diff options
| author | user <user@clank> | 2026-09-29 14:25:34 +0200 |
|---|---|---|
| committer | user <user@clank> | 2026-09-29 14:25:34 +0200 |
| commit | 3760c4e99aff06d09edb0575d7596d0607298e71 (patch) | |
| tree | c6e1a676cdccb449e1f59309dd622bd14a7bb544 /src | |
| parent | skill: the verbs that shipped since, and the bare form's other shape (diff) | |
| download | slosh-3760c4e99aff06d09edb0575d7596d0607298e71.tar.gz slosh-3760c4e99aff06d09edb0575d7596d0607298e71.tar.xz slosh-3760c4e99aff06d09edb0575d7596d0607298e71.zip | |
mouse: encode pixel reports against the cell size we advertise
A pane is told how big a cell is twice -- in mode 2048's resize report and
in the XTWINOPS answers -- and the mouse encoder was told something else: a
1x1 cell, so that cell coordinates and "pixels" were the same number.
That held until a program turned on SGR-pixel mouse (mode 1016) by itself.
Textual does, the moment in-band resize works, and then divides every
coordinate it is sent by the pixels-per-cell we advertised: a click 30 cells
across and 9 down arrived as (3, 0), so every click in a Textual app landed
in the top-left corner.
Hand the encoder the cell size the pane already has, and put the pointer in
the middle of its cell. A size that contradicts what the same pane was told
about itself is the lie, not the precision.
Diffstat (limited to 'src')
| -rw-r--r-- | src/pane.c | 28 |
1 files changed, 20 insertions, 8 deletions
@@ -846,17 +846,29 @@ void pane_send_key(pane_t *p, const input_event_t *ev) { void pane_send_mouse(pane_t *p, const input_event_t *ev) { ghostty_mouse_encoder_setopt_from_terminal(p->menc, p->term); - /* The encoder works in surface pixels. We have no pixels, so we declare a - * 1x1-pixel cell: cell coordinates and "pixels" become the same number, and - * SGR-pixel mode degrades to cell precision instead of lying. */ + /* The encoder works in surface pixels, so we hand it the cell size we + * already tell the pane about (mode 2048's resize reports and the XTWINOPS + * answers in on_size_report) and put the pointer in the middle of its cell. + * + * It used to declare a 1x1 cell, so that cell coordinates and "pixels" were + * the same number -- which was fine until a program turned on SGR-pixel mouse + * (mode 1016). Textual does that by itself the moment in-band resize works, + * and then divides every reported coordinate by the pixels-per-cell it got + * from OUR resize report (8x16): a click 30 cells across and 9 down arrived + * as (3, 0), so every click in a Textual app landed in the top-left corner. + * A size that contradicts what the same pane was told about itself is the + * lie, not the precision. */ + uint16_t cw = p->cell_w ? p->cell_w : 1; + uint16_t ch = p->cell_h ? p->cell_h : 1; GhosttyMouseEncoderSize size = GHOSTTY_INIT_SIZED(GhosttyMouseEncoderSize); - size.screen_width = p->cols; - size.screen_height = p->rows_n; - size.cell_width = 1; - size.cell_height = 1; + size.screen_width = (uint32_t)p->cols * cw; + size.screen_height = (uint32_t)p->rows_n * ch; + size.cell_width = cw; + size.cell_height = ch; ghostty_mouse_encoder_setopt(p->menc, GHOSTTY_MOUSE_ENCODER_OPT_SIZE, &size); - GhosttyMousePosition pos = {.x = (float)ev->mx, .y = (float)ev->my}; + GhosttyMousePosition pos = {.x = (float)ev->mx * cw + cw / 2.0f, + .y = (float)ev->my * ch + ch / 2.0f}; ghostty_mouse_event_set_position(p->mev, pos); /* "no button held" is its own thing, not button zero: bare motion (hover) * must be encoded as none, or a pane in any-event tracking sees nothing. */ |
