Sorry for the noise.

For future reference, I had been disabling evil in ghostel buffers like this:

(with-eval-after-load 'evil
  (evil-set-initial-state 'ghostel-mode 'emacs))

But emacs state still keeps evil active, so the cursor was being redrawn. The fix was to exclude the buffer from evil entirely:

(with-eval-after-load 'evil
  (add-to-list 'evil-buffer-regexps '("\\`\\*ghostel" . nil)))

Thanks for the patient review and for creating this package.