foot/render.c

4605 lines
152 KiB
C
Raw Normal View History

#include "render.h"
#include <string.h>
#include <wctype.h>
#include <unistd.h>
#include <signal.h>
2019-08-01 20:09:16 +02:00
#include <sys/ioctl.h>
#include <sys/time.h>
2019-07-21 20:11:20 +02:00
#include <sys/timerfd.h>
#include <sys/epoll.h>
#include <pthread.h>
#include "macros.h"
#if HAS_INCLUDE(<pthread_np.h>)
#include <pthread_np.h>
#define pthread_setname_np(thread, name) (pthread_set_name_np(thread, name), 0)
#elif defined(__NetBSD__)
#define pthread_setname_np(thread, name) pthread_setname_np(thread, "%s", (void *)name)
#endif
#include <wayland-cursor.h>
#include <xdg-shell.h>
#include <presentation-time.h>
#include <fcft/fcft.h>
#define LOG_MODULE "render"
#define LOG_ENABLE_DBG 0
#include "log.h"
#include "box-drawing.h"
#include "char32.h"
#include "config.h"
#include "cursor-shape.h"
#include "grid.h"
#include "hsl.h"
#include "ime.h"
#include "quirks.h"
#include "search.h"
#include "selection.h"
#include "shm.h"
#include "sixel.h"
#include "url-mode.h"
#include "util.h"
#include "xmalloc.h"
#define TIME_SCROLL_DAMAGE 0
struct renderer {
struct fdm *fdm;
struct wayland *wayl;
};
static struct {
size_t total;
size_t zero; /* commits presented in less than one frame interval */
size_t one; /* commits presented in one frame interval */
size_t two; /* commits presented in two or more frame intervals */
2020-08-23 07:42:20 +02:00
} presentation_statistics = {0};
static void fdm_hook_refresh_pending_terminals(struct fdm *fdm, void *data);
struct renderer *
render_init(struct fdm *fdm, struct wayland *wayl)
{
struct renderer *renderer = malloc(sizeof(*renderer));
if (unlikely(renderer == NULL)) {
LOG_ERRNO("malloc() failed");
return NULL;
}
*renderer = (struct renderer) {
.fdm = fdm,
.wayl = wayl,
};
if (!fdm_hook_add(fdm, &fdm_hook_refresh_pending_terminals, renderer,
FDM_HOOK_PRIORITY_NORMAL))
{
LOG_ERR("failed to register FDM hook");
free(renderer);
return NULL;
}
return renderer;
}
void
render_destroy(struct renderer *renderer)
{
if (renderer == NULL)
return;
fdm_hook_del(renderer->fdm, &fdm_hook_refresh_pending_terminals,
FDM_HOOK_PRIORITY_NORMAL);
free(renderer);
}
static void DESTRUCTOR
log_presentation_statistics(void)
{
if (presentation_statistics.total == 0)
return;
const size_t total = presentation_statistics.total;
LOG_INFO("presentation statistics: zero=%f%%, one=%f%%, two=%f%%",
100. * presentation_statistics.zero / total,
100. * presentation_statistics.one / total,
100. * presentation_statistics.two / total);
}
static void
sync_output(void *data,
struct wp_presentation_feedback *wp_presentation_feedback,
struct wl_output *output)
{
}
struct presentation_context {
struct terminal *term;
struct timeval input;
struct timeval commit;
};
static void
presented(void *data,
struct wp_presentation_feedback *wp_presentation_feedback,
uint32_t tv_sec_hi, uint32_t tv_sec_lo, uint32_t tv_nsec,
uint32_t refresh, uint32_t seq_hi, uint32_t seq_lo, uint32_t flags)
{
struct presentation_context *ctx = data;
struct terminal *term = ctx->term;
const struct timeval *input = &ctx->input;
const struct timeval *commit = &ctx->commit;
2020-01-01 11:37:47 +01:00
const struct timeval presented = {
.tv_sec = (uint64_t)tv_sec_hi << 32 | tv_sec_lo,
.tv_usec = tv_nsec / 1000,
};
bool use_input = (input->tv_sec > 0 || input->tv_usec > 0) &&
timercmp(&presented, input, >);
char msg[1024];
int chars = 0;
if (use_input && timercmp(&presented, input, <))
2020-01-01 11:37:47 +01:00
return;
else if (timercmp(&presented, commit, <))
2020-01-01 11:37:47 +01:00
return;
LOG_DBG("commit: %lu s %lu µs, presented: %lu s %lu µs",
commit->tv_sec, commit->tv_usec, presented.tv_sec, presented.tv_usec);
2020-01-01 11:37:47 +01:00
if (use_input) {
struct timeval diff;
timersub(commit, input, &diff);
chars += snprintf(
&msg[chars], sizeof(msg) - chars,
"input - %llu µs -> ", (unsigned long long)diff.tv_usec);
}
struct timeval diff;
timersub(&presented, commit, &diff);
chars += snprintf(
&msg[chars], sizeof(msg) - chars,
"commit - %llu µs -> ", (unsigned long long)diff.tv_usec);
2020-01-01 11:37:47 +01:00
if (use_input) {
xassert(timercmp(&presented, input, >));
timersub(&presented, input, &diff);
2020-01-01 11:37:47 +01:00
} else {
xassert(timercmp(&presented, commit, >));
timersub(&presented, commit, &diff);
2020-01-01 11:37:47 +01:00
}
chars += snprintf(
&msg[chars], sizeof(msg) - chars,
"presented (total: %llu µs)", (unsigned long long)diff.tv_usec);
unsigned frame_count = 0;
if (tll_length(term->window->on_outputs) > 0) {
const struct monitor *mon = tll_front(term->window->on_outputs);
frame_count = (diff.tv_sec * 1000000. + diff.tv_usec) / (1000000. / mon->refresh);
}
presentation_statistics.total++;
if (frame_count >= 2)
presentation_statistics.two++;
else if (frame_count >= 1)
presentation_statistics.one++;
else
presentation_statistics.zero++;
#define _log_fmt "%s (more than %u frames)"
if (frame_count >= 2)
LOG_ERR(_log_fmt, msg, frame_count);
else if (frame_count >= 1)
LOG_WARN(_log_fmt, msg, frame_count);
else
LOG_INFO(_log_fmt, msg, frame_count);
#undef _log_fmt
wp_presentation_feedback_destroy(wp_presentation_feedback);
free(ctx);
}
static void
discarded(void *data, struct wp_presentation_feedback *wp_presentation_feedback)
{
struct presentation_context *ctx = data;
wp_presentation_feedback_destroy(wp_presentation_feedback);
free(ctx);
}
static const struct wp_presentation_feedback_listener presentation_feedback_listener = {
.sync_output = &sync_output,
.presented = &presented,
.discarded = &discarded,
};
static struct fcft_font *
2019-12-19 07:28:33 +01:00
attrs_to_font(const struct terminal *term, const struct attributes *attrs)
{
int idx = attrs->italic << 1 | attrs->bold;
return term->fonts[idx];
}
static inline pixman_color_t
color_hex_to_pixman_with_alpha(uint32_t color, uint16_t alpha)
{
return (pixman_color_t){
.red = ((color >> 16 & 0xff) | (color >> 8 & 0xff00)) * alpha / 0xffff,
.green = ((color >> 8 & 0xff) | (color >> 0 & 0xff00)) * alpha / 0xffff,
.blue = ((color >> 0 & 0xff) | (color << 8 & 0xff00)) * alpha / 0xffff,
.alpha = alpha,
};
}
static inline pixman_color_t
color_hex_to_pixman(uint32_t color)
{
/* Count on the compiler optimizing this */
return color_hex_to_pixman_with_alpha(color, 0xffff);
}
static inline uint32_t
color_decrease_luminance(uint32_t color)
{
uint32_t alpha = color & 0xff000000;
int hue, sat, lum;
rgb_to_hsl(color, &hue, &sat, &lum);
return alpha | hsl_to_rgb(hue, sat, lum / 1.5);
}
static inline uint32_t
color_dim(const struct terminal *term, uint32_t color)
{
const struct config *conf = term->conf;
const uint8_t custom_dim = conf->colors.use_custom.dim;
if (likely(custom_dim == 0))
return color_decrease_luminance(color);
for (size_t i = 0; i < 8; i++) {
if (((custom_dim >> i) & 1) == 0)
continue;
if (term->colors.table[0 + i] == color) {
/* “Regular” color, return the corresponding “dim” */
return conf->colors.dim[i];
}
else if (term->colors.table[8 + i] == color) {
/* “Bright” color, return the corresponding “regular” */
return term->colors.table[i];
}
}
return color_decrease_luminance(color);
}
static inline uint32_t
color_brighten(const struct terminal *term, uint32_t color)
{
/*
* First try to match the color against the base 8 colors. If we
* find a match, return the corresponding bright color.
*/
if (term->conf->bold_in_bright.palette_based) {
for (size_t i = 0; i < 8; i++) {
if (term->colors.table[i] == color)
return term->colors.table[i + 8];
}
return color;
}
int hue, sat, lum;
rgb_to_hsl(color, &hue, &sat, &lum);
lum = (int)roundf(lum * term->conf->bold_in_bright.amount);
return hsl_to_rgb(hue, sat, min(lum, 100));
}
static void
draw_unfocused_block(const struct terminal *term, pixman_image_t *pix,
const pixman_color_t *color, int x, int y, int cell_cols)
{
const int scale = (int)roundf(term->scale);
const int width = min(min(scale, term->cell_width), term->cell_height);
pixman_image_fill_rectangles(
PIXMAN_OP_SRC, pix, color, 4,
(pixman_rectangle16_t []){
{x, y, cell_cols * term->cell_width, width}, /* top */
{x, y, width, term->cell_height}, /* left */
{x + cell_cols * term->cell_width - width, y, width, term->cell_height}, /* right */
{x, y + term->cell_height - width, cell_cols * term->cell_width, width}, /* bottom */
});
}
static void
draw_beam_cursor(const struct terminal *term, pixman_image_t *pix,
const struct fcft_font *font,
const pixman_color_t *color, int x, int y)
{
int baseline = y + term->font_baseline - term->fonts[0]->ascent;
pixman_image_fill_rectangles(
PIXMAN_OP_SRC, pix, color,
1, &(pixman_rectangle16_t){
x, baseline,
term_pt_or_px_as_pixels(term, &term->conf->cursor.beam_thickness),
term->fonts[0]->ascent + term->fonts[0]->descent});
}
static int
underline_offset(const struct terminal *term, const struct fcft_font *font)
{
return term->font_baseline -
(term->conf->use_custom_underline_offset
? -term_pt_or_px_as_pixels(term, &term->conf->underline_offset)
: font->underline.position);
}
static void
draw_underline_cursor(const struct terminal *term, pixman_image_t *pix,
const struct fcft_font *font,
const pixman_color_t *color, int x, int y, int cols)
{
int thickness = term->conf->cursor.underline_thickness.px >= 0
? term_pt_or_px_as_pixels(
term, &term->conf->cursor.underline_thickness)
: font->underline.thickness;
/* Make sure the line isn't positioned below the cell */
const int y_ofs = min(underline_offset(term, font) + thickness,
term->cell_height - thickness);
pixman_image_fill_rectangles(
PIXMAN_OP_SRC, pix, color,
1, &(pixman_rectangle16_t){
x, y + y_ofs, cols * term->cell_width, thickness});
}
static void
draw_underline(const struct terminal *term, pixman_image_t *pix,
const struct fcft_font *font,
const pixman_color_t *color, int x, int y, int cols)
{
const int thickness = term->conf->underline_thickness.px >= 0
? term_pt_or_px_as_pixels(
term, &term->conf->underline_thickness)
: font->underline.thickness;
/* Make sure the line isn't positioned below the cell */
const int y_ofs = min(underline_offset(term, font),
term->cell_height - thickness);
pixman_image_fill_rectangles(
PIXMAN_OP_SRC, pix, color,
1, &(pixman_rectangle16_t){
x, y + y_ofs, cols * term->cell_width, thickness});
}
static void
draw_strikeout(const struct terminal *term, pixman_image_t *pix,
const struct fcft_font *font,
const pixman_color_t *color, int x, int y, int cols)
{
pixman_image_fill_rectangles(
PIXMAN_OP_SRC, pix, color,
1, &(pixman_rectangle16_t){
x, y + term->font_baseline - font->strikeout.position,
cols * term->cell_width, font->strikeout.thickness});
}
static void
cursor_colors_for_cell(const struct terminal *term, const struct cell *cell,
const pixman_color_t *fg, const pixman_color_t *bg,
pixman_color_t *cursor_color, pixman_color_t *text_color)
{
if (term->cursor_color.cursor >> 31) {
xassert(term->cursor_color.text >> 31);
*cursor_color = color_hex_to_pixman(term->cursor_color.cursor);
*text_color = color_hex_to_pixman(term->cursor_color.text);
} else {
*cursor_color = *fg;
*text_color = *bg;
if (unlikely(text_color->alpha != 0xffff)) {
/* The *only* color that can have transparency is the
* default background color */
*text_color = color_hex_to_pixman(term->colors.bg);
}
}
}
static void
draw_cursor(const struct terminal *term, const struct cell *cell,
const struct fcft_font *font, pixman_image_t *pix, pixman_color_t *fg,
const pixman_color_t *bg, int x, int y, int cols)
{
pixman_color_t cursor_color;
pixman_color_t text_color;
cursor_colors_for_cell(term, cell, fg, bg, &cursor_color, &text_color);
switch (term->cursor_style) {
case CURSOR_BLOCK:
if (unlikely(!term->kbd_focus))
draw_unfocused_block(term, pix, &cursor_color, x, y, cols);
else if (likely(term->cursor_blink.state == CURSOR_BLINK_ON)) {
*fg = text_color;
pixman_image_fill_rectangles(
PIXMAN_OP_SRC, pix, &cursor_color, 1,
&(pixman_rectangle16_t){x, y, cols * term->cell_width, term->cell_height});
}
break;
case CURSOR_BEAM:
if (likely(term->cursor_blink.state == CURSOR_BLINK_ON ||
!term->kbd_focus))
{
draw_beam_cursor(term, pix, font, &cursor_color, x, y);
}
break;
case CURSOR_UNDERLINE:
if (likely(term->cursor_blink.state == CURSOR_BLINK_ON ||
!term->kbd_focus))
{
draw_underline_cursor(term, pix, font, &cursor_color, x, y, cols);
}
break;
}
}
static int
render_cell(struct terminal *term, pixman_image_t *pix, pixman_region32_t *damage,
struct row *row, int row_no, int col, bool has_cursor)
{
struct cell *cell = &row->cells[col];
if (cell->attrs.clean)
return 0;
cell->attrs.clean = 1;
cell->attrs.confined = true;
int width = term->cell_width;
int height = term->cell_height;
const int x = term->margins.left + col * width;
const int y = term->margins.top + row_no * height;
bool is_selected = cell->attrs.selected;
uint32_t _fg = 0;
uint32_t _bg = 0;
uint16_t alpha = 0xffff;
if (is_selected && term->colors.use_custom_selection) {
_fg = term->colors.selection_fg;
_bg = term->colors.selection_bg;
} else {
/* Use cell specific color, if set, otherwise the default colors (possible reversed) */
switch (cell->attrs.fg_src) {
case COLOR_RGB:
_fg = cell->attrs.fg;
break;
case COLOR_BASE16:
case COLOR_BASE256:
xassert(cell->attrs.fg < ALEN(term->colors.table));
_fg = term->colors.table[cell->attrs.fg];
break;
case COLOR_DEFAULT:
_fg = term->reverse ? term->colors.bg : term->colors.fg;
break;
}
switch (cell->attrs.bg_src) {
case COLOR_RGB:
_bg = cell->attrs.bg;
break;
case COLOR_BASE16:
case COLOR_BASE256:
xassert(cell->attrs.bg < ALEN(term->colors.table));
_bg = term->colors.table[cell->attrs.bg];
break;
case COLOR_DEFAULT:
_bg = term->reverse ? term->colors.fg : term->colors.bg;
break;
}
if (cell->attrs.reverse ^ is_selected) {
uint32_t swap = _fg;
_fg = _bg;
_bg = swap;
}
else if (cell->attrs.bg_src == COLOR_DEFAULT) {
if (term->window->is_fullscreen) {
/*
* Note: disable transparency when fullscreened.
*
* This is because the wayland protocol mandates no
* screen content is shown behind the fullscreened
* window.
*
* The _intent_ of the specification is that a black
* (or other static color) should be used as
* background.
*
* There's a bit of gray area however, and some
* compositors have chosen to interpret the
* specification in a way that allows wallpapers to be
* seen through a fullscreen window.
*
* Given that a) the intent of the specification, and
* b) we don't know what the compositor will do, we
* simply disable transparency while in fullscreen.
*
* To see why, consider what happens if we keep our
* transparency. For example, if the background color
* is white, and alpha is 0.5, then the window will be
* drawn in a shade of gray while fullscreened.
*
* See
* https://gitlab.freedesktop.org/wayland/wayland-protocols/-/issues/116
* for a discussion on whether transparent, fullscreen
* windows should be allowed in some way or not.
*
* NOTE: if changing this, also update render_margin()
*/
xassert(alpha == 0xffff);
} else {
alpha = term->colors.alpha;
}
}
}
if (unlikely(is_selected && _fg == _bg)) {
/* Invert bg when selected/highlighted text has same fg/bg */
_bg = ~_bg;
alpha = 0xffff;
}
if (cell->attrs.dim)
_fg = color_dim(term, _fg);
if (term->conf->bold_in_bright.enabled && cell->attrs.bold)
_fg = color_brighten(term, _fg);
if (cell->attrs.blink && term->blink.state == BLINK_OFF)
_fg = color_decrease_luminance(_fg);
pixman_color_t fg = color_hex_to_pixman(_fg);
pixman_color_t bg = color_hex_to_pixman_with_alpha(_bg, alpha);
struct fcft_font *font = attrs_to_font(term, &cell->attrs);
const struct composed *composed = NULL;
2020-08-20 19:25:35 +02:00
const struct fcft_grapheme *grapheme = NULL;
const struct fcft_glyph *single = NULL;
const struct fcft_glyph **glyphs = NULL;
unsigned glyph_count = 0;
char32_t base = cell->wc;
int cell_cols = 1;
if (base != 0) {
if (unlikely(
/* Classic box drawings */
(base >= GLYPH_BOX_DRAWING_FIRST &&
base <= GLYPH_BOX_DRAWING_LAST) ||
box-drawing: add braille characters Render braille ourselves, instead of using font glyphs. Decoding a braille character is easy enough; there are 256 codepoints, represented by an 8-bit integer (i.e. subtract the Unicode codepoint offset, 0x2800, and you’re left with an integer in the range 0-255). Each bit corresponds to a dot. The first 6 bits represent the upper 6 dots, while the two last bits represent the fourth (and last) row of dots. The hard part is sizing the dots and the spacing between them. The aim is to have the spacing between the dots be the same size as the dots themselves, and to have the margins on each side be half the size of the dots. In a perfectly sized cell, this means two braille characters next to each other will be evenly spaced. This is however almost never the case. The layout logic currently: * Set dot size to either the width / 4, or height / 8, depending on which one is smallest. * Horizontal spacing is initialized to the width / 4 * Vertical spacing is initialized to the height / 8 * Horizontal margins are initialized to the horizontal spacing / 2 * Vertical margins are initialized to the vertical spacing / 2. Next, we calculate the number of “remaining” pixels. That is, if we add the left margin, two dots and the spacing between, how many pixels are left on the horizontal axis? These pixels are distributed in the following order (we “stop” as soon as we run out of pixels): * If the dot size is 0 (happens for very small font sizes), increase it to 1. * If the margins are 0, increase them to 1. * If we have enough pixels (need at 2 horizontal and 4 vertical), increase the dot size. * Increase spacing. * Increase margins. Closes #702
2021-09-02 14:55:26 +02:00
/* Braille */
(base >= GLYPH_BRAILLE_FIRST &&
base <= GLYPH_BRAILLE_LAST) ||
box-drawing: add braille characters Render braille ourselves, instead of using font glyphs. Decoding a braille character is easy enough; there are 256 codepoints, represented by an 8-bit integer (i.e. subtract the Unicode codepoint offset, 0x2800, and you’re left with an integer in the range 0-255). Each bit corresponds to a dot. The first 6 bits represent the upper 6 dots, while the two last bits represent the fourth (and last) row of dots. The hard part is sizing the dots and the spacing between them. The aim is to have the spacing between the dots be the same size as the dots themselves, and to have the margins on each side be half the size of the dots. In a perfectly sized cell, this means two braille characters next to each other will be evenly spaced. This is however almost never the case. The layout logic currently: * Set dot size to either the width / 4, or height / 8, depending on which one is smallest. * Horizontal spacing is initialized to the width / 4 * Vertical spacing is initialized to the height / 8 * Horizontal margins are initialized to the horizontal spacing / 2 * Vertical margins are initialized to the vertical spacing / 2. Next, we calculate the number of “remaining” pixels. That is, if we add the left margin, two dots and the spacing between, how many pixels are left on the horizontal axis? These pixels are distributed in the following order (we “stop” as soon as we run out of pixels): * If the dot size is 0 (happens for very small font sizes), increase it to 1. * If the margins are 0, increase them to 1. * If we have enough pixels (need at 2 horizontal and 4 vertical), increase the dot size. * Increase spacing. * Increase margins. Closes #702
2021-09-02 14:55:26 +02:00
/*
* Unicode 13 "Symbols for Legacy Computing"
* sub-ranges below.
*
* Note, the full range is U+1FB00 - U+1FBF9
*/
(base >= GLYPH_LEGACY_FIRST &&
base <= GLYPH_LEGACY_LAST)) &&
likely(!term->conf->box_drawings_uses_font_glyphs))
{
struct fcft_glyph ***arr;
size_t count;
size_t idx;
if (base >= GLYPH_LEGACY_FIRST) {
arr = &term->custom_glyphs.legacy;
count = GLYPH_LEGACY_COUNT;
idx = base - GLYPH_LEGACY_FIRST;
} else if (base >= GLYPH_BRAILLE_FIRST) {
arr = &term->custom_glyphs.braille;
count = GLYPH_BRAILLE_COUNT;
idx = base - GLYPH_BRAILLE_FIRST;
} else {
arr = &term->custom_glyphs.box_drawing;
count = GLYPH_BOX_DRAWING_COUNT;
idx = base - GLYPH_BOX_DRAWING_FIRST;
}
if (unlikely(*arr == NULL))
*arr = xcalloc(count, sizeof((*arr)[0]));
if (likely((*arr)[idx] != NULL))
single = (*arr)[idx];
else {
mtx_lock(&term->render.workers.lock);
box-drawing: add braille characters Render braille ourselves, instead of using font glyphs. Decoding a braille character is easy enough; there are 256 codepoints, represented by an 8-bit integer (i.e. subtract the Unicode codepoint offset, 0x2800, and you’re left with an integer in the range 0-255). Each bit corresponds to a dot. The first 6 bits represent the upper 6 dots, while the two last bits represent the fourth (and last) row of dots. The hard part is sizing the dots and the spacing between them. The aim is to have the spacing between the dots be the same size as the dots themselves, and to have the margins on each side be half the size of the dots. In a perfectly sized cell, this means two braille characters next to each other will be evenly spaced. This is however almost never the case. The layout logic currently: * Set dot size to either the width / 4, or height / 8, depending on which one is smallest. * Horizontal spacing is initialized to the width / 4 * Vertical spacing is initialized to the height / 8 * Horizontal margins are initialized to the horizontal spacing / 2 * Vertical margins are initialized to the vertical spacing / 2. Next, we calculate the number of “remaining” pixels. That is, if we add the left margin, two dots and the spacing between, how many pixels are left on the horizontal axis? These pixels are distributed in the following order (we “stop” as soon as we run out of pixels): * If the dot size is 0 (happens for very small font sizes), increase it to 1. * If the margins are 0, increase them to 1. * If we have enough pixels (need at 2 horizontal and 4 vertical), increase the dot size. * Increase spacing. * Increase margins. Closes #702
2021-09-02 14:55:26 +02:00
/* Other thread may have instantiated it while we
2021-09-03 21:38:08 +02:00
* acquired the lock */
single = (*arr)[idx];
if (likely(single == NULL))
single = (*arr)[idx] = box_drawing(term, base);
mtx_unlock(&term->render.workers.lock);
}
2020-08-20 19:25:35 +02:00
if (single != NULL) {
glyph_count = 1;
glyphs = &single;
cell_cols = single->cols;
}
2020-08-20 19:25:35 +02:00
}
composed: store compose chains in a binary search tree The previous implementation stored compose chains in a dynamically allocated array. Adding a chain was easy: resize the array and append the new chain at the end. Looking up a compose chain given a compose chain key/index was also easy: just index into the array. However, searching for a pre-existing chain given a codepoint sequence was very slow. Since the array wasn’t sorted, we typically had to scan through the entire array, just to realize that there is no pre-existing chain, and that we need to add a new one. Since this happens for *each* codepoint in a grapheme cluster, things quickly became really slow. Things were ok:ish as long as the compose chain struct was small, as that made it possible to hold all the chains in the cache. Once the number of chains reached a certain point, or when we were forced to bump maximum number of allowed codepoints in a chain, we started thrashing the cache and things got much much worse. So what can we do? We can’t sort the array, because a) that would invalidate all existing chain keys in the grid (and iterating the entire scrollback and updating compose keys is *not* an option). b) inserting a chain becomes slow as we need to first find _where_ to insert it, and then memmove() the rest of the array. This patch uses a binary search tree to store the chains instead of a simple array. The tree is sorted on a “key”, which is the XOR of all codepoints, truncated to the CELL_COMB_CHARS_HI-CELL_COMB_CHARS_LO range. The grid now stores CELL_COMB_CHARS_LO+key, instead of CELL_COMB_CHARS_LO+index. Since the key is truncated, collisions may occur. This is handled by incrementing the key by 1. Lookup is of course slower than before, O(log n) instead of O(1). Insertion is slightly slower as well: technically it’s O(log n) instead of O(1). However, we also need to take into account the re-allocating the array will occasionally force a full copy of the array when it cannot simply be growed. But finding a pre-existing chain is now *much* faster: O(log n) instead of O(n). In most cases, the first lookup will either succeed (return a true match), or fail (return NULL). However, since key collisions are possible, it may also return false matches. This means we need to verify the contents of the chain before deciding to use it instead of inserting a new chain. But remember that this comparison was being done for each and every chain in the previous implementation. With lookups being much faster, and in particular, no longer requiring us to check the chain contents for every singlec chain, we can now use a dynamically allocated ‘chars’ array in the chain. This was previously a hardcoded array of 10 chars. Using a dynamic allocated array means looking in the array is slower, since we now need two loads: one to load the pointer, and a second to load _from_ the pointer. As a result, the base size of a compose chain (i.e. an “empty” chain) has now been reduced from 48 bytes to 32. A chain with two codepoints is 40 bytes. This means we have up to 4 codepoints while still using less, or the same amount, of memory as before. Furthermore, the Unicode random test (i.e. write random “unicode” chars) is now **faster** than current master (i.e. before text-shaping support was added), **with** test-shaping enabled. With text-shaping disabled, we’re _even_ faster.
2021-06-24 13:17:07 +02:00
else if (base >= CELL_COMB_CHARS_LO && base <= CELL_COMB_CHARS_HI)
2020-08-20 19:25:35 +02:00
{
composed: store compose chains in a binary search tree The previous implementation stored compose chains in a dynamically allocated array. Adding a chain was easy: resize the array and append the new chain at the end. Looking up a compose chain given a compose chain key/index was also easy: just index into the array. However, searching for a pre-existing chain given a codepoint sequence was very slow. Since the array wasn’t sorted, we typically had to scan through the entire array, just to realize that there is no pre-existing chain, and that we need to add a new one. Since this happens for *each* codepoint in a grapheme cluster, things quickly became really slow. Things were ok:ish as long as the compose chain struct was small, as that made it possible to hold all the chains in the cache. Once the number of chains reached a certain point, or when we were forced to bump maximum number of allowed codepoints in a chain, we started thrashing the cache and things got much much worse. So what can we do? We can’t sort the array, because a) that would invalidate all existing chain keys in the grid (and iterating the entire scrollback and updating compose keys is *not* an option). b) inserting a chain becomes slow as we need to first find _where_ to insert it, and then memmove() the rest of the array. This patch uses a binary search tree to store the chains instead of a simple array. The tree is sorted on a “key”, which is the XOR of all codepoints, truncated to the CELL_COMB_CHARS_HI-CELL_COMB_CHARS_LO range. The grid now stores CELL_COMB_CHARS_LO+key, instead of CELL_COMB_CHARS_LO+index. Since the key is truncated, collisions may occur. This is handled by incrementing the key by 1. Lookup is of course slower than before, O(log n) instead of O(1). Insertion is slightly slower as well: technically it’s O(log n) instead of O(1). However, we also need to take into account the re-allocating the array will occasionally force a full copy of the array when it cannot simply be growed. But finding a pre-existing chain is now *much* faster: O(log n) instead of O(n). In most cases, the first lookup will either succeed (return a true match), or fail (return NULL). However, since key collisions are possible, it may also return false matches. This means we need to verify the contents of the chain before deciding to use it instead of inserting a new chain. But remember that this comparison was being done for each and every chain in the previous implementation. With lookups being much faster, and in particular, no longer requiring us to check the chain contents for every singlec chain, we can now use a dynamically allocated ‘chars’ array in the chain. This was previously a hardcoded array of 10 chars. Using a dynamic allocated array means looking in the array is slower, since we now need two loads: one to load the pointer, and a second to load _from_ the pointer. As a result, the base size of a compose chain (i.e. an “empty” chain) has now been reduced from 48 bytes to 32. A chain with two codepoints is 40 bytes. This means we have up to 4 codepoints while still using less, or the same amount, of memory as before. Furthermore, the Unicode random test (i.e. write random “unicode” chars) is now **faster** than current master (i.e. before text-shaping support was added), **with** test-shaping enabled. With text-shaping disabled, we’re _even_ faster.
2021-06-24 13:17:07 +02:00
composed = composed_lookup(term->composed, base - CELL_COMB_CHARS_LO);
2020-08-20 19:25:35 +02:00
base = composed->chars[0];
if (term->conf->can_shape_grapheme && term->conf->tweak.grapheme_shaping) {
grapheme = fcft_rasterize_grapheme_utf32(
font, composed->count, composed->chars, term->font_subpixel);
2020-08-20 19:25:35 +02:00
}
if (grapheme != NULL) {
cell_cols = composed->width;
2020-08-20 19:25:35 +02:00
composed = NULL;
glyphs = grapheme->glyphs;
glyph_count = grapheme->count;
}
2020-08-20 19:25:35 +02:00
}
if (single == NULL && grapheme == NULL) {
if (unlikely(base >= CELL_SPACER)) {
glyph_count = 0;
cell_cols = 1;
} else {
xassert(base != 0);
single = fcft_rasterize_char_utf32(font, base, term->font_subpixel);
if (single == NULL) {
glyph_count = 0;
cell_cols = 1;
} else {
glyph_count = 1;
glyphs = &single;
cell_cols = single->cols;
}
}
2020-08-20 19:25:35 +02:00
}
}
2020-08-20 19:25:35 +02:00
assert(glyph_count == 0 || glyphs != NULL);
const int cols_left = term->cols - col;
cell_cols = max(1, min(cell_cols, cols_left));
/*
* Determine cells that will bleed into their right neighbor and remember
* them for cleanup in the next frame.
*/
int render_width = cell_cols * width;
if (term->conf->tweak.overflowing_glyphs &&
glyph_count > 0 &&
cols_left > cell_cols)
{
int glyph_width = 0, advance = 0;
for (size_t i = 0; i < glyph_count; i++) {
glyph_width = max(glyph_width,
advance + glyphs[i]->x + glyphs[i]->width);
advance += glyphs[i]->advance.x;
}
if (glyph_width > render_width) {
render_width = min(glyph_width, render_width + width);
for (int i = 0; i < cell_cols; i++)
row->cells[col + i].attrs.confined = false;
}
}
pixman_region32_t clip;
pixman_region32_init_rect(
&clip, x, y,
render_width, term->cell_height);
pixman_image_set_clip_region32(pix, &clip);
if (damage != NULL) {
pixman_region32_union_rect(
damage, damage, x, y, render_width, term->cell_height);
}
2021-09-04 20:08:23 +02:00
pixman_region32_fini(&clip);
/* Background */
pixman_image_fill_rectangles(
PIXMAN_OP_SRC, pix, &bg, 1,
&(pixman_rectangle16_t){x, y, cell_cols * width, height});
if (cell->attrs.blink && term->blink.fd < 0) {
/* TODO: use a custom lock for this? */
mtx_lock(&term->render.workers.lock);
term_arm_blink_timer(term);
mtx_unlock(&term->render.workers.lock);
}
2019-07-21 20:11:20 +02:00
if (unlikely(has_cursor && term->cursor_style == CURSOR_BLOCK && term->kbd_focus))
draw_cursor(term, cell, font, pix, &fg, &bg, x, y, cell_cols);
if (cell->wc == 0 || cell->wc >= CELL_SPACER || cell->wc == U'\t' ||
(unlikely(cell->attrs.conceal) && !is_selected))
{
goto draw_cursor;
}
pixman_image_t *clr_pix = pixman_image_create_solid_fill(&fg);
int pen_x = x;
2020-08-20 19:25:35 +02:00
for (unsigned i = 0; i < glyph_count; i++) {
const int letter_x_ofs = i == 0 ? term->font_x_ofs : 0;
const struct fcft_glyph *glyph = glyphs[i];
if (glyph == NULL)
continue;
int g_x = glyph->x;
int g_y = glyph->y;
if (i > 0 && glyph->x >= 0)
g_x -= term->cell_width;
if (unlikely(pixman_image_get_format(glyph->pix) == PIXMAN_a8r8g8b8)) {
/* Glyph surface is a pre-rendered image (typically a color emoji...) */
if (!(cell->attrs.blink && term->blink.state == BLINK_OFF)) {
pixman_image_composite32(
PIXMAN_OP_OVER, glyph->pix, NULL, pix, 0, 0, 0, 0,
pen_x + letter_x_ofs + g_x, y + term->font_baseline - g_y,
glyph->width, glyph->height);
}
} else {
pixman_image_composite32(
PIXMAN_OP_OVER, clr_pix, glyph->pix, pix, 0, 0, 0, 0,
pen_x + letter_x_ofs + g_x, y + term->font_baseline - g_y,
glyph->width, glyph->height);
2020-08-20 19:25:35 +02:00
/* Combining characters */
if (composed != NULL) {
assert(glyph_count == 1);
2020-08-20 19:25:35 +02:00
for (size_t i = 1; i < composed->count; i++) {
const struct fcft_glyph *g = fcft_rasterize_char_utf32(
2020-08-20 19:25:35 +02:00
font, composed->chars[i], term->font_subpixel);
2020-08-20 19:25:35 +02:00
if (g == NULL)
continue;
/*
* Fonts _should_ assume the pen position is now
* *after* the base glyph, and thus use negative
* offsets for combining glyphs.
*
* Not all fonts behave like this however, and we
* try to accommodate both variants.
*
* Since we haven't moved our pen position yet, we
* add a full cell width to the offset (or two, in
* case of double-width characters).
*
* If the font does *not* use negative offsets,
* we'd normally use an offset of 0. However, to
* somewhat deal with double-width glyphs we use
* an offset of *one* cell.
*/
2020-08-20 19:25:35 +02:00
int x_ofs = g->x < 0
? cell_cols * term->cell_width
: (cell_cols - 1) * term->cell_width;
pixman_image_composite32(
PIXMAN_OP_OVER, clr_pix, g->pix, pix, 0, 0, 0, 0,
/* Some fonts use a negative offset, while others use a
* "normal" offset */
pen_x + x_ofs + g->x,
y + term->font_baseline - g->y,
2020-08-20 19:25:35 +02:00
g->width, g->height);
}
}
}
2020-08-20 19:25:35 +02:00
pen_x += glyph->advance.x;
}
pixman_image_unref(clr_pix);
2019-07-16 14:20:39 +02:00
/* Underline */
if (cell->attrs.underline)
draw_underline(term, pix, font, &fg, x, y, cell_cols);
2019-07-16 15:08:02 +02:00
if (cell->attrs.strikethrough)
draw_strikeout(term, pix, font, &fg, x, y, cell_cols);
if (unlikely(cell->attrs.url)) {
pixman_color_t url_color = color_hex_to_pixman(
term->conf->colors.use_custom.url
? term->conf->colors.url
: term->colors.table[3]
);
draw_underline(term, pix, font, &url_color, x, y, cell_cols);
}
draw_cursor:
if (has_cursor && (term->cursor_style != CURSOR_BLOCK || !term->kbd_focus))
draw_cursor(term, cell, font, pix, &fg, &bg, x, y, cell_cols);
pixman_image_set_clip_region32(pix, NULL);
return cell_cols;
}
static void
render_row(struct terminal *term, pixman_image_t *pix, pixman_region32_t *damage,
struct row *row, int row_no, int cursor_col)
{
for (int col = term->cols - 1; col >= 0; col--)
render_cell(term, pix, damage, row, row_no, col, cursor_col == col);
}
static void
render_urgency(struct terminal *term, struct buffer *buf)
{
uint32_t red = term->colors.table[1];
pixman_color_t bg = color_hex_to_pixman(red);
int width = min(min(term->margins.left, term->margins.right),
min(term->margins.top, term->margins.bottom));
pixman_image_fill_rectangles(
PIXMAN_OP_SRC, buf->pix[0], &bg, 4,
(pixman_rectangle16_t[]){
/* Top */
{0, 0, term->width, width},
/* Bottom */
{0, term->height - width, term->width, width},
/* Left */
{0, width, width, term->height - 2 * width},
/* Right */
{term->width - width, width, width, term->height - 2 * width},
});
}
static void
render_margin(struct terminal *term, struct buffer *buf,
int start_line, int end_line, bool apply_damage)
{
/* Fill area outside the cell grid with the default background color */
const int rmargin = term->width - term->margins.right;
const int bmargin = term->height - term->margins.bottom;
const int line_count = end_line - start_line;
const uint32_t _bg = !term->reverse ? term->colors.bg : term->colors.fg;
uint16_t alpha = term->colors.alpha;
if (term->window->is_fullscreen) {
/* Disable alpha in fullscreen - see render_cell() for details */
alpha = 0xffff;
}
pixman_color_t bg = color_hex_to_pixman_with_alpha(_bg, alpha);
pixman_image_fill_rectangles(
PIXMAN_OP_SRC, buf->pix[0], &bg, 4,
(pixman_rectangle16_t[]){
/* Top */
{0, 0, term->width, term->margins.top},
/* Bottom */
{0, bmargin, term->width, term->margins.bottom},
/* Left */
{0, term->margins.top + start_line * term->cell_height,
term->margins.left, line_count * term->cell_height},
/* Right */
{rmargin, term->margins.top + start_line * term->cell_height,
term->margins.right, line_count * term->cell_height},
});
if (term->render.urgency)
render_urgency(term, buf);
/* Ensure the updated regions are copied to the next frame's
* buffer when we're double buffering */
pixman_region32_union_rect(
&buf->dirty[0], &buf->dirty[0], 0, 0, term->width, term->margins.top);
pixman_region32_union_rect(
&buf->dirty[0], &buf->dirty[0], 0, bmargin, term->width, term->margins.bottom);
pixman_region32_union_rect(
&buf->dirty[0], &buf->dirty[0], 0, 0, term->margins.left, term->height);
pixman_region32_union_rect(
&buf->dirty[0], &buf->dirty[0],
rmargin, 0, term->margins.right, term->height);
if (apply_damage) {
/* Top */
wl_surface_damage_buffer(
term->window->surface.surf, 0, 0, term->width, term->margins.top);
/* Bottom */
wl_surface_damage_buffer(
term->window->surface.surf, 0, bmargin, term->width, term->margins.bottom);
/* Left */
wl_surface_damage_buffer(
term->window->surface.surf,
0, term->margins.top + start_line * term->cell_height,
term->margins.left, line_count * term->cell_height);
/* Right */
wl_surface_damage_buffer(
term->window->surface.surf,
rmargin, term->margins.top + start_line * term->cell_height,
term->margins.right, line_count * term->cell_height);
}
}
static void
grid_render_scroll(struct terminal *term, struct buffer *buf,
const struct damage *dmg)
{
LOG_DBG(
"damage: SCROLL: %d-%d by %d lines",
dmg->region.start, dmg->region.end, dmg->lines);
const int region_size = dmg->region.end - dmg->region.start;
if (dmg->lines >= region_size) {
/* The entire scroll region will be scrolled out (i.e. replaced) */
return;
}
const int height = (region_size - dmg->lines) * term->cell_height;
xassert(height > 0);
#if TIME_SCROLL_DAMAGE
struct timespec start_time;
clock_gettime(CLOCK_MONOTONIC, &start_time);
#endif
int dst_y = term->margins.top + (dmg->region.start + 0) * term->cell_height;
int src_y = term->margins.top + (dmg->region.start + dmg->lines) * term->cell_height;
/*
* SHM scrolling can be *much* faster, but it depends on how many
* lines we're scrolling, and how much repairing we need to do.
*
* In short, scrolling a *large* number of rows is faster with a
* memmove, while scrolling a *small* number of lines is faster
* with SHM scrolling.
*
* However, since we need to restore the scrolling regions when
* SHM scrolling, we also need to take this into account.
*
* Finally, we also have to restore the window margins, and this
* is a *huge* performance hit when scrolling a large number of
* lines (in addition to the sloweness of SHM scrolling as
* method).
*
* So, we need to figure out when to SHM scroll, and when to
* memmove.
*
2020-08-15 19:39:00 +01:00
* For now, assume that the both methods perform roughly the same,
* given an equal number of bytes to move/allocate, and use the
* method that results in the least amount of bytes to touch.
*
* Since number of lines directly translates to bytes, we can
* simply count lines.
*
* SHM scrolling needs to first "move" (punch hole + allocate)
* dmg->lines number of lines, and then we need to restore
* the bottom scroll region.
*
* If the total number of lines is less than half the screen - use
* SHM. Otherwise use memmove.
*/
bool try_shm_scroll =
shm_can_scroll(buf) && (
dmg->lines +
dmg->region.start +
(term->rows - dmg->region.end)) < term->rows / 2;
bool did_shm_scroll = false;
//try_shm_scroll = false;
//try_shm_scroll = true;
if (try_shm_scroll) {
did_shm_scroll = shm_scroll(
buf, dmg->lines * term->cell_height,
term->margins.top, dmg->region.start * term->cell_height,
term->margins.bottom, (term->rows - dmg->region.end) * term->cell_height);
}
if (did_shm_scroll) {
/* Restore margins */
render_margin(
term, buf, dmg->region.end - dmg->lines, term->rows, false);
} else {
/* Fallback for when we either cannot do SHM scrolling, or it failed */
uint8_t *raw = buf->data;
memmove(raw + dst_y * buf->stride,
raw + src_y * buf->stride,
height * buf->stride);
}
#if TIME_SCROLL_DAMAGE
struct timespec end_time;
clock_gettime(CLOCK_MONOTONIC, &end_time);
struct timespec memmove_time;
timespec_sub(&end_time, &start_time, &memmove_time);
LOG_INFO("scrolled %dKB (%d lines) using %s in %lds %ldns",
height * buf->stride / 1024, dmg->lines,
did_shm_scroll ? "SHM" : try_shm_scroll ? "memmove (SHM failed)" : "memmove",
(long)memmove_time.tv_sec, memmove_time.tv_nsec);
#endif
wl_surface_damage_buffer(
term->window->surface.surf, term->margins.left, dst_y,
term->width - term->margins.left - term->margins.right, height);
/*
* TODO: remove this if re-enabling scroll damage when re-applying
* last frames damage (see reapply_old_damage()
*/
pixman_region32_union_rect(
&buf->dirty[0], &buf->dirty[0], 0, dst_y, buf->width, height);
}
static void
grid_render_scroll_reverse(struct terminal *term, struct buffer *buf,
const struct damage *dmg)
{
LOG_DBG(
"damage: SCROLL REVERSE: %d-%d by %d lines",
dmg->region.start, dmg->region.end, dmg->lines);
const int region_size = dmg->region.end - dmg->region.start;
if (dmg->lines >= region_size) {
/* The entire scroll region will be scrolled out (i.e. replaced) */
return;
}
const int height = (region_size - dmg->lines) * term->cell_height;
xassert(height > 0);
#if TIME_SCROLL_DAMAGE
struct timespec start_time;
clock_gettime(CLOCK_MONOTONIC, &start_time);
#endif
int src_y = term->margins.top + (dmg->region.start + 0) * term->cell_height;
int dst_y = term->margins.top + (dmg->region.start + dmg->lines) * term->cell_height;
bool try_shm_scroll =
shm_can_scroll(buf) && (
dmg->lines +
dmg->region.start +
(term->rows - dmg->region.end)) < term->rows / 2;
bool did_shm_scroll = false;
if (try_shm_scroll) {
did_shm_scroll = shm_scroll(
buf, -dmg->lines * term->cell_height,
term->margins.top, dmg->region.start * term->cell_height,
term->margins.bottom, (term->rows - dmg->region.end) * term->cell_height);
}
if (did_shm_scroll) {
/* Restore margins */
render_margin(
term, buf, dmg->region.start, dmg->region.start + dmg->lines, false);
} else {
/* Fallback for when we either cannot do SHM scrolling, or it failed */
uint8_t *raw = buf->data;
memmove(raw + dst_y * buf->stride,
raw + src_y * buf->stride,
height * buf->stride);
}
#if TIME_SCROLL_DAMAGE
struct timespec end_time;
clock_gettime(CLOCK_MONOTONIC, &end_time);
struct timespec memmove_time;
timespec_sub(&end_time, &start_time, &memmove_time);
LOG_INFO("scrolled REVERSE %dKB (%d lines) using %s in %lds %ldns",
height * buf->stride / 1024, dmg->lines,
did_shm_scroll ? "SHM" : try_shm_scroll ? "memmove (SHM failed)" : "memmove",
(long)memmove_time.tv_sec, memmove_time.tv_nsec);
#endif
wl_surface_damage_buffer(
term->window->surface.surf, term->margins.left, dst_y,
term->width - term->margins.left - term->margins.right, height);
/*
* TODO: remove this if re-enabling scroll damage when re-applying
* last frames damage (see reapply_old_damage()
*/
pixman_region32_union_rect(
&buf->dirty[0], &buf->dirty[0], 0, dst_y, buf->width, height);
}
2020-02-21 23:48:45 +01:00
static void
render_sixel_chunk(struct terminal *term, pixman_image_t *pix, const struct sixel *sixel,
int term_start_row, int img_start_row, int count)
2020-02-21 23:48:45 +01:00
{
/* Translate row/column to x/y pixel values */
const int x = term->margins.left + sixel->pos.col * term->cell_width;
const int y = term->margins.top + term_start_row * term->cell_height;
2020-02-21 23:48:45 +01:00
/* Width/height, in pixels - and don't touch the window margins */
const int width = max(
0,
min(sixel->width,
term->width - x - term->margins.right));
const int height = max(
0,
min(
min(count * term->cell_height, /* 'count' number of rows */
sixel->height - img_start_row * term->cell_height), /* What remains of the sixel */
term->height - y - term->margins.bottom));
2020-02-21 23:48:45 +01:00
/* Verify we're not stepping outside the grid */
xassert(x >= term->margins.left);
xassert(y >= term->margins.top);
xassert(width == 0 || x + width <= term->width - term->margins.right);
xassert(height == 0 || y + height <= term->height - term->margins.bottom);
2020-02-21 23:48:45 +01:00
//LOG_DBG("sixel chunk: %dx%d %dx%d", x, y, width, height);
pixman_image_composite32(
sixel->opaque ? PIXMAN_OP_SRC : PIXMAN_OP_OVER,
2020-02-21 23:48:45 +01:00
sixel->pix,
NULL,
pix,
0, img_start_row * term->cell_height,
2020-02-21 23:48:45 +01:00
0, 0,
x, y,
width, height);
wl_surface_damage_buffer(term->window->surface.surf, x, y, width, height);
2020-02-21 23:48:45 +01:00
}
static void
render_sixel(struct terminal *term, pixman_image_t *pix,
pixman_region32_t *damage, const struct coord *cursor,
const struct sixel *sixel)
{
xassert(sixel->pix != NULL);
xassert(sixel->width >= 0);
xassert(sixel->height >= 0);
const int view_end = (term->grid->view + term->rows - 1) & (term->grid->num_rows - 1);
const bool last_row_needs_erase = sixel->height % term->cell_height != 0;
const bool last_col_needs_erase = sixel->width % term->cell_width != 0;
int chunk_img_start = -1; /* Image-relative start row of chunk */
int chunk_term_start = -1; /* Viewport relative start row of chunk */
int chunk_row_count = 0; /* Number of rows to emit */
#define maybe_emit_sixel_chunk_then_reset() \
if (chunk_row_count != 0) { \
render_sixel_chunk( \
term, pix, sixel, \
chunk_term_start, chunk_img_start, chunk_row_count); \
chunk_term_start = chunk_img_start = -1; \
chunk_row_count = 0; \
}
/*
* Iterate all sixel rows:
*
* - ignore rows that aren't visible on-screen
* - ignore rows that aren't dirty (they have already been rendered)
* - chunk consecutive dirty rows into a 'chunk'
* - emit (render) chunk as soon as a row isn't visible, or is clean
* - emit final chunk after we've iterated all rows
*
* The purpose of this is to reduce the amount of pixels that
* needs to be composited and marked as damaged for the
* compositor.
*
* Since we do CPU based composition, rendering is a slow and
* heavy task for foot, and thus it is important to not re-render
* things unnecessarily.
*/
for (int _abs_row_no = sixel->pos.row;
_abs_row_no < sixel->pos.row + sixel->rows;
_abs_row_no++)
{
const int abs_row_no = _abs_row_no & (term->grid->num_rows - 1);
const int term_row_no =
(abs_row_no - term->grid->view + term->grid->num_rows) &
(term->grid->num_rows - 1);
/* Check if row is in the visible viewport */
if (view_end >= term->grid->view) {
/* Not wrapped */
if (!(abs_row_no >= term->grid->view && abs_row_no <= view_end)) {
/* Not visible */
maybe_emit_sixel_chunk_then_reset();
continue;
}
} else {
/* Wrapped */
if (!(abs_row_no >= term->grid->view || abs_row_no <= view_end)) {
/* Not visible */
maybe_emit_sixel_chunk_then_reset();
continue;
}
}
/* Is the row dirty? */
struct row *row = term->grid->rows[abs_row_no];
xassert(row != NULL); /* Should be visible */
if (!row->dirty) {
maybe_emit_sixel_chunk_then_reset();
continue;
}
int cursor_col = cursor->row == term_row_no ? cursor->col : -1;
/*
* If image contains transparent parts, render all (dirty)
* cells beneath it.
*
* If image is opaque, loop cells and set their 'clean' bit,
* to prevent the grid rendered from overwriting the sixel
*
* If the last sixel row only partially covers the cell row,
* 'erase' the cell by rendering them.
*
* In all cases, do *not* clear the dirty bit on the row, to
* ensure the regular renderer includes them in the damage
* rect.
*/
if (!sixel->opaque) {
/* TODO: multithreading */
render_row(term, pix, damage, row, term_row_no, cursor_col);
} else {
for (int col = sixel->pos.col;
col < min(sixel->pos.col + sixel->cols, term->cols);
col++)
{
struct cell *cell = &row->cells[col];
if (!cell->attrs.clean) {
bool last_row = abs_row_no == sixel->pos.row + sixel->rows - 1;
bool last_col = col == sixel->pos.col + sixel->cols - 1;
if ((last_row_needs_erase && last_row) ||
(last_col_needs_erase && last_col))
{
render_cell(term, pix, damage, row, term_row_no, col, cursor_col == col);
} else {
cell->attrs.clean = 1;
cell->attrs.confined = 1;
}
}
}
}
if (chunk_term_start == -1) {
xassert(chunk_img_start == -1);
chunk_term_start = term_row_no;
chunk_img_start = _abs_row_no - sixel->pos.row;
chunk_row_count = 1;
} else
chunk_row_count++;
}
maybe_emit_sixel_chunk_then_reset();
#undef maybe_emit_sixel_chunk_then_reset
}
2020-02-21 23:48:45 +01:00
static void
render_sixel_images(struct terminal *term, pixman_image_t *pix,
pixman_region32_t *damage,
const struct coord *cursor)
2020-02-21 23:48:45 +01:00
{
if (likely(tll_length(term->grid->sixel_images)) == 0)
return;
const int scrollback_end
= (term->grid->offset + term->rows) & (term->grid->num_rows - 1);
const int view_start
= (term->grid->view
- scrollback_end
+ term->grid->num_rows) & (term->grid->num_rows - 1);
const int view_end = view_start + term->rows - 1;
//LOG_DBG("SIXELS: %zu images, view=%d-%d",
// tll_length(term->grid->sixel_images), view_start, view_end);
tll_foreach(term->grid->sixel_images, it) {
const struct sixel *six = &it->item;
const int start
= (six->pos.row
- scrollback_end
+ term->grid->num_rows) & (term->grid->num_rows - 1);
const int end = start + six->rows - 1;
//LOG_DBG(" sixel: %d-%d", start, end);
if (start > view_end) {
/* Sixel starts after view ends, no need to try to render it */
continue;
} else if (end < view_start) {
/* Image ends before view starts. Since the image list is
* sorted, we can safely stop here */
break;
}
sixel_sync_cache(term, &it->item);
render_sixel(term, pix, damage, cursor, &it->item);
}
2020-02-21 23:48:45 +01:00
}
#if defined(FOOT_IME_ENABLED) && FOOT_IME_ENABLED
static void
render_ime_preedit_for_seat(struct terminal *term, struct seat *seat,
struct buffer *buf)
{
if (likely(seat->ime.preedit.cells == NULL))
return;
if (unlikely(term->is_searching))
return;
/* Adjust cursor position to viewport */
struct coord cursor;
cursor = term->grid->cursor.point;
cursor.row += term->grid->offset;
cursor.row -= term->grid->view;
cursor.row &= term->grid->num_rows - 1;
if (cursor.row < 0 || cursor.row >= term->rows)
return;
int cells_needed = seat->ime.preedit.count;
if (seat->ime.preedit.cursor.start == cells_needed &&
seat->ime.preedit.cursor.end == cells_needed)
{
/* Cursor will be drawn *after* the pre-edit string, i.e. in
* the cell *after*. This means we need to copy, and dirty,
* one extra cell from the original grid, or well leave
* trailing cursors after us if the user deletes text while
* pre-editing */
cells_needed++;
}
int row_idx = cursor.row;
int col_idx = cursor.col;
2020-12-08 19:17:46 +01:00
int ime_ofs = 0; /* Offset into pre-edit string to start rendering at */
int cells_left = term->cols - cursor.col;
int cells_used = min(cells_needed, term->cols);
/* Adjust start of pre-edit text to the left if string doesn't fit on row */
if (cells_left < cells_used)
col_idx -= cells_used - cells_left;
2020-12-08 19:17:46 +01:00
if (cells_needed > cells_used) {
int start = seat->ime.preedit.cursor.start;
int end = seat->ime.preedit.cursor.end;
2020-12-08 19:17:46 +01:00
if (start == end) {
/* Ensure *end* of pre-edit string is visible */
ime_ofs = cells_needed - cells_used;
} else {
/* Ensure the *beginning* of the cursor-area is visible */
ime_ofs = start;
/* Display as much as possible of the pre-edit string */
if (cells_needed - ime_ofs < cells_used)
ime_ofs = cells_needed - cells_used;
}
/* Make sure we don't start in the middle of a character */
while (ime_ofs < cells_needed &&
seat->ime.preedit.cells[ime_ofs].wc >= CELL_SPACER)
2020-12-08 19:17:46 +01:00
{
ime_ofs++;
}
}
xassert(col_idx >= 0);
xassert(col_idx < term->cols);
struct row *row = grid_row_in_view(term->grid, row_idx);
/* Don't start pre-edit text in the middle of a double-width character */
while (col_idx > 0 && row->cells[col_idx].wc >= CELL_SPACER) {
cells_used++;
col_idx--;
}
/*
* Copy original content (render_cell() reads cell data directly
* from grid), and mark all cells as dirty. This ensures they are
* re-rendered when the pre-edit text is modified or removed.
*/
struct cell *real_cells = xmalloc(cells_used * sizeof(real_cells[0]));
for (int i = 0; i < cells_used; i++) {
xassert(col_idx + i < term->cols);
real_cells[i] = row->cells[col_idx + i];
real_cells[i].attrs.clean = 0;
}
row->dirty = true;
/* Render pre-edit text */
xassert(seat->ime.preedit.cells[ime_ofs].wc < CELL_SPACER);
for (int i = 0, idx = ime_ofs; idx < seat->ime.preedit.count; i++, idx++) {
const struct cell *cell = &seat->ime.preedit.cells[idx];
if (cell->wc >= CELL_SPACER)
continue;
int width = max(1, c32width(cell->wc));
if (col_idx + i + width > term->cols)
break;
2020-12-08 19:17:46 +01:00
row->cells[col_idx + i] = *cell;
render_cell(term, buf->pix[0], NULL, row, row_idx, col_idx + i, false);
}
int start = seat->ime.preedit.cursor.start - ime_ofs;
int end = seat->ime.preedit.cursor.end - ime_ofs;
if (!seat->ime.preedit.cursor.hidden) {
const struct cell *start_cell = &seat->ime.preedit.cells[0];
pixman_color_t fg = color_hex_to_pixman(term->colors.fg);
pixman_color_t bg = color_hex_to_pixman(term->colors.bg);
pixman_color_t cursor_color, text_color;
cursor_colors_for_cell(
term, start_cell, &fg, &bg, &cursor_color, &text_color);
int x = term->margins.left + (col_idx + start) * term->cell_width;
int y = term->margins.top + row_idx * term->cell_height;
if (end == start) {
/* Bar */
2020-12-08 19:17:46 +01:00
if (start >= 0) {
struct fcft_font *font = attrs_to_font(term, &start_cell->attrs);
draw_beam_cursor(term, buf->pix[0], font, &cursor_color, x, y);
2020-12-08 19:17:46 +01:00
}
term_ime_set_cursor_rect(term, x, y, 1, term->cell_height);
}
else if (end > start) {
/* Hollow cursor */
2020-12-08 19:17:46 +01:00
if (start >= 0 && end <= term->cols) {
int cols = end - start;
draw_unfocused_block(term, buf->pix[0], &cursor_color, x, y, cols);
}
term_ime_set_cursor_rect(
term, x, y, (end - start) * term->cell_width, term->cell_height);
}
}
/* Restore original content (but do not render) */
for (int i = 0; i < cells_used; i++)
row->cells[col_idx + i] = real_cells[i];
free(real_cells);
wl_surface_damage_buffer(
term->window->surface.surf,
term->margins.left,
term->margins.top + row_idx * term->cell_height,
term->width - term->margins.left - term->margins.right,
1 * term->cell_height);
}
#endif
static void
render_ime_preedit(struct terminal *term, struct buffer *buf)
{
#if defined(FOOT_IME_ENABLED) && FOOT_IME_ENABLED
tll_foreach(term->wl->seats, it) {
if (it->item.kbd_focus == term)
render_ime_preedit_for_seat(term, &it->item, buf);
}
2020-12-03 18:36:56 +01:00
#endif
}
render: implement ‘flash’ and search mode’s ‘dimming’ with a sub-surface Search mode and ‘flash’ (OSC-555) both achieves similar visual effects: flash tints the entire window yellow, and search mode dims it (except the search match). But, they do so in completely different ways. Search mode is detected in render_cell(), and the colors are then dimmed there. Flash is implemented by blending a yellow, semi-transparent color on top of the rendered grid. This patch replaces those two implementations with a single one. We add a new sub-surface, called the ‘overlay’. In normal mode, it’s unmapped. When either search mode, or flash, is enabled, we enable it, and fill it with a semi-transparent color. Yellow for ‘flash’, and “black” (i.e. no color) for search mode. The compositor then blends it with the grid. Hopefully on the GPU, meaning it’ll be faster than if we blend in software. There are more performance benefits however. By using a separate surface, we can do much better damage tracking. The normal grid rendering code no longer have to care about neither search mode, nor flash. Thus, we get rid of a couple of ‘if’ statements in render_cell(), which is nice. But more importantly, we can drop full grid repaints in a couple of circumstances: * Entering/exiting search mode * Every frame while flash is active Now, when rendering the search mode overlay, we do want to do some damage tracking, also of the overlay. This, since search mode doesn’t dim the *entire* window. The search match is *not* dimmed. This is implemented by punching a hole in the overlay sub-surface. That is, we make part of it *fully* transparent. The basic idea is to set a clip region that excludes the search match, and then dim the rest of the overlay. It’s slightly more complicated than that however, if we want to reuse the last frame’s overlay buffer (i.e we don’t want to re-render the *entire* overlay every frame). In short, we need to: * Clear (punch hole) in areas that are part of this frame’s search match, but not the last frame’s (since those parts are _already_ cleared). * Dim the areas that were part of the last frame’s search match, but aren’t anymore (the rest of the overlay should already be dimmed). To do this, we save the last frame’s “holes” (as a pixman region). Then, when rendering the next frame, we first calculate the new frame’s “holes” region. The region to clear is “this frame’s holes minus last frame’s holes” The region to dim is “last frame’s holes minus this frames holes”. Finally, we compute the bounding box of all modified cells by taking the union of the two diff regions mentioned above. This allows us to limit the buffer damage sent to the compositor.
2022-04-16 17:49:46 +02:00
static void
render_overlay(struct terminal *term)
{
struct wayl_sub_surface *overlay = &term->window->overlay;
bool unicode_mode_active = false;
/* Check if unicode mode is active on at least one seat focusing
* this terminal instance */
tll_foreach(term->wl->seats, it) {
if (it->item.unicode_mode.active) {
unicode_mode_active = true;
break;
}
}
render: implement ‘flash’ and search mode’s ‘dimming’ with a sub-surface Search mode and ‘flash’ (OSC-555) both achieves similar visual effects: flash tints the entire window yellow, and search mode dims it (except the search match). But, they do so in completely different ways. Search mode is detected in render_cell(), and the colors are then dimmed there. Flash is implemented by blending a yellow, semi-transparent color on top of the rendered grid. This patch replaces those two implementations with a single one. We add a new sub-surface, called the ‘overlay’. In normal mode, it’s unmapped. When either search mode, or flash, is enabled, we enable it, and fill it with a semi-transparent color. Yellow for ‘flash’, and “black” (i.e. no color) for search mode. The compositor then blends it with the grid. Hopefully on the GPU, meaning it’ll be faster than if we blend in software. There are more performance benefits however. By using a separate surface, we can do much better damage tracking. The normal grid rendering code no longer have to care about neither search mode, nor flash. Thus, we get rid of a couple of ‘if’ statements in render_cell(), which is nice. But more importantly, we can drop full grid repaints in a couple of circumstances: * Entering/exiting search mode * Every frame while flash is active Now, when rendering the search mode overlay, we do want to do some damage tracking, also of the overlay. This, since search mode doesn’t dim the *entire* window. The search match is *not* dimmed. This is implemented by punching a hole in the overlay sub-surface. That is, we make part of it *fully* transparent. The basic idea is to set a clip region that excludes the search match, and then dim the rest of the overlay. It’s slightly more complicated than that however, if we want to reuse the last frame’s overlay buffer (i.e we don’t want to re-render the *entire* overlay every frame). In short, we need to: * Clear (punch hole) in areas that are part of this frame’s search match, but not the last frame’s (since those parts are _already_ cleared). * Dim the areas that were part of the last frame’s search match, but aren’t anymore (the rest of the overlay should already be dimmed). To do this, we save the last frame’s “holes” (as a pixman region). Then, when rendering the next frame, we first calculate the new frame’s “holes” region. The region to clear is “this frame’s holes minus last frame’s holes” The region to dim is “last frame’s holes minus this frames holes”. Finally, we compute the bounding box of all modified cells by taking the union of the two diff regions mentioned above. This allows us to limit the buffer damage sent to the compositor.
2022-04-16 17:49:46 +02:00
const enum overlay_style style =
term->is_searching ? OVERLAY_SEARCH :
term->flash.active ? OVERLAY_FLASH :
unicode_mode_active ? OVERLAY_UNICODE_MODE :
render: implement ‘flash’ and search mode’s ‘dimming’ with a sub-surface Search mode and ‘flash’ (OSC-555) both achieves similar visual effects: flash tints the entire window yellow, and search mode dims it (except the search match). But, they do so in completely different ways. Search mode is detected in render_cell(), and the colors are then dimmed there. Flash is implemented by blending a yellow, semi-transparent color on top of the rendered grid. This patch replaces those two implementations with a single one. We add a new sub-surface, called the ‘overlay’. In normal mode, it’s unmapped. When either search mode, or flash, is enabled, we enable it, and fill it with a semi-transparent color. Yellow for ‘flash’, and “black” (i.e. no color) for search mode. The compositor then blends it with the grid. Hopefully on the GPU, meaning it’ll be faster than if we blend in software. There are more performance benefits however. By using a separate surface, we can do much better damage tracking. The normal grid rendering code no longer have to care about neither search mode, nor flash. Thus, we get rid of a couple of ‘if’ statements in render_cell(), which is nice. But more importantly, we can drop full grid repaints in a couple of circumstances: * Entering/exiting search mode * Every frame while flash is active Now, when rendering the search mode overlay, we do want to do some damage tracking, also of the overlay. This, since search mode doesn’t dim the *entire* window. The search match is *not* dimmed. This is implemented by punching a hole in the overlay sub-surface. That is, we make part of it *fully* transparent. The basic idea is to set a clip region that excludes the search match, and then dim the rest of the overlay. It’s slightly more complicated than that however, if we want to reuse the last frame’s overlay buffer (i.e we don’t want to re-render the *entire* overlay every frame). In short, we need to: * Clear (punch hole) in areas that are part of this frame’s search match, but not the last frame’s (since those parts are _already_ cleared). * Dim the areas that were part of the last frame’s search match, but aren’t anymore (the rest of the overlay should already be dimmed). To do this, we save the last frame’s “holes” (as a pixman region). Then, when rendering the next frame, we first calculate the new frame’s “holes” region. The region to clear is “this frame’s holes minus last frame’s holes” The region to dim is “last frame’s holes minus this frames holes”. Finally, we compute the bounding box of all modified cells by taking the union of the two diff regions mentioned above. This allows us to limit the buffer damage sent to the compositor.
2022-04-16 17:49:46 +02:00
OVERLAY_NONE;
if (likely(style == OVERLAY_NONE)) {
if (term->render.last_overlay_style != OVERLAY_NONE) {
/* Unmap overlay sub-surface */
wl_surface_attach(overlay->surface.surf, NULL, 0, 0);
wl_surface_commit(overlay->surface.surf);
render: implement ‘flash’ and search mode’s ‘dimming’ with a sub-surface Search mode and ‘flash’ (OSC-555) both achieves similar visual effects: flash tints the entire window yellow, and search mode dims it (except the search match). But, they do so in completely different ways. Search mode is detected in render_cell(), and the colors are then dimmed there. Flash is implemented by blending a yellow, semi-transparent color on top of the rendered grid. This patch replaces those two implementations with a single one. We add a new sub-surface, called the ‘overlay’. In normal mode, it’s unmapped. When either search mode, or flash, is enabled, we enable it, and fill it with a semi-transparent color. Yellow for ‘flash’, and “black” (i.e. no color) for search mode. The compositor then blends it with the grid. Hopefully on the GPU, meaning it’ll be faster than if we blend in software. There are more performance benefits however. By using a separate surface, we can do much better damage tracking. The normal grid rendering code no longer have to care about neither search mode, nor flash. Thus, we get rid of a couple of ‘if’ statements in render_cell(), which is nice. But more importantly, we can drop full grid repaints in a couple of circumstances: * Entering/exiting search mode * Every frame while flash is active Now, when rendering the search mode overlay, we do want to do some damage tracking, also of the overlay. This, since search mode doesn’t dim the *entire* window. The search match is *not* dimmed. This is implemented by punching a hole in the overlay sub-surface. That is, we make part of it *fully* transparent. The basic idea is to set a clip region that excludes the search match, and then dim the rest of the overlay. It’s slightly more complicated than that however, if we want to reuse the last frame’s overlay buffer (i.e we don’t want to re-render the *entire* overlay every frame). In short, we need to: * Clear (punch hole) in areas that are part of this frame’s search match, but not the last frame’s (since those parts are _already_ cleared). * Dim the areas that were part of the last frame’s search match, but aren’t anymore (the rest of the overlay should already be dimmed). To do this, we save the last frame’s “holes” (as a pixman region). Then, when rendering the next frame, we first calculate the new frame’s “holes” region. The region to clear is “this frame’s holes minus last frame’s holes” The region to dim is “last frame’s holes minus this frames holes”. Finally, we compute the bounding box of all modified cells by taking the union of the two diff regions mentioned above. This allows us to limit the buffer damage sent to the compositor.
2022-04-16 17:49:46 +02:00
term->render.last_overlay_style = OVERLAY_NONE;
term->render.last_overlay_buf = NULL;
/* Work around Sway bug - unmapping a sub-surface does not
* damage the underlying surface */
quirk_sway_subsurface_unmap(term);
render: implement ‘flash’ and search mode’s ‘dimming’ with a sub-surface Search mode and ‘flash’ (OSC-555) both achieves similar visual effects: flash tints the entire window yellow, and search mode dims it (except the search match). But, they do so in completely different ways. Search mode is detected in render_cell(), and the colors are then dimmed there. Flash is implemented by blending a yellow, semi-transparent color on top of the rendered grid. This patch replaces those two implementations with a single one. We add a new sub-surface, called the ‘overlay’. In normal mode, it’s unmapped. When either search mode, or flash, is enabled, we enable it, and fill it with a semi-transparent color. Yellow for ‘flash’, and “black” (i.e. no color) for search mode. The compositor then blends it with the grid. Hopefully on the GPU, meaning it’ll be faster than if we blend in software. There are more performance benefits however. By using a separate surface, we can do much better damage tracking. The normal grid rendering code no longer have to care about neither search mode, nor flash. Thus, we get rid of a couple of ‘if’ statements in render_cell(), which is nice. But more importantly, we can drop full grid repaints in a couple of circumstances: * Entering/exiting search mode * Every frame while flash is active Now, when rendering the search mode overlay, we do want to do some damage tracking, also of the overlay. This, since search mode doesn’t dim the *entire* window. The search match is *not* dimmed. This is implemented by punching a hole in the overlay sub-surface. That is, we make part of it *fully* transparent. The basic idea is to set a clip region that excludes the search match, and then dim the rest of the overlay. It’s slightly more complicated than that however, if we want to reuse the last frame’s overlay buffer (i.e we don’t want to re-render the *entire* overlay every frame). In short, we need to: * Clear (punch hole) in areas that are part of this frame’s search match, but not the last frame’s (since those parts are _already_ cleared). * Dim the areas that were part of the last frame’s search match, but aren’t anymore (the rest of the overlay should already be dimmed). To do this, we save the last frame’s “holes” (as a pixman region). Then, when rendering the next frame, we first calculate the new frame’s “holes” region. The region to clear is “this frame’s holes minus last frame’s holes” The region to dim is “last frame’s holes minus this frames holes”. Finally, we compute the bounding box of all modified cells by taking the union of the two diff regions mentioned above. This allows us to limit the buffer damage sent to the compositor.
2022-04-16 17:49:46 +02:00
}
return;
}
struct buffer *buf = shm_get_buffer(
term->render.chains.overlay, term->width, term->height);
pixman_image_set_clip_region32(buf->pix[0], NULL);
pixman_color_t color;
switch (style) {
case OVERLAY_NONE:
break;
case OVERLAY_SEARCH:
case OVERLAY_UNICODE_MODE:
color = (pixman_color_t){0, 0, 0, 0x7fff};
break;
case OVERLAY_FLASH:
color = color_hex_to_pixman_with_alpha(
term->conf->colors.flash,
term->conf->colors.flash_alpha);
break;
}
render: implement ‘flash’ and search mode’s ‘dimming’ with a sub-surface Search mode and ‘flash’ (OSC-555) both achieves similar visual effects: flash tints the entire window yellow, and search mode dims it (except the search match). But, they do so in completely different ways. Search mode is detected in render_cell(), and the colors are then dimmed there. Flash is implemented by blending a yellow, semi-transparent color on top of the rendered grid. This patch replaces those two implementations with a single one. We add a new sub-surface, called the ‘overlay’. In normal mode, it’s unmapped. When either search mode, or flash, is enabled, we enable it, and fill it with a semi-transparent color. Yellow for ‘flash’, and “black” (i.e. no color) for search mode. The compositor then blends it with the grid. Hopefully on the GPU, meaning it’ll be faster than if we blend in software. There are more performance benefits however. By using a separate surface, we can do much better damage tracking. The normal grid rendering code no longer have to care about neither search mode, nor flash. Thus, we get rid of a couple of ‘if’ statements in render_cell(), which is nice. But more importantly, we can drop full grid repaints in a couple of circumstances: * Entering/exiting search mode * Every frame while flash is active Now, when rendering the search mode overlay, we do want to do some damage tracking, also of the overlay. This, since search mode doesn’t dim the *entire* window. The search match is *not* dimmed. This is implemented by punching a hole in the overlay sub-surface. That is, we make part of it *fully* transparent. The basic idea is to set a clip region that excludes the search match, and then dim the rest of the overlay. It’s slightly more complicated than that however, if we want to reuse the last frame’s overlay buffer (i.e we don’t want to re-render the *entire* overlay every frame). In short, we need to: * Clear (punch hole) in areas that are part of this frame’s search match, but not the last frame’s (since those parts are _already_ cleared). * Dim the areas that were part of the last frame’s search match, but aren’t anymore (the rest of the overlay should already be dimmed). To do this, we save the last frame’s “holes” (as a pixman region). Then, when rendering the next frame, we first calculate the new frame’s “holes” region. The region to clear is “this frame’s holes minus last frame’s holes” The region to dim is “last frame’s holes minus this frames holes”. Finally, we compute the bounding box of all modified cells by taking the union of the two diff regions mentioned above. This allows us to limit the buffer damage sent to the compositor.
2022-04-16 17:49:46 +02:00
/* Bounding rectangle of damaged areas - for wl_surface_damage_buffer() */
pixman_box32_t damage_bounds;
if (style == OVERLAY_SEARCH) {
/*
* When possible, we only update the areas that have *changed*
* since the last frame. That means:
*
* - clearing/erasing cells that are now selected, but werent
* in the last frame
* - dimming cells that were selected, but arent anymore
*
* To do this, we save the last frames selected cells as a
* pixman region.
*
* Then, we calculate the corresponding region for this
* frames selected cells.
*
* Last frames region minus this frames region gives us the
* region that needs to be *dimmed* in this frame
*
* This frames region minus last frames region gives us the
* region that needs to be *cleared* in this frame.
*
* Finally, the union of the two diff regions above, gives
* us the total region affected by a change, in either way. We
render: implement ‘flash’ and search mode’s ‘dimming’ with a sub-surface Search mode and ‘flash’ (OSC-555) both achieves similar visual effects: flash tints the entire window yellow, and search mode dims it (except the search match). But, they do so in completely different ways. Search mode is detected in render_cell(), and the colors are then dimmed there. Flash is implemented by blending a yellow, semi-transparent color on top of the rendered grid. This patch replaces those two implementations with a single one. We add a new sub-surface, called the ‘overlay’. In normal mode, it’s unmapped. When either search mode, or flash, is enabled, we enable it, and fill it with a semi-transparent color. Yellow for ‘flash’, and “black” (i.e. no color) for search mode. The compositor then blends it with the grid. Hopefully on the GPU, meaning it’ll be faster than if we blend in software. There are more performance benefits however. By using a separate surface, we can do much better damage tracking. The normal grid rendering code no longer have to care about neither search mode, nor flash. Thus, we get rid of a couple of ‘if’ statements in render_cell(), which is nice. But more importantly, we can drop full grid repaints in a couple of circumstances: * Entering/exiting search mode * Every frame while flash is active Now, when rendering the search mode overlay, we do want to do some damage tracking, also of the overlay. This, since search mode doesn’t dim the *entire* window. The search match is *not* dimmed. This is implemented by punching a hole in the overlay sub-surface. That is, we make part of it *fully* transparent. The basic idea is to set a clip region that excludes the search match, and then dim the rest of the overlay. It’s slightly more complicated than that however, if we want to reuse the last frame’s overlay buffer (i.e we don’t want to re-render the *entire* overlay every frame). In short, we need to: * Clear (punch hole) in areas that are part of this frame’s search match, but not the last frame’s (since those parts are _already_ cleared). * Dim the areas that were part of the last frame’s search match, but aren’t anymore (the rest of the overlay should already be dimmed). To do this, we save the last frame’s “holes” (as a pixman region). Then, when rendering the next frame, we first calculate the new frame’s “holes” region. The region to clear is “this frame’s holes minus last frame’s holes” The region to dim is “last frame’s holes minus this frames holes”. Finally, we compute the bounding box of all modified cells by taking the union of the two diff regions mentioned above. This allows us to limit the buffer damage sent to the compositor.
2022-04-16 17:49:46 +02:00
* use this as the bounding box for the
* wl_surface_damage_buffer() call.
*/
pixman_region32_t *see_through = &term->render.last_overlay_clip;
pixman_region32_t old_see_through;
const bool buffer_reuse =
buf == term->render.last_overlay_buf &&
style == term->render.last_overlay_style &&
buf->age == 0;
render: implement ‘flash’ and search mode’s ‘dimming’ with a sub-surface Search mode and ‘flash’ (OSC-555) both achieves similar visual effects: flash tints the entire window yellow, and search mode dims it (except the search match). But, they do so in completely different ways. Search mode is detected in render_cell(), and the colors are then dimmed there. Flash is implemented by blending a yellow, semi-transparent color on top of the rendered grid. This patch replaces those two implementations with a single one. We add a new sub-surface, called the ‘overlay’. In normal mode, it’s unmapped. When either search mode, or flash, is enabled, we enable it, and fill it with a semi-transparent color. Yellow for ‘flash’, and “black” (i.e. no color) for search mode. The compositor then blends it with the grid. Hopefully on the GPU, meaning it’ll be faster than if we blend in software. There are more performance benefits however. By using a separate surface, we can do much better damage tracking. The normal grid rendering code no longer have to care about neither search mode, nor flash. Thus, we get rid of a couple of ‘if’ statements in render_cell(), which is nice. But more importantly, we can drop full grid repaints in a couple of circumstances: * Entering/exiting search mode * Every frame while flash is active Now, when rendering the search mode overlay, we do want to do some damage tracking, also of the overlay. This, since search mode doesn’t dim the *entire* window. The search match is *not* dimmed. This is implemented by punching a hole in the overlay sub-surface. That is, we make part of it *fully* transparent. The basic idea is to set a clip region that excludes the search match, and then dim the rest of the overlay. It’s slightly more complicated than that however, if we want to reuse the last frame’s overlay buffer (i.e we don’t want to re-render the *entire* overlay every frame). In short, we need to: * Clear (punch hole) in areas that are part of this frame’s search match, but not the last frame’s (since those parts are _already_ cleared). * Dim the areas that were part of the last frame’s search match, but aren’t anymore (the rest of the overlay should already be dimmed). To do this, we save the last frame’s “holes” (as a pixman region). Then, when rendering the next frame, we first calculate the new frame’s “holes” region. The region to clear is “this frame’s holes minus last frame’s holes” The region to dim is “last frame’s holes minus this frames holes”. Finally, we compute the bounding box of all modified cells by taking the union of the two diff regions mentioned above. This allows us to limit the buffer damage sent to the compositor.
2022-04-16 17:49:46 +02:00
if (!buffer_reuse) {
2023-10-03 14:11:55 +02:00
/* Cant reuse last frames damage - set to full window,
render: implement ‘flash’ and search mode’s ‘dimming’ with a sub-surface Search mode and ‘flash’ (OSC-555) both achieves similar visual effects: flash tints the entire window yellow, and search mode dims it (except the search match). But, they do so in completely different ways. Search mode is detected in render_cell(), and the colors are then dimmed there. Flash is implemented by blending a yellow, semi-transparent color on top of the rendered grid. This patch replaces those two implementations with a single one. We add a new sub-surface, called the ‘overlay’. In normal mode, it’s unmapped. When either search mode, or flash, is enabled, we enable it, and fill it with a semi-transparent color. Yellow for ‘flash’, and “black” (i.e. no color) for search mode. The compositor then blends it with the grid. Hopefully on the GPU, meaning it’ll be faster than if we blend in software. There are more performance benefits however. By using a separate surface, we can do much better damage tracking. The normal grid rendering code no longer have to care about neither search mode, nor flash. Thus, we get rid of a couple of ‘if’ statements in render_cell(), which is nice. But more importantly, we can drop full grid repaints in a couple of circumstances: * Entering/exiting search mode * Every frame while flash is active Now, when rendering the search mode overlay, we do want to do some damage tracking, also of the overlay. This, since search mode doesn’t dim the *entire* window. The search match is *not* dimmed. This is implemented by punching a hole in the overlay sub-surface. That is, we make part of it *fully* transparent. The basic idea is to set a clip region that excludes the search match, and then dim the rest of the overlay. It’s slightly more complicated than that however, if we want to reuse the last frame’s overlay buffer (i.e we don’t want to re-render the *entire* overlay every frame). In short, we need to: * Clear (punch hole) in areas that are part of this frame’s search match, but not the last frame’s (since those parts are _already_ cleared). * Dim the areas that were part of the last frame’s search match, but aren’t anymore (the rest of the overlay should already be dimmed). To do this, we save the last frame’s “holes” (as a pixman region). Then, when rendering the next frame, we first calculate the new frame’s “holes” region. The region to clear is “this frame’s holes minus last frame’s holes” The region to dim is “last frame’s holes minus this frames holes”. Finally, we compute the bounding box of all modified cells by taking the union of the two diff regions mentioned above. This allows us to limit the buffer damage sent to the compositor.
2022-04-16 17:49:46 +02:00
* to ensure *everything* is updated */
pixman_region32_init_rect(
&old_see_through, 0, 0, buf->width, buf->height);
} else {
/* Use last frames saved region */
pixman_region32_init(&old_see_through);
pixman_region32_copy(&old_see_through, see_through);
}
pixman_region32_clear(see_through);
/* Build region consisting of all current search matches */
struct search_match_iterator iter = search_matches_new_iter(term);
for (struct range match = search_matches_next(&iter);
match.start.row >= 0;
match = search_matches_next(&iter))
{
int r = match.start.row;
int start_col = match.start.col;
const int end_row = match.end.row;
while (true) {
const int end_col =
r == end_row ? match.end.col : term->cols - 1;
int x = term->margins.left + start_col * term->cell_width;
int y = term->margins.top + r * term->cell_height;
int width = (end_col + 1 - start_col) * term->cell_width;
int height = 1 * term->cell_height;
pixman_region32_union_rect(
see_through, see_through, x, y, width, height);
if (++r > end_row)
break;
start_col = 0;
render: implement ‘flash’ and search mode’s ‘dimming’ with a sub-surface Search mode and ‘flash’ (OSC-555) both achieves similar visual effects: flash tints the entire window yellow, and search mode dims it (except the search match). But, they do so in completely different ways. Search mode is detected in render_cell(), and the colors are then dimmed there. Flash is implemented by blending a yellow, semi-transparent color on top of the rendered grid. This patch replaces those two implementations with a single one. We add a new sub-surface, called the ‘overlay’. In normal mode, it’s unmapped. When either search mode, or flash, is enabled, we enable it, and fill it with a semi-transparent color. Yellow for ‘flash’, and “black” (i.e. no color) for search mode. The compositor then blends it with the grid. Hopefully on the GPU, meaning it’ll be faster than if we blend in software. There are more performance benefits however. By using a separate surface, we can do much better damage tracking. The normal grid rendering code no longer have to care about neither search mode, nor flash. Thus, we get rid of a couple of ‘if’ statements in render_cell(), which is nice. But more importantly, we can drop full grid repaints in a couple of circumstances: * Entering/exiting search mode * Every frame while flash is active Now, when rendering the search mode overlay, we do want to do some damage tracking, also of the overlay. This, since search mode doesn’t dim the *entire* window. The search match is *not* dimmed. This is implemented by punching a hole in the overlay sub-surface. That is, we make part of it *fully* transparent. The basic idea is to set a clip region that excludes the search match, and then dim the rest of the overlay. It’s slightly more complicated than that however, if we want to reuse the last frame’s overlay buffer (i.e we don’t want to re-render the *entire* overlay every frame). In short, we need to: * Clear (punch hole) in areas that are part of this frame’s search match, but not the last frame’s (since those parts are _already_ cleared). * Dim the areas that were part of the last frame’s search match, but aren’t anymore (the rest of the overlay should already be dimmed). To do this, we save the last frame’s “holes” (as a pixman region). Then, when rendering the next frame, we first calculate the new frame’s “holes” region. The region to clear is “this frame’s holes minus last frame’s holes” The region to dim is “last frame’s holes minus this frames holes”. Finally, we compute the bounding box of all modified cells by taking the union of the two diff regions mentioned above. This allows us to limit the buffer damage sent to the compositor.
2022-04-16 17:49:46 +02:00
}
}
/* Areas that need to be cleared: cells that were dimmed in
* the last frame but is now see-through */
render: implement ‘flash’ and search mode’s ‘dimming’ with a sub-surface Search mode and ‘flash’ (OSC-555) both achieves similar visual effects: flash tints the entire window yellow, and search mode dims it (except the search match). But, they do so in completely different ways. Search mode is detected in render_cell(), and the colors are then dimmed there. Flash is implemented by blending a yellow, semi-transparent color on top of the rendered grid. This patch replaces those two implementations with a single one. We add a new sub-surface, called the ‘overlay’. In normal mode, it’s unmapped. When either search mode, or flash, is enabled, we enable it, and fill it with a semi-transparent color. Yellow for ‘flash’, and “black” (i.e. no color) for search mode. The compositor then blends it with the grid. Hopefully on the GPU, meaning it’ll be faster than if we blend in software. There are more performance benefits however. By using a separate surface, we can do much better damage tracking. The normal grid rendering code no longer have to care about neither search mode, nor flash. Thus, we get rid of a couple of ‘if’ statements in render_cell(), which is nice. But more importantly, we can drop full grid repaints in a couple of circumstances: * Entering/exiting search mode * Every frame while flash is active Now, when rendering the search mode overlay, we do want to do some damage tracking, also of the overlay. This, since search mode doesn’t dim the *entire* window. The search match is *not* dimmed. This is implemented by punching a hole in the overlay sub-surface. That is, we make part of it *fully* transparent. The basic idea is to set a clip region that excludes the search match, and then dim the rest of the overlay. It’s slightly more complicated than that however, if we want to reuse the last frame’s overlay buffer (i.e we don’t want to re-render the *entire* overlay every frame). In short, we need to: * Clear (punch hole) in areas that are part of this frame’s search match, but not the last frame’s (since those parts are _already_ cleared). * Dim the areas that were part of the last frame’s search match, but aren’t anymore (the rest of the overlay should already be dimmed). To do this, we save the last frame’s “holes” (as a pixman region). Then, when rendering the next frame, we first calculate the new frame’s “holes” region. The region to clear is “this frame’s holes minus last frame’s holes” The region to dim is “last frame’s holes minus this frames holes”. Finally, we compute the bounding box of all modified cells by taking the union of the two diff regions mentioned above. This allows us to limit the buffer damage sent to the compositor.
2022-04-16 17:49:46 +02:00
pixman_region32_t new_see_through;
pixman_region32_init(&new_see_through);
if (buffer_reuse)
pixman_region32_subtract(&new_see_through, see_through, &old_see_through);
else {
/* Buffer content is unknown - explicitly clear *all*
* current see-through areas */
pixman_region32_copy(&new_see_through, see_through);
}
render: implement ‘flash’ and search mode’s ‘dimming’ with a sub-surface Search mode and ‘flash’ (OSC-555) both achieves similar visual effects: flash tints the entire window yellow, and search mode dims it (except the search match). But, they do so in completely different ways. Search mode is detected in render_cell(), and the colors are then dimmed there. Flash is implemented by blending a yellow, semi-transparent color on top of the rendered grid. This patch replaces those two implementations with a single one. We add a new sub-surface, called the ‘overlay’. In normal mode, it’s unmapped. When either search mode, or flash, is enabled, we enable it, and fill it with a semi-transparent color. Yellow for ‘flash’, and “black” (i.e. no color) for search mode. The compositor then blends it with the grid. Hopefully on the GPU, meaning it’ll be faster than if we blend in software. There are more performance benefits however. By using a separate surface, we can do much better damage tracking. The normal grid rendering code no longer have to care about neither search mode, nor flash. Thus, we get rid of a couple of ‘if’ statements in render_cell(), which is nice. But more importantly, we can drop full grid repaints in a couple of circumstances: * Entering/exiting search mode * Every frame while flash is active Now, when rendering the search mode overlay, we do want to do some damage tracking, also of the overlay. This, since search mode doesn’t dim the *entire* window. The search match is *not* dimmed. This is implemented by punching a hole in the overlay sub-surface. That is, we make part of it *fully* transparent. The basic idea is to set a clip region that excludes the search match, and then dim the rest of the overlay. It’s slightly more complicated than that however, if we want to reuse the last frame’s overlay buffer (i.e we don’t want to re-render the *entire* overlay every frame). In short, we need to: * Clear (punch hole) in areas that are part of this frame’s search match, but not the last frame’s (since those parts are _already_ cleared). * Dim the areas that were part of the last frame’s search match, but aren’t anymore (the rest of the overlay should already be dimmed). To do this, we save the last frame’s “holes” (as a pixman region). Then, when rendering the next frame, we first calculate the new frame’s “holes” region. The region to clear is “this frame’s holes minus last frame’s holes” The region to dim is “last frame’s holes minus this frames holes”. Finally, we compute the bounding box of all modified cells by taking the union of the two diff regions mentioned above. This allows us to limit the buffer damage sent to the compositor.
2022-04-16 17:49:46 +02:00
pixman_image_set_clip_region32(buf->pix[0], &new_see_through);
/* Areas that need to be dimmed: cells that were cleared in
* the last frame but is not anymore */
render: implement ‘flash’ and search mode’s ‘dimming’ with a sub-surface Search mode and ‘flash’ (OSC-555) both achieves similar visual effects: flash tints the entire window yellow, and search mode dims it (except the search match). But, they do so in completely different ways. Search mode is detected in render_cell(), and the colors are then dimmed there. Flash is implemented by blending a yellow, semi-transparent color on top of the rendered grid. This patch replaces those two implementations with a single one. We add a new sub-surface, called the ‘overlay’. In normal mode, it’s unmapped. When either search mode, or flash, is enabled, we enable it, and fill it with a semi-transparent color. Yellow for ‘flash’, and “black” (i.e. no color) for search mode. The compositor then blends it with the grid. Hopefully on the GPU, meaning it’ll be faster than if we blend in software. There are more performance benefits however. By using a separate surface, we can do much better damage tracking. The normal grid rendering code no longer have to care about neither search mode, nor flash. Thus, we get rid of a couple of ‘if’ statements in render_cell(), which is nice. But more importantly, we can drop full grid repaints in a couple of circumstances: * Entering/exiting search mode * Every frame while flash is active Now, when rendering the search mode overlay, we do want to do some damage tracking, also of the overlay. This, since search mode doesn’t dim the *entire* window. The search match is *not* dimmed. This is implemented by punching a hole in the overlay sub-surface. That is, we make part of it *fully* transparent. The basic idea is to set a clip region that excludes the search match, and then dim the rest of the overlay. It’s slightly more complicated than that however, if we want to reuse the last frame’s overlay buffer (i.e we don’t want to re-render the *entire* overlay every frame). In short, we need to: * Clear (punch hole) in areas that are part of this frame’s search match, but not the last frame’s (since those parts are _already_ cleared). * Dim the areas that were part of the last frame’s search match, but aren’t anymore (the rest of the overlay should already be dimmed). To do this, we save the last frame’s “holes” (as a pixman region). Then, when rendering the next frame, we first calculate the new frame’s “holes” region. The region to clear is “this frame’s holes minus last frame’s holes” The region to dim is “last frame’s holes minus this frames holes”. Finally, we compute the bounding box of all modified cells by taking the union of the two diff regions mentioned above. This allows us to limit the buffer damage sent to the compositor.
2022-04-16 17:49:46 +02:00
pixman_region32_t new_dimmed;
pixman_region32_init(&new_dimmed);
pixman_region32_subtract(&new_dimmed, &old_see_through, see_through);
pixman_region32_fini(&old_see_through);
/* Total affected area */
render: implement ‘flash’ and search mode’s ‘dimming’ with a sub-surface Search mode and ‘flash’ (OSC-555) both achieves similar visual effects: flash tints the entire window yellow, and search mode dims it (except the search match). But, they do so in completely different ways. Search mode is detected in render_cell(), and the colors are then dimmed there. Flash is implemented by blending a yellow, semi-transparent color on top of the rendered grid. This patch replaces those two implementations with a single one. We add a new sub-surface, called the ‘overlay’. In normal mode, it’s unmapped. When either search mode, or flash, is enabled, we enable it, and fill it with a semi-transparent color. Yellow for ‘flash’, and “black” (i.e. no color) for search mode. The compositor then blends it with the grid. Hopefully on the GPU, meaning it’ll be faster than if we blend in software. There are more performance benefits however. By using a separate surface, we can do much better damage tracking. The normal grid rendering code no longer have to care about neither search mode, nor flash. Thus, we get rid of a couple of ‘if’ statements in render_cell(), which is nice. But more importantly, we can drop full grid repaints in a couple of circumstances: * Entering/exiting search mode * Every frame while flash is active Now, when rendering the search mode overlay, we do want to do some damage tracking, also of the overlay. This, since search mode doesn’t dim the *entire* window. The search match is *not* dimmed. This is implemented by punching a hole in the overlay sub-surface. That is, we make part of it *fully* transparent. The basic idea is to set a clip region that excludes the search match, and then dim the rest of the overlay. It’s slightly more complicated than that however, if we want to reuse the last frame’s overlay buffer (i.e we don’t want to re-render the *entire* overlay every frame). In short, we need to: * Clear (punch hole) in areas that are part of this frame’s search match, but not the last frame’s (since those parts are _already_ cleared). * Dim the areas that were part of the last frame’s search match, but aren’t anymore (the rest of the overlay should already be dimmed). To do this, we save the last frame’s “holes” (as a pixman region). Then, when rendering the next frame, we first calculate the new frame’s “holes” region. The region to clear is “this frame’s holes minus last frame’s holes” The region to dim is “last frame’s holes minus this frames holes”. Finally, we compute the bounding box of all modified cells by taking the union of the two diff regions mentioned above. This allows us to limit the buffer damage sent to the compositor.
2022-04-16 17:49:46 +02:00
pixman_region32_t damage;
pixman_region32_init(&damage);
pixman_region32_union(&damage, &new_see_through, &new_dimmed);
damage_bounds = damage.extents;
/* Clear cells that became selected in this frame. */
pixman_image_fill_rectangles(
PIXMAN_OP_SRC, buf->pix[0], &(pixman_color_t){0}, 1,
&(pixman_rectangle16_t){0, 0, term->width, term->height});
/* Set clip region for the newly dimmed cells. The actual
* paint call is done below */
pixman_image_set_clip_region32(buf->pix[0], &new_dimmed);
pixman_region32_fini(&new_see_through);
pixman_region32_fini(&new_dimmed);
pixman_region32_fini(&damage);
}
else if (buf == term->render.last_overlay_buf &&
style == term->render.last_overlay_style)
{
xassert(style == OVERLAY_FLASH || style == OVERLAY_UNICODE_MODE);
render: implement ‘flash’ and search mode’s ‘dimming’ with a sub-surface Search mode and ‘flash’ (OSC-555) both achieves similar visual effects: flash tints the entire window yellow, and search mode dims it (except the search match). But, they do so in completely different ways. Search mode is detected in render_cell(), and the colors are then dimmed there. Flash is implemented by blending a yellow, semi-transparent color on top of the rendered grid. This patch replaces those two implementations with a single one. We add a new sub-surface, called the ‘overlay’. In normal mode, it’s unmapped. When either search mode, or flash, is enabled, we enable it, and fill it with a semi-transparent color. Yellow for ‘flash’, and “black” (i.e. no color) for search mode. The compositor then blends it with the grid. Hopefully on the GPU, meaning it’ll be faster than if we blend in software. There are more performance benefits however. By using a separate surface, we can do much better damage tracking. The normal grid rendering code no longer have to care about neither search mode, nor flash. Thus, we get rid of a couple of ‘if’ statements in render_cell(), which is nice. But more importantly, we can drop full grid repaints in a couple of circumstances: * Entering/exiting search mode * Every frame while flash is active Now, when rendering the search mode overlay, we do want to do some damage tracking, also of the overlay. This, since search mode doesn’t dim the *entire* window. The search match is *not* dimmed. This is implemented by punching a hole in the overlay sub-surface. That is, we make part of it *fully* transparent. The basic idea is to set a clip region that excludes the search match, and then dim the rest of the overlay. It’s slightly more complicated than that however, if we want to reuse the last frame’s overlay buffer (i.e we don’t want to re-render the *entire* overlay every frame). In short, we need to: * Clear (punch hole) in areas that are part of this frame’s search match, but not the last frame’s (since those parts are _already_ cleared). * Dim the areas that were part of the last frame’s search match, but aren’t anymore (the rest of the overlay should already be dimmed). To do this, we save the last frame’s “holes” (as a pixman region). Then, when rendering the next frame, we first calculate the new frame’s “holes” region. The region to clear is “this frame’s holes minus last frame’s holes” The region to dim is “last frame’s holes minus this frames holes”. Finally, we compute the bounding box of all modified cells by taking the union of the two diff regions mentioned above. This allows us to limit the buffer damage sent to the compositor.
2022-04-16 17:49:46 +02:00
shm_did_not_use_buf(buf);
return;
} else {
pixman_image_set_clip_region32(buf->pix[0], NULL);
damage_bounds = (pixman_box32_t){0, 0, buf->width, buf->height};
}
pixman_image_fill_rectangles(
PIXMAN_OP_SRC, buf->pix[0], &color, 1,
&(pixman_rectangle16_t){0, 0, term->width, term->height});
2022-04-17 11:04:27 +02:00
quirk_weston_subsurface_desync_on(overlay->sub);
wayl_surface_scale(
term->window, &overlay->surface, buf, term->scale);
render: implement ‘flash’ and search mode’s ‘dimming’ with a sub-surface Search mode and ‘flash’ (OSC-555) both achieves similar visual effects: flash tints the entire window yellow, and search mode dims it (except the search match). But, they do so in completely different ways. Search mode is detected in render_cell(), and the colors are then dimmed there. Flash is implemented by blending a yellow, semi-transparent color on top of the rendered grid. This patch replaces those two implementations with a single one. We add a new sub-surface, called the ‘overlay’. In normal mode, it’s unmapped. When either search mode, or flash, is enabled, we enable it, and fill it with a semi-transparent color. Yellow for ‘flash’, and “black” (i.e. no color) for search mode. The compositor then blends it with the grid. Hopefully on the GPU, meaning it’ll be faster than if we blend in software. There are more performance benefits however. By using a separate surface, we can do much better damage tracking. The normal grid rendering code no longer have to care about neither search mode, nor flash. Thus, we get rid of a couple of ‘if’ statements in render_cell(), which is nice. But more importantly, we can drop full grid repaints in a couple of circumstances: * Entering/exiting search mode * Every frame while flash is active Now, when rendering the search mode overlay, we do want to do some damage tracking, also of the overlay. This, since search mode doesn’t dim the *entire* window. The search match is *not* dimmed. This is implemented by punching a hole in the overlay sub-surface. That is, we make part of it *fully* transparent. The basic idea is to set a clip region that excludes the search match, and then dim the rest of the overlay. It’s slightly more complicated than that however, if we want to reuse the last frame’s overlay buffer (i.e we don’t want to re-render the *entire* overlay every frame). In short, we need to: * Clear (punch hole) in areas that are part of this frame’s search match, but not the last frame’s (since those parts are _already_ cleared). * Dim the areas that were part of the last frame’s search match, but aren’t anymore (the rest of the overlay should already be dimmed). To do this, we save the last frame’s “holes” (as a pixman region). Then, when rendering the next frame, we first calculate the new frame’s “holes” region. The region to clear is “this frame’s holes minus last frame’s holes” The region to dim is “last frame’s holes minus this frames holes”. Finally, we compute the bounding box of all modified cells by taking the union of the two diff regions mentioned above. This allows us to limit the buffer damage sent to the compositor.
2022-04-16 17:49:46 +02:00
wl_subsurface_set_position(overlay->sub, 0, 0);
wl_surface_attach(overlay->surface.surf, buf->wl_buf, 0, 0);
render: implement ‘flash’ and search mode’s ‘dimming’ with a sub-surface Search mode and ‘flash’ (OSC-555) both achieves similar visual effects: flash tints the entire window yellow, and search mode dims it (except the search match). But, they do so in completely different ways. Search mode is detected in render_cell(), and the colors are then dimmed there. Flash is implemented by blending a yellow, semi-transparent color on top of the rendered grid. This patch replaces those two implementations with a single one. We add a new sub-surface, called the ‘overlay’. In normal mode, it’s unmapped. When either search mode, or flash, is enabled, we enable it, and fill it with a semi-transparent color. Yellow for ‘flash’, and “black” (i.e. no color) for search mode. The compositor then blends it with the grid. Hopefully on the GPU, meaning it’ll be faster than if we blend in software. There are more performance benefits however. By using a separate surface, we can do much better damage tracking. The normal grid rendering code no longer have to care about neither search mode, nor flash. Thus, we get rid of a couple of ‘if’ statements in render_cell(), which is nice. But more importantly, we can drop full grid repaints in a couple of circumstances: * Entering/exiting search mode * Every frame while flash is active Now, when rendering the search mode overlay, we do want to do some damage tracking, also of the overlay. This, since search mode doesn’t dim the *entire* window. The search match is *not* dimmed. This is implemented by punching a hole in the overlay sub-surface. That is, we make part of it *fully* transparent. The basic idea is to set a clip region that excludes the search match, and then dim the rest of the overlay. It’s slightly more complicated than that however, if we want to reuse the last frame’s overlay buffer (i.e we don’t want to re-render the *entire* overlay every frame). In short, we need to: * Clear (punch hole) in areas that are part of this frame’s search match, but not the last frame’s (since those parts are _already_ cleared). * Dim the areas that were part of the last frame’s search match, but aren’t anymore (the rest of the overlay should already be dimmed). To do this, we save the last frame’s “holes” (as a pixman region). Then, when rendering the next frame, we first calculate the new frame’s “holes” region. The region to clear is “this frame’s holes minus last frame’s holes” The region to dim is “last frame’s holes minus this frames holes”. Finally, we compute the bounding box of all modified cells by taking the union of the two diff regions mentioned above. This allows us to limit the buffer damage sent to the compositor.
2022-04-16 17:49:46 +02:00
wl_surface_damage_buffer(
overlay->surface.surf,
render: implement ‘flash’ and search mode’s ‘dimming’ with a sub-surface Search mode and ‘flash’ (OSC-555) both achieves similar visual effects: flash tints the entire window yellow, and search mode dims it (except the search match). But, they do so in completely different ways. Search mode is detected in render_cell(), and the colors are then dimmed there. Flash is implemented by blending a yellow, semi-transparent color on top of the rendered grid. This patch replaces those two implementations with a single one. We add a new sub-surface, called the ‘overlay’. In normal mode, it’s unmapped. When either search mode, or flash, is enabled, we enable it, and fill it with a semi-transparent color. Yellow for ‘flash’, and “black” (i.e. no color) for search mode. The compositor then blends it with the grid. Hopefully on the GPU, meaning it’ll be faster than if we blend in software. There are more performance benefits however. By using a separate surface, we can do much better damage tracking. The normal grid rendering code no longer have to care about neither search mode, nor flash. Thus, we get rid of a couple of ‘if’ statements in render_cell(), which is nice. But more importantly, we can drop full grid repaints in a couple of circumstances: * Entering/exiting search mode * Every frame while flash is active Now, when rendering the search mode overlay, we do want to do some damage tracking, also of the overlay. This, since search mode doesn’t dim the *entire* window. The search match is *not* dimmed. This is implemented by punching a hole in the overlay sub-surface. That is, we make part of it *fully* transparent. The basic idea is to set a clip region that excludes the search match, and then dim the rest of the overlay. It’s slightly more complicated than that however, if we want to reuse the last frame’s overlay buffer (i.e we don’t want to re-render the *entire* overlay every frame). In short, we need to: * Clear (punch hole) in areas that are part of this frame’s search match, but not the last frame’s (since those parts are _already_ cleared). * Dim the areas that were part of the last frame’s search match, but aren’t anymore (the rest of the overlay should already be dimmed). To do this, we save the last frame’s “holes” (as a pixman region). Then, when rendering the next frame, we first calculate the new frame’s “holes” region. The region to clear is “this frame’s holes minus last frame’s holes” The region to dim is “last frame’s holes minus this frames holes”. Finally, we compute the bounding box of all modified cells by taking the union of the two diff regions mentioned above. This allows us to limit the buffer damage sent to the compositor.
2022-04-16 17:49:46 +02:00
damage_bounds.x1, damage_bounds.y1,
damage_bounds.x2 - damage_bounds.x1,
damage_bounds.y2 - damage_bounds.y1);
wl_surface_commit(overlay->surface.surf);
2022-04-17 11:04:27 +02:00
quirk_weston_subsurface_desync_off(overlay->sub);
render: implement ‘flash’ and search mode’s ‘dimming’ with a sub-surface Search mode and ‘flash’ (OSC-555) both achieves similar visual effects: flash tints the entire window yellow, and search mode dims it (except the search match). But, they do so in completely different ways. Search mode is detected in render_cell(), and the colors are then dimmed there. Flash is implemented by blending a yellow, semi-transparent color on top of the rendered grid. This patch replaces those two implementations with a single one. We add a new sub-surface, called the ‘overlay’. In normal mode, it’s unmapped. When either search mode, or flash, is enabled, we enable it, and fill it with a semi-transparent color. Yellow for ‘flash’, and “black” (i.e. no color) for search mode. The compositor then blends it with the grid. Hopefully on the GPU, meaning it’ll be faster than if we blend in software. There are more performance benefits however. By using a separate surface, we can do much better damage tracking. The normal grid rendering code no longer have to care about neither search mode, nor flash. Thus, we get rid of a couple of ‘if’ statements in render_cell(), which is nice. But more importantly, we can drop full grid repaints in a couple of circumstances: * Entering/exiting search mode * Every frame while flash is active Now, when rendering the search mode overlay, we do want to do some damage tracking, also of the overlay. This, since search mode doesn’t dim the *entire* window. The search match is *not* dimmed. This is implemented by punching a hole in the overlay sub-surface. That is, we make part of it *fully* transparent. The basic idea is to set a clip region that excludes the search match, and then dim the rest of the overlay. It’s slightly more complicated than that however, if we want to reuse the last frame’s overlay buffer (i.e we don’t want to re-render the *entire* overlay every frame). In short, we need to: * Clear (punch hole) in areas that are part of this frame’s search match, but not the last frame’s (since those parts are _already_ cleared). * Dim the areas that were part of the last frame’s search match, but aren’t anymore (the rest of the overlay should already be dimmed). To do this, we save the last frame’s “holes” (as a pixman region). Then, when rendering the next frame, we first calculate the new frame’s “holes” region. The region to clear is “this frame’s holes minus last frame’s holes” The region to dim is “last frame’s holes minus this frames holes”. Finally, we compute the bounding box of all modified cells by taking the union of the two diff regions mentioned above. This allows us to limit the buffer damage sent to the compositor.
2022-04-16 17:49:46 +02:00
buf->age = 0;
term->render.last_overlay_buf = buf;
term->render.last_overlay_style = style;
}
2019-07-29 20:13:26 +02:00
int
render_worker_thread(void *_ctx)
{
struct render_worker_context *ctx = _ctx;
struct terminal *term = ctx->term;
const int my_id = ctx->my_id;
free(ctx);
2019-07-29 20:13:26 +02:00
sigset_t mask;
sigfillset(&mask);
pthread_sigmask(SIG_SETMASK, &mask, NULL);
2019-08-01 20:09:16 +02:00
char proc_title[16];
snprintf(proc_title, sizeof(proc_title), "foot:render:%d", my_id);
if (pthread_setname_np(pthread_self(), proc_title) < 0)
2019-08-01 20:09:16 +02:00
LOG_ERRNO("render worker %d: failed to set process title", my_id);
2019-07-29 20:13:26 +02:00
sem_t *start = &term->render.workers.start;
sem_t *done = &term->render.workers.done;
mtx_t *lock = &term->render.workers.lock;
while (true) {
sem_wait(start);
struct buffer *buf = term->render.workers.buf;
bool frame_done = false;
/* Translate offset-relative cursor row to view-relative */
struct coord cursor = {-1, -1};
if (!term->hide_cursor) {
cursor = term->grid->cursor.point;
cursor.row += term->grid->offset;
cursor.row -= term->grid->view;
cursor.row &= term->grid->num_rows - 1;
}
2019-07-29 20:13:26 +02:00
while (!frame_done) {
mtx_lock(lock);
xassert(tll_length(term->render.workers.queue) > 0);
2019-07-29 20:13:26 +02:00
int row_no = tll_pop_front(term->render.workers.queue);
mtx_unlock(lock);
switch (row_no) {
default: {
xassert(buf != NULL);
struct row *row = grid_row_in_view(term->grid, row_no);
int cursor_col = cursor.row == row_no ? cursor.col : -1;
render_row(term, buf->pix[my_id], &buf->dirty[my_id],
row, row_no, cursor_col);
2019-07-29 20:13:26 +02:00
break;
}
2019-07-29 20:13:26 +02:00
case -1:
frame_done = true;
sem_post(done);
break;
case -2:
return 0;
}
}
};
return -1;
2019-07-24 20:31:21 +02:00
}
struct csd_data
get_csd_data(const struct terminal *term, enum csd_surface surf_idx)
{
xassert(term->window->csd_mode == CSD_YES);
const bool borders_visible = wayl_win_csd_borders_visible(term->window);
const bool title_visible = wayl_win_csd_titlebar_visible(term->window);
const float scale = term->scale;
const int border_width = borders_visible
? roundf(term->conf->csd.border_width * scale) : 0;
const int title_height = title_visible
? roundf(term->conf->csd.title_height * scale) : 0;
const int button_width = title_visible
? roundf(term->conf->csd.button_width * scale) : 0;
const int button_close_width = term->width >= 1 * button_width
? button_width : 0;
const int button_maximize_width =
term->width >= 2 * button_width && term->window->wm_capabilities.maximize
? button_width : 0;
const int button_minimize_width =
term->width >= 3 * button_width && term->window->wm_capabilities.minimize
? button_width : 0;
/*
* With fractional scaling, we must ensure the offset, when
* divided by the scale (in set_position()), and the scaled back
* (by the compositor), matches the actual pixel count made up by
* the titlebar and the border.
*/
const int top_offset = roundf(
scale * (roundf(-title_height / scale) - roundf(border_width / scale)));
const int top_bottom_width = roundf(
scale * (roundf(term->width / scale) + 2 * roundf(border_width / scale)));
const int left_right_height = roundf(
scale * (roundf(title_height / scale) + roundf(term->height / scale)));
switch (surf_idx) {
case CSD_SURF_TITLE: return (struct csd_data){ 0, -title_height, term->width, title_height};
case CSD_SURF_LEFT: return (struct csd_data){-border_width, -title_height, border_width, left_right_height};
case CSD_SURF_RIGHT: return (struct csd_data){ term->width, -title_height, border_width, left_right_height};
case CSD_SURF_TOP: return (struct csd_data){-border_width, top_offset, top_bottom_width, border_width};
case CSD_SURF_BOTTOM: return (struct csd_data){-border_width, term->height, top_bottom_width, border_width};
/* Positioned relative to CSD_SURF_TITLE */
case CSD_SURF_MINIMIZE: return (struct csd_data){term->width - 3 * button_width, 0, button_minimize_width, title_height};
case CSD_SURF_MAXIMIZE: return (struct csd_data){term->width - 2 * button_width, 0, button_maximize_width, title_height};
case CSD_SURF_CLOSE: return (struct csd_data){term->width - 1 * button_width, 0, button_close_width, title_height};
case CSD_SURF_COUNT:
break;
}
BUG("Invalid csd_surface type");
2020-08-23 07:42:20 +02:00
return (struct csd_data){0};
}
static void
csd_commit(struct terminal *term, struct wayl_surface *surf, struct buffer *buf)
{
wayl_surface_scale(term->window, surf, buf, term->scale);
wl_surface_attach(surf->surf, buf->wl_buf, 0, 0);
wl_surface_damage_buffer(surf->surf, 0, 0, buf->width, buf->height);
wl_surface_commit(surf->surf);
}
static void
render_csd_part(struct terminal *term,
struct wl_surface *surf, struct buffer *buf,
int width, int height, pixman_color_t *color)
{
xassert(term->window->csd_mode == CSD_YES);
2020-02-23 14:17:48 +01:00
pixman_image_fill_rectangles(
PIXMAN_OP_SRC, buf->pix[0], color, 1,
&(pixman_rectangle16_t){0, 0, buf->width, buf->height});
}
2020-02-25 19:09:29 +01:00
static void
render_osd(struct terminal *term, const struct wayl_sub_surface *sub_surf,
struct fcft_font *font, struct buffer *buf,
const char32_t *text, uint32_t _fg, uint32_t _bg,
unsigned x)
{
pixman_region32_t clip;
pixman_region32_init_rect(&clip, 0, 0, buf->width, buf->height);
pixman_image_set_clip_region32(buf->pix[0], &clip);
pixman_region32_fini(&clip);
uint16_t alpha = _bg >> 24 | (_bg >> 24 << 8);
pixman_color_t bg = color_hex_to_pixman_with_alpha(_bg, alpha);
pixman_image_fill_rectangles(
PIXMAN_OP_SRC, buf->pix[0], &bg, 1,
&(pixman_rectangle16_t){0, 0, buf->width, buf->height});
pixman_color_t fg = color_hex_to_pixman(_fg);
const int x_ofs = term->font_x_ofs;
const size_t len = c32len(text);
struct fcft_text_run *text_run = NULL;
const struct fcft_glyph **glyphs = NULL;
const struct fcft_glyph *_glyphs[len];
size_t glyph_count = 0;
if (fcft_capabilities() & FCFT_CAPABILITY_TEXT_RUN_SHAPING) {
text_run = fcft_rasterize_text_run_utf32(
font, len, (const char32_t *)text, term->font_subpixel);
if (text_run != NULL) {
glyphs = text_run->glyphs;
glyph_count = text_run->count;
}
}
if (glyphs == NULL) {
for (size_t i = 0; i < len; i++) {
const struct fcft_glyph *glyph = fcft_rasterize_char_utf32(
font, text[i], term->font_subpixel);
if (glyph == NULL)
continue;
_glyphs[glyph_count++] = glyph;
}
glyphs = _glyphs;
}
pixman_image_t *src = pixman_image_create_solid_fill(&fg);
/* Calculate baseline */
unsigned y;
{
const int line_height = buf->height;
const int font_height = max(font->height, font->ascent + font->descent);
const int glyph_top_y = round((line_height - font_height) / 2.);
y = term->font_y_ofs + glyph_top_y + font->ascent;
}
for (size_t i = 0; i < glyph_count; i++) {
const struct fcft_glyph *glyph = glyphs[i];
if (pixman_image_get_format(glyph->pix) == PIXMAN_a8r8g8b8) {
pixman_image_composite32(
PIXMAN_OP_OVER, glyph->pix, NULL, buf->pix[0], 0, 0, 0, 0,
x + x_ofs + glyph->x, y - glyph->y,
glyph->width, glyph->height);
} else {
pixman_image_composite32(
PIXMAN_OP_OVER, src, glyph->pix, buf->pix[0], 0, 0, 0, 0,
x + x_ofs + glyph->x, y - glyph->y,
glyph->width, glyph->height);
}
x += glyph->advance.x;
}
fcft_text_run_destroy(text_run);
pixman_image_unref(src);
pixman_image_set_clip_region32(buf->pix[0], NULL);
quirk_weston_subsurface_desync_on(sub_surf->sub);
wayl_surface_scale(term->window, &sub_surf->surface, buf, term->scale);
wl_surface_attach(sub_surf->surface.surf, buf->wl_buf, 0, 0);
wl_surface_damage_buffer(sub_surf->surface.surf, 0, 0, buf->width, buf->height);
if (alpha == 0xffff) {
struct wl_region *region = wl_compositor_create_region(term->wl->compositor);
if (region != NULL) {
wl_region_add(region, 0, 0, buf->width, buf->height);
wl_surface_set_opaque_region(sub_surf->surface.surf, region);
wl_region_destroy(region);
}
} else
wl_surface_set_opaque_region(sub_surf->surface.surf, NULL);
wl_surface_commit(sub_surf->surface.surf);
quirk_weston_subsurface_desync_off(sub_surf->sub);
}
static void
render_csd_title(struct terminal *term, const struct csd_data *info,
struct buffer *buf)
{
xassert(term->window->csd_mode == CSD_YES);
struct wayl_sub_surface *surf = &term->window->csd.surface[CSD_SURF_TITLE];
if (info->width == 0 || info->height == 0)
return;
2020-02-23 14:17:48 +01:00
uint32_t bg = term->conf->csd.color.title_set
? term->conf->csd.color.title
: 0xffu << 24 | term->conf->colors.fg;
uint32_t fg = term->conf->csd.color.buttons_set
? term->conf->csd.color.buttons
: term->conf->colors.bg;
if (!term->visual_focus) {
bg = color_dim(term, bg);
fg = color_dim(term, fg);
}
2020-02-25 19:09:29 +01:00
char32_t *_title_text = ambstoc32(term->window_title);
const char32_t *title_text = _title_text != NULL ? _title_text : U"";
struct wl_window *win = term->window;
const struct fcft_glyph *M = fcft_rasterize_char_utf32(
win->csd.font, U'M', term->font_subpixel);
const int margin = M != NULL ? M->advance.x : win->csd.font->max_advance.x;
render_osd(term, surf, win->csd.font, buf, title_text, fg, bg, margin);
csd_commit(term, &surf->surface, buf);
free(_title_text);
}
static void
render_csd_border(struct terminal *term, enum csd_surface surf_idx,
const struct csd_data *info, struct buffer *buf)
{
xassert(term->window->csd_mode == CSD_YES);
xassert(surf_idx >= CSD_SURF_LEFT && surf_idx <= CSD_SURF_BOTTOM);
struct wayl_surface *surf = &term->window->csd.surface[surf_idx].surface;
if (info->width == 0 || info->height == 0)
return;
{
pixman_color_t color = color_hex_to_pixman_with_alpha(0, 0);
render_csd_part(term, surf->surf, buf, info->width, info->height, &color);
}
/*
* The visible border.
*/
float scale = term->scale;
int bwidth = (int)roundf(term->conf->csd.border_width * scale);
int vwidth = (int)roundf(term->conf->csd.border_width_visible * scale); /* Visible size */
xassert(bwidth >= vwidth);
if (vwidth > 0) {
const struct config *conf = term->conf;
int x = 0, y = 0, w = 0, h = 0;
switch (surf_idx) {
case CSD_SURF_TOP:
case CSD_SURF_BOTTOM:
x = bwidth - vwidth;
y = surf_idx == CSD_SURF_TOP ? info->height - vwidth : 0;
w = info->width - 2 * x;
h = vwidth;
break;
case CSD_SURF_LEFT:
case CSD_SURF_RIGHT:
x = surf_idx == CSD_SURF_LEFT ? bwidth - vwidth : 0;
y = 0;
w = vwidth;
h = info->height;
break;
case CSD_SURF_TITLE:
case CSD_SURF_MINIMIZE:
case CSD_SURF_MAXIMIZE:
case CSD_SURF_CLOSE:
case CSD_SURF_COUNT:
BUG("unexpected CSD surface type");
}
xassert(x >= 0);
xassert(y >= 0);
xassert(w >= 0);
xassert(h >= 0);
xassert(x + w <= info->width);
xassert(y + h <= info->height);
uint32_t _color =
conf->csd.color.border_set ? conf->csd.color.border :
conf->csd.color.title_set ? conf->csd.color.title :
0xffu << 24 | term->conf->colors.fg;
if (!term->visual_focus)
_color = color_dim(term, _color);
uint16_t alpha = _color >> 24 | (_color >> 24 << 8);
pixman_color_t color = color_hex_to_pixman_with_alpha(_color, alpha);
pixman_image_fill_rectangles(
PIXMAN_OP_SRC, buf->pix[0], &color, 1,
&(pixman_rectangle16_t){x, y, w, h});
}
csd_commit(term, surf, buf);
}
static pixman_color_t
get_csd_button_fg_color(const struct config *conf)
{
uint32_t _color = conf->colors.bg;
uint16_t alpha = 0xffff;
if (conf->csd.color.buttons_set) {
_color = conf->csd.color.buttons;
alpha = _color >> 24 | (_color >> 24 << 8);
}
return color_hex_to_pixman_with_alpha(_color, alpha);
}
static void
render_csd_button_minimize(struct terminal *term, struct buffer *buf)
{
pixman_color_t color = get_csd_button_fg_color(term->conf);
pixman_image_t *src = pixman_image_create_solid_fill(&color);
2023-04-25 03:43:36 +00:00
const int max_height = buf->height / 3;
const int max_width = buf->width / 3;
2023-04-25 03:43:36 +00:00
int width = min(max_height, max_width);
int thick = min(width / 2, 1 * term->scale);
2023-04-25 03:43:36 +00:00
const int x_margin = (buf->width - width) / 2;
const int y_margin = (buf->height - width) / 2;
2023-04-25 03:43:36 +00:00
xassert(x_margin + width - thick >= 0);
xassert(width - 2 * thick >= 0);
xassert(y_margin + width - thick >= 0);
pixman_image_fill_rectangles(
PIXMAN_OP_SRC, buf->pix[0], &color, 1,
(pixman_rectangle16_t[]) {
{x_margin, y_margin + width - thick, width, thick}
});
pixman_image_unref(src);
}
static void
render_csd_button_maximize_maximized(
struct terminal *term, struct buffer *buf)
{
pixman_color_t color = get_csd_button_fg_color(term->conf);
pixman_image_t *src = pixman_image_create_solid_fill(&color);
const int max_height = buf->height / 3;
const int max_width = buf->width / 3;
int width = min(max_height, max_width);
int thick = min(width / 2, 1 * term->scale);
const int x_margin = (buf->width - width) / 2;
const int y_margin = (buf->height - width) / 2;
2023-04-25 03:43:36 +00:00
const int shrink = 1;
xassert(x_margin + width - thick >= 0);
xassert(width - 2 * thick >= 0);
xassert(y_margin + width - thick >= 0);
pixman_image_fill_rectangles(
PIXMAN_OP_SRC, buf->pix[0], &color, 4,
(pixman_rectangle16_t[]){
2023-04-25 03:43:36 +00:00
{x_margin + shrink, y_margin + shrink, width - 2 * shrink, thick},
{ x_margin + shrink, y_margin + thick, thick, width - 2 * thick - shrink },
{ x_margin + width - thick - shrink, y_margin + thick, thick, width - 2 * thick - shrink },
{ x_margin + shrink, y_margin + width - thick - shrink, width - 2 * shrink, thick }});
pixman_image_unref(src);
}
static void
render_csd_button_maximize_window(
struct terminal *term, struct buffer *buf)
{
pixman_color_t color = get_csd_button_fg_color(term->conf);
pixman_image_t *src = pixman_image_create_solid_fill(&color);
2023-04-25 03:43:36 +00:00
const int max_height = buf->height / 3;
const int max_width = buf->width / 3;
2023-04-25 03:43:36 +00:00
int width = min(max_height, max_width);
int thick = min(width / 2, 1 * term->scale);
2023-04-25 03:43:36 +00:00
const int x_margin = (buf->width - width) / 2;
const int y_margin = (buf->height - width) / 2;
2023-04-25 03:43:36 +00:00
xassert(x_margin + width - thick >= 0);
xassert(width - 2 * thick >= 0);
xassert(y_margin + width - thick >= 0);
2023-04-25 03:43:36 +00:00
pixman_image_fill_rectangles(
PIXMAN_OP_SRC, buf->pix[0], &color, 4,
(pixman_rectangle16_t[]) {
{x_margin, y_margin, width, thick},
{ x_margin, y_margin + thick, thick, width - 2 * thick },
{ x_margin + width - thick, y_margin + thick, thick, width - 2 * thick },
{ x_margin, y_margin + width - thick, width, thick }
});
pixman_image_unref(src);
}
static void
render_csd_button_maximize(struct terminal *term, struct buffer *buf)
{
if (term->window->is_maximized)
render_csd_button_maximize_maximized(term, buf);
else
render_csd_button_maximize_window(term, buf);
}
static void
render_csd_button_close(struct terminal *term, struct buffer *buf)
{
pixman_color_t color = get_csd_button_fg_color(term->conf);
pixman_image_t *src = pixman_image_create_solid_fill(&color);
const int max_height = buf->height / 3;
const int max_width = buf->width / 3;
int width = min(max_height, max_width);
2023-04-25 03:43:36 +00:00
int thick = min(width / 2, 1 * term->scale);
const int x_margin = (buf->width - width) / 2;
const int y_margin = (buf->height - width) / 2;
2023-04-25 03:43:36 +00:00
xassert(x_margin + width - thick >= 0);
xassert(width - 2 * thick >= 0);
xassert(y_margin + width - thick >= 0);
pixman_triangle_t tri[4] = {
{
.p1 = {
.x = pixman_int_to_fixed(x_margin),
.y = pixman_int_to_fixed(y_margin + thick),
},
.p2 = {
.x = pixman_int_to_fixed(x_margin + width - thick),
.y = pixman_int_to_fixed(y_margin + width),
},
.p3 = {
.x = pixman_int_to_fixed(x_margin + thick),
.y = pixman_int_to_fixed(y_margin),
},
},
{
.p1 = {
.x = pixman_int_to_fixed(x_margin + width),
.y = pixman_int_to_fixed(y_margin + width - thick),
},
.p2 = {
.x = pixman_int_to_fixed(x_margin + thick),
.y = pixman_int_to_fixed(y_margin),
},
.p3 = {
.x = pixman_int_to_fixed(x_margin + width - thick),
.y = pixman_int_to_fixed(y_margin + width),
},
},
{
.p1 = {
.x = pixman_int_to_fixed(x_margin),
.y = pixman_int_to_fixed(y_margin + width - thick),
},
.p2 = {
.x = pixman_int_to_fixed(x_margin + width),
.y = pixman_int_to_fixed(y_margin + thick),
},
.p3 = {
.x = pixman_int_to_fixed(x_margin + thick),
.y = pixman_int_to_fixed(y_margin + width),
},
},
{
.p1 = {
.x = pixman_int_to_fixed(x_margin + width),
.y = pixman_int_to_fixed(y_margin + thick),
},
.p2 = {
.x = pixman_int_to_fixed(x_margin),
.y = pixman_int_to_fixed(y_margin + width - thick),
},
.p3 = {
.x = pixman_int_to_fixed(x_margin + width - thick),
.y = pixman_int_to_fixed(y_margin),
},
},
};
pixman_composite_triangles(
PIXMAN_OP_OVER, src, buf->pix[0], PIXMAN_a1,
0, 0, 0, 0, 4, tri);
pixman_image_unref(src);
}
static void
render_csd_button(struct terminal *term, enum csd_surface surf_idx,
const struct csd_data *info, struct buffer *buf)
{
xassert(term->window->csd_mode == CSD_YES);
xassert(surf_idx >= CSD_SURF_MINIMIZE && surf_idx <= CSD_SURF_CLOSE);
struct wayl_surface *surf = &term->window->csd.surface[surf_idx].surface;
if (info->width == 0 || info->height == 0)
return;
uint32_t _color;
uint16_t alpha = 0xffff;
bool is_active = false;
bool is_set = false;
const uint32_t *conf_color = NULL;
switch (surf_idx) {
case CSD_SURF_MINIMIZE:
_color = term->conf->colors.table[4]; /* blue */
is_set = term->conf->csd.color.minimize_set;
conf_color = &term->conf->csd.color.minimize;
is_active = term->active_surface == TERM_SURF_BUTTON_MINIMIZE;
break;
case CSD_SURF_MAXIMIZE:
_color = term->conf->colors.table[2]; /* green */
is_set = term->conf->csd.color.maximize_set;
conf_color = &term->conf->csd.color.maximize;
is_active = term->active_surface == TERM_SURF_BUTTON_MAXIMIZE;
break;
case CSD_SURF_CLOSE:
_color = term->conf->colors.table[1]; /* red */
is_set = term->conf->csd.color.close_set;
conf_color = &term->conf->csd.color.quit;
is_active = term->active_surface == TERM_SURF_BUTTON_CLOSE;
break;
default:
BUG("unhandled surface type: %u", (unsigned)surf_idx);
break;
}
if (is_active) {
if (is_set) {
_color = *conf_color;
alpha = _color >> 24 | (_color >> 24 << 8);
}
} else {
_color = 0;
alpha = 0;
}
if (!term->visual_focus)
_color = color_dim(term, _color);
pixman_color_t color = color_hex_to_pixman_with_alpha(_color, alpha);
render_csd_part(term, surf->surf, buf, info->width, info->height, &color);
switch (surf_idx) {
case CSD_SURF_MINIMIZE: render_csd_button_minimize(term, buf); break;
case CSD_SURF_MAXIMIZE: render_csd_button_maximize(term, buf); break;
case CSD_SURF_CLOSE: render_csd_button_close(term, buf); break;
break;
default:
BUG("unhandled surface type: %u", (unsigned)surf_idx);
break;
}
csd_commit(term, surf, buf);
}
static void
render_csd(struct terminal *term)
{
xassert(term->window->csd_mode == CSD_YES);
if (term->window->is_fullscreen)
return;
const float scale = term->scale;
struct csd_data infos[CSD_SURF_COUNT];
int widths[CSD_SURF_COUNT];
int heights[CSD_SURF_COUNT];
for (size_t i = 0; i < CSD_SURF_COUNT; i++) {
infos[i] = get_csd_data(term, i);
const int x = infos[i].x;
const int y = infos[i].y;
const int width = infos[i].width;
const int height = infos[i].height;
struct wl_surface *surf = term->window->csd.surface[i].surface.surf;
struct wl_subsurface *sub = term->window->csd.surface[i].sub;
xassert(surf != NULL);
xassert(sub != NULL);
if (width == 0 || height == 0) {
widths[i] = heights[i] = 0;
wl_subsurface_set_position(sub, 0, 0);
wl_surface_attach(surf, NULL, 0, 0);
wl_surface_commit(surf);
continue;
}
widths[i] = width;
heights[i] = height;
wl_subsurface_set_position(sub, roundf(x / scale), roundf(y / scale));
}
struct buffer *bufs[CSD_SURF_COUNT];
shm_get_many(term->render.chains.csd, CSD_SURF_COUNT, widths, heights, bufs);
for (size_t i = CSD_SURF_LEFT; i <= CSD_SURF_BOTTOM; i++)
render_csd_border(term, i, &infos[i], bufs[i]);
for (size_t i = CSD_SURF_MINIMIZE; i <= CSD_SURF_CLOSE; i++)
render_csd_button(term, i, &infos[i], bufs[i]);
render_csd_title(term, &infos[CSD_SURF_TITLE], bufs[CSD_SURF_TITLE]);
2020-02-23 14:17:48 +01:00
}
static void
render_scrollback_position(struct terminal *term)
{
if (term->conf->scrollback.indicator.position == SCROLLBACK_INDICATOR_POSITION_NONE)
return;
struct wl_window *win = term->window;
if (term->grid->view == term->grid->offset) {
if (win->scrollback_indicator.surface.surf != NULL) {
wayl_win_subsurface_destroy(&win->scrollback_indicator);
/* Work around Sway bug - unmapping a sub-surface does not damage
* the underlying surface */
quirk_sway_subsurface_unmap(term);
}
return;
}
if (win->scrollback_indicator.surface.surf == NULL) {
if (!wayl_win_subsurface_new(
win, &win->scrollback_indicator, false))
{
LOG_ERR("failed to create scrollback indicator surface");
return;
}
}
xassert(win->scrollback_indicator.surface.surf != NULL);
xassert(win->scrollback_indicator.sub != NULL);
/* Find absolute row number of the scrollback start */
int scrollback_start = term->grid->offset + term->rows;
int empty_rows = 0;
while (term->grid->rows[scrollback_start & (term->grid->num_rows - 1)] == NULL) {
scrollback_start++;
empty_rows++;
}
/* Rebase viewport against scrollback start (so that 0 is at
* the beginning of the scrollback) */
int rebased_view = term->grid->view - scrollback_start + term->grid->num_rows;
rebased_view &= term->grid->num_rows - 1;
/* How much of the scrollback is actually used? */
int populated_rows = term->grid->num_rows - empty_rows;
xassert(populated_rows > 0);
xassert(populated_rows <= term->grid->num_rows);
/*
* How far down in the scrollback we are.
*
* 0% -> at the beginning of the scrollback
* 100% -> at the bottom, i.e. where new lines are inserted
*/
double percent =
rebased_view + term->rows == populated_rows
? 1.0
: (double)rebased_view / (populated_rows - term->rows);
char32_t _text[64];
const char32_t *text = _text;
int cell_count = 0;
/* *What* to render */
switch (term->conf->scrollback.indicator.format) {
case SCROLLBACK_INDICATOR_FORMAT_PERCENTAGE: {
char percent_str[8];
snprintf(percent_str, sizeof(percent_str), "%u%%", (int)(100 * percent));
mbstoc32(_text, percent_str, ALEN(_text));
cell_count = 3;
break;
}
case SCROLLBACK_INDICATOR_FORMAT_LINENO: {
char lineno_str[64];
snprintf(lineno_str, sizeof(lineno_str), "%d", rebased_view + 1);
mbstoc32(_text, lineno_str, ALEN(_text));
cell_count = (int)ceilf(log10f(term->grid->num_rows));
break;
}
case SCROLLBACK_INDICATOR_FORMAT_TEXT:
text = term->conf->scrollback.indicator.text;
cell_count = c32len(text);
break;
}
const float scale = term->scale;
const int margin = (int)roundf(3. * scale);
int width = margin + cell_count * term->cell_width + margin;
int height = margin + term->cell_height + margin;
width = roundf(scale * ceilf(width / scale));
height = roundf(scale * ceilf(height / scale));
/* *Where* to render - parent relative coordinates */
int surf_top = 0;
switch (term->conf->scrollback.indicator.position) {
case SCROLLBACK_INDICATOR_POSITION_NONE:
BUG("Invalid scrollback indicator position type");
return;
case SCROLLBACK_INDICATOR_POSITION_FIXED:
surf_top = term->cell_height - margin;
break;
case SCROLLBACK_INDICATOR_POSITION_RELATIVE: {
int lines = term->rows - 2; /* Avoid using first and last rows */
if (term->is_searching) {
/* Make sure we don't collide with the scrollback search box */
lines--;
}
lines = max(lines, 0);
int pixels = max(lines * term->cell_height - height + 2 * margin, 0);
surf_top = term->cell_height - margin + (int)(percent * pixels);
break;
}
}
int x = term->width - margin - width;
int y = term->margins.top + surf_top;
x = roundf(scale * ceilf(x / scale));
y = roundf(scale * ceilf(y / scale));
if (y + height > term->height) {
wl_surface_attach(win->scrollback_indicator.surface.surf, NULL, 0, 0);
wl_surface_commit(win->scrollback_indicator.surface.surf);
return;
}
struct buffer_chain *chain = term->render.chains.scrollback_indicator;
struct buffer *buf = shm_get_buffer(chain, width, height);
wl_subsurface_set_position(
win->scrollback_indicator.sub, roundf(x / scale), roundf(y / scale));
2021-10-20 20:03:15 +02:00
uint32_t fg = term->colors.table[0];
uint32_t bg = term->colors.table[8 + 4];
if (term->conf->colors.use_custom.scrollback_indicator) {
fg = term->conf->colors.scrollback_indicator.fg;
bg = term->conf->colors.scrollback_indicator.bg;
}
render_osd(
term,
&win->scrollback_indicator,
term->fonts[0], buf, text,
fg, 0xffu << 24 | bg,
width - margin - c32len(text) * term->cell_width);
}
static void
render_render_timer(struct terminal *term, struct timespec render_time)
{
struct wl_window *win = term->window;
char usecs_str[256];
double usecs = render_time.tv_sec * 1000000 + render_time.tv_nsec / 1000.0;
snprintf(usecs_str, sizeof(usecs_str), "%.2f µs", usecs);
char32_t text[256];
mbstoc32(text, usecs_str, ALEN(text));
const float scale = term->scale;
const int cell_count = c32len(text);
const int margin = (int)roundf(3. * scale);
int width = margin + cell_count * term->cell_width + margin;
int height = margin + term->cell_height + margin;
width = roundf(scale * ceilf(width / scale));
height = roundf(scale * ceilf(height / scale));
struct buffer_chain *chain = term->render.chains.render_timer;
struct buffer *buf = shm_get_buffer(chain, width, height);
wl_subsurface_set_position(
win->render_timer.sub,
roundf(margin / scale),
roundf((term->margins.top + term->cell_height - margin) / scale));
render_osd(
term,
&win->render_timer,
term->fonts[0], buf, text,
term->colors.table[0], 0xffu << 24 | term->colors.table[8 + 1],
margin);
}
static void frame_callback(
void *data, struct wl_callback *wl_callback, uint32_t callback_data);
static const struct wl_callback_listener frame_listener = {
.done = &frame_callback,
};
static void
force_full_repaint(struct terminal *term, struct buffer *buf)
{
tll_free(term->grid->scroll_damage);
render_margin(term, buf, 0, term->rows, true);
term_damage_view(term);
}
static void
reapply_old_damage(struct terminal *term, struct buffer *new, struct buffer *old)
{
static int counter = 0;
static bool have_warned = false;
if (!have_warned && ++counter > 5) {
LOG_WARN("compositor is not releasing buffers immediately; "
"expect lower rendering performance");
have_warned = true;
}
if (new->age > 1) {
memcpy(new->data, old->data, new->height * new->stride);
return;
}
pixman_region32_t dirty;
pixman_region32_init(&dirty);
/*
* Figure out current frames damage region
*
* If current frame doesnt have any scroll damage, we can simply
* subtract this frames damage from the last frames damage. That
* way, we dont have to copy areas from the old frame thatll
* just get overwritten by current frame.
*
* Note that this is row based. A half damaged row is not
* excluded. I.e. the entire row will be copied from the old frame
* to the new, and then when actually rendering the new frame, the
* updated cells will overwrite parts of the copied row.
*
* Since were scanning the entire viewport anyway, we also track
* whether *all* cells are to be updated. In this case, just force
* a full re-rendering, and dont copy anything from the old
* frame.
*/
bool full_repaint_needed = true;
for (int r = 0; r < term->rows; r++) {
const struct row *row = grid_row_in_view(term->grid, r);
bool row_all_dirty = true;
for (int c = 0; c < term->cols; c++) {
if (row->cells[c].attrs.clean) {
row_all_dirty = false;
full_repaint_needed = false;
break;
}
}
if (row_all_dirty) {
pixman_region32_union_rect(
&dirty, &dirty,
term->margins.left,
term->margins.top + r * term->cell_height,
term->width - term->margins.left - term->margins.right,
term->cell_height);
}
}
if (full_repaint_needed) {
force_full_repaint(term, new);
return;
}
/*
* TODO: re-apply last frames scroll damage
*
* We used to do this, but it turned out to be buggy. If we decide
* to re-add it, this is where to do it. Note that wed also have
* to remove the updates to buf->dirty from grid_render_scroll()
* and grid_render_scroll_reverse().
*/
if (tll_length(term->grid->scroll_damage) == 0) {
/*
* We can only subtract current frames damage from the old
* frames if we dont have any scroll damage.
*
* If we do have scroll damage, the damage region we
* calculated above is not (yet) valid - we need to apply the
* current frames scroll damage *first*. This is done later,
* when rendering the frame.
*/
pixman_region32_subtract(&dirty, &old->dirty[0], &dirty);
pixman_image_set_clip_region32(new->pix[0], &dirty);
} else {
/* Copy *all* of last frames damaged areas */
pixman_image_set_clip_region32(new->pix[0], &old->dirty[0]);
}
pixman_image_composite32(
PIXMAN_OP_SRC, old->pix[0], NULL, new->pix[0],
0, 0, 0, 0, 0, 0, term->width, term->height);
pixman_image_set_clip_region32(new->pix[0], NULL);
pixman_region32_fini(&dirty);
}
static void
dirty_old_cursor(struct terminal *term)
{
render: wip: re-apply last frame’s damage when forced to double buffer When we are forced to swap between two buffers, re-apply the old frame’s damage to the current buffer, before applying the current frame’s damage. First, while applying this frame’s scroll damage, copy it to the buffer’s scroll damage list (so that we can access it via term->render.last_buf). Also, when iterating and rendering the grid, build a pixman region of the damaged regions. This is currently done on a per-row basis. This is also stored in the buffer. Now, when being forced to double buffer, first iterate the old buffer’s damage, and re-apply it to the current buffer. Then, composite the old buffer on top of the current buffer, using the old frame’s damage region as clip region. This effectively copies everything that was rendered to the last frame. Remember, this is on a per-row basis. Then we go on and render the frame as usual. Note that it would be _really_ nice if we could subtract the current frame’s damage region from the clip region (no point in copying areas we’re going to overwrite anyway). Unfortunately, that’s harder than it looks; the current frame’s damage region is only valid *after* this frame’s scroll damage have been applied, while the last frame’s damage region is only valid *before* it’s been applied. Translating one to the other isn’t easy, since scroll damage isn’t just about counting lines - there may be multiple scroll damage records, each with its own scrolling region. This creates very complex scenarios.
2021-05-07 20:21:27 +02:00
if (term->render.last_cursor.row != NULL && !term->render.last_cursor.hidden) {
struct row *row = term->render.last_cursor.row;
struct cell *cell = &row->cells[term->render.last_cursor.col];
cell->attrs.clean = 0;
row->dirty = true;
}
/* Remember current cursor position, for the next frame */
term->render.last_cursor.row = grid_row(term->grid, term->grid->cursor.point.row);
term->render.last_cursor.col = term->grid->cursor.point.col;
term->render.last_cursor.hidden = term->hide_cursor;
}
render: wip: re-apply last frame’s damage when forced to double buffer When we are forced to swap between two buffers, re-apply the old frame’s damage to the current buffer, before applying the current frame’s damage. First, while applying this frame’s scroll damage, copy it to the buffer’s scroll damage list (so that we can access it via term->render.last_buf). Also, when iterating and rendering the grid, build a pixman region of the damaged regions. This is currently done on a per-row basis. This is also stored in the buffer. Now, when being forced to double buffer, first iterate the old buffer’s damage, and re-apply it to the current buffer. Then, composite the old buffer on top of the current buffer, using the old frame’s damage region as clip region. This effectively copies everything that was rendered to the last frame. Remember, this is on a per-row basis. Then we go on and render the frame as usual. Note that it would be _really_ nice if we could subtract the current frame’s damage region from the clip region (no point in copying areas we’re going to overwrite anyway). Unfortunately, that’s harder than it looks; the current frame’s damage region is only valid *after* this frame’s scroll damage have been applied, while the last frame’s damage region is only valid *before* it’s been applied. Translating one to the other isn’t easy, since scroll damage isn’t just about counting lines - there may be multiple scroll damage records, each with its own scrolling region. This creates very complex scenarios.
2021-05-07 20:21:27 +02:00
static void
dirty_cursor(struct terminal *term)
{
if (term->hide_cursor)
return;
render: wip: re-apply last frame’s damage when forced to double buffer When we are forced to swap between two buffers, re-apply the old frame’s damage to the current buffer, before applying the current frame’s damage. First, while applying this frame’s scroll damage, copy it to the buffer’s scroll damage list (so that we can access it via term->render.last_buf). Also, when iterating and rendering the grid, build a pixman region of the damaged regions. This is currently done on a per-row basis. This is also stored in the buffer. Now, when being forced to double buffer, first iterate the old buffer’s damage, and re-apply it to the current buffer. Then, composite the old buffer on top of the current buffer, using the old frame’s damage region as clip region. This effectively copies everything that was rendered to the last frame. Remember, this is on a per-row basis. Then we go on and render the frame as usual. Note that it would be _really_ nice if we could subtract the current frame’s damage region from the clip region (no point in copying areas we’re going to overwrite anyway). Unfortunately, that’s harder than it looks; the current frame’s damage region is only valid *after* this frame’s scroll damage have been applied, while the last frame’s damage region is only valid *before* it’s been applied. Translating one to the other isn’t easy, since scroll damage isn’t just about counting lines - there may be multiple scroll damage records, each with its own scrolling region. This creates very complex scenarios.
2021-05-07 20:21:27 +02:00
const struct coord *cursor = &term->grid->cursor.point;
struct row *row = grid_row(term->grid, cursor->row);
struct cell *cell = &row->cells[cursor->col];
cell->attrs.clean = 0;
row->dirty = true;
}
static void
grid_render(struct terminal *term)
{
if (term->shutdown.in_progress)
return;
render: wip: re-apply last frame’s damage when forced to double buffer When we are forced to swap between two buffers, re-apply the old frame’s damage to the current buffer, before applying the current frame’s damage. First, while applying this frame’s scroll damage, copy it to the buffer’s scroll damage list (so that we can access it via term->render.last_buf). Also, when iterating and rendering the grid, build a pixman region of the damaged regions. This is currently done on a per-row basis. This is also stored in the buffer. Now, when being forced to double buffer, first iterate the old buffer’s damage, and re-apply it to the current buffer. Then, composite the old buffer on top of the current buffer, using the old frame’s damage region as clip region. This effectively copies everything that was rendered to the last frame. Remember, this is on a per-row basis. Then we go on and render the frame as usual. Note that it would be _really_ nice if we could subtract the current frame’s damage region from the clip region (no point in copying areas we’re going to overwrite anyway). Unfortunately, that’s harder than it looks; the current frame’s damage region is only valid *after* this frame’s scroll damage have been applied, while the last frame’s damage region is only valid *before* it’s been applied. Translating one to the other isn’t easy, since scroll damage isn’t just about counting lines - there may be multiple scroll damage records, each with its own scrolling region. This creates very complex scenarios.
2021-05-07 20:21:27 +02:00
struct timespec start_time, start_double_buffering = {0}, stop_double_buffering = {0};
if (term->conf->tweak.render_timer != RENDER_TIMER_NONE)
clock_gettime(CLOCK_MONOTONIC, &start_time);
xassert(term->width > 0);
xassert(term->height > 0);
render: wip: re-apply last frame’s damage when forced to double buffer When we are forced to swap between two buffers, re-apply the old frame’s damage to the current buffer, before applying the current frame’s damage. First, while applying this frame’s scroll damage, copy it to the buffer’s scroll damage list (so that we can access it via term->render.last_buf). Also, when iterating and rendering the grid, build a pixman region of the damaged regions. This is currently done on a per-row basis. This is also stored in the buffer. Now, when being forced to double buffer, first iterate the old buffer’s damage, and re-apply it to the current buffer. Then, composite the old buffer on top of the current buffer, using the old frame’s damage region as clip region. This effectively copies everything that was rendered to the last frame. Remember, this is on a per-row basis. Then we go on and render the frame as usual. Note that it would be _really_ nice if we could subtract the current frame’s damage region from the clip region (no point in copying areas we’re going to overwrite anyway). Unfortunately, that’s harder than it looks; the current frame’s damage region is only valid *after* this frame’s scroll damage have been applied, while the last frame’s damage region is only valid *before* it’s been applied. Translating one to the other isn’t easy, since scroll damage isn’t just about counting lines - there may be multiple scroll damage records, each with its own scrolling region. This creates very complex scenarios.
2021-05-07 20:21:27 +02:00
struct buffer_chain *chain = term->render.chains.grid;
struct buffer *buf = shm_get_buffer(chain, term->width, term->height);
render: wip: re-apply last frame’s damage when forced to double buffer When we are forced to swap between two buffers, re-apply the old frame’s damage to the current buffer, before applying the current frame’s damage. First, while applying this frame’s scroll damage, copy it to the buffer’s scroll damage list (so that we can access it via term->render.last_buf). Also, when iterating and rendering the grid, build a pixman region of the damaged regions. This is currently done on a per-row basis. This is also stored in the buffer. Now, when being forced to double buffer, first iterate the old buffer’s damage, and re-apply it to the current buffer. Then, composite the old buffer on top of the current buffer, using the old frame’s damage region as clip region. This effectively copies everything that was rendered to the last frame. Remember, this is on a per-row basis. Then we go on and render the frame as usual. Note that it would be _really_ nice if we could subtract the current frame’s damage region from the clip region (no point in copying areas we’re going to overwrite anyway). Unfortunately, that’s harder than it looks; the current frame’s damage region is only valid *after* this frame’s scroll damage have been applied, while the last frame’s damage region is only valid *before* it’s been applied. Translating one to the other isn’t easy, since scroll damage isn’t just about counting lines - there may be multiple scroll damage records, each with its own scrolling region. This creates very complex scenarios.
2021-05-07 20:21:27 +02:00
/* Dirty old and current cursor cell, to ensure theyre repainted */
dirty_old_cursor(term);
dirty_cursor(term);
render: wip: re-apply last frame’s damage when forced to double buffer When we are forced to swap between two buffers, re-apply the old frame’s damage to the current buffer, before applying the current frame’s damage. First, while applying this frame’s scroll damage, copy it to the buffer’s scroll damage list (so that we can access it via term->render.last_buf). Also, when iterating and rendering the grid, build a pixman region of the damaged regions. This is currently done on a per-row basis. This is also stored in the buffer. Now, when being forced to double buffer, first iterate the old buffer’s damage, and re-apply it to the current buffer. Then, composite the old buffer on top of the current buffer, using the old frame’s damage region as clip region. This effectively copies everything that was rendered to the last frame. Remember, this is on a per-row basis. Then we go on and render the frame as usual. Note that it would be _really_ nice if we could subtract the current frame’s damage region from the clip region (no point in copying areas we’re going to overwrite anyway). Unfortunately, that’s harder than it looks; the current frame’s damage region is only valid *after* this frame’s scroll damage have been applied, while the last frame’s damage region is only valid *before* it’s been applied. Translating one to the other isn’t easy, since scroll damage isn’t just about counting lines - there may be multiple scroll damage records, each with its own scrolling region. This creates very complex scenarios.
2021-05-07 20:21:27 +02:00
if (term->render.last_buf == NULL ||
term->render.last_buf->width != buf->width ||
term->render.last_buf->height != buf->height ||
term->render.margins)
{
force_full_repaint(term, buf);
}
render: wip: re-apply last frame’s damage when forced to double buffer When we are forced to swap between two buffers, re-apply the old frame’s damage to the current buffer, before applying the current frame’s damage. First, while applying this frame’s scroll damage, copy it to the buffer’s scroll damage list (so that we can access it via term->render.last_buf). Also, when iterating and rendering the grid, build a pixman region of the damaged regions. This is currently done on a per-row basis. This is also stored in the buffer. Now, when being forced to double buffer, first iterate the old buffer’s damage, and re-apply it to the current buffer. Then, composite the old buffer on top of the current buffer, using the old frame’s damage region as clip region. This effectively copies everything that was rendered to the last frame. Remember, this is on a per-row basis. Then we go on and render the frame as usual. Note that it would be _really_ nice if we could subtract the current frame’s damage region from the clip region (no point in copying areas we’re going to overwrite anyway). Unfortunately, that’s harder than it looks; the current frame’s damage region is only valid *after* this frame’s scroll damage have been applied, while the last frame’s damage region is only valid *before* it’s been applied. Translating one to the other isn’t easy, since scroll damage isn’t just about counting lines - there may be multiple scroll damage records, each with its own scrolling region. This creates very complex scenarios.
2021-05-07 20:21:27 +02:00
else if (buf->age > 0) {
LOG_DBG("buffer age: %u (%p)", buf->age, (void *)buf);
xassert(term->render.last_buf != NULL);
xassert(term->render.last_buf != buf);
xassert(term->render.last_buf->width == buf->width);
xassert(term->render.last_buf->height == buf->height);
render: wip: re-apply last frame’s damage when forced to double buffer When we are forced to swap between two buffers, re-apply the old frame’s damage to the current buffer, before applying the current frame’s damage. First, while applying this frame’s scroll damage, copy it to the buffer’s scroll damage list (so that we can access it via term->render.last_buf). Also, when iterating and rendering the grid, build a pixman region of the damaged regions. This is currently done on a per-row basis. This is also stored in the buffer. Now, when being forced to double buffer, first iterate the old buffer’s damage, and re-apply it to the current buffer. Then, composite the old buffer on top of the current buffer, using the old frame’s damage region as clip region. This effectively copies everything that was rendered to the last frame. Remember, this is on a per-row basis. Then we go on and render the frame as usual. Note that it would be _really_ nice if we could subtract the current frame’s damage region from the clip region (no point in copying areas we’re going to overwrite anyway). Unfortunately, that’s harder than it looks; the current frame’s damage region is only valid *after* this frame’s scroll damage have been applied, while the last frame’s damage region is only valid *before* it’s been applied. Translating one to the other isn’t easy, since scroll damage isn’t just about counting lines - there may be multiple scroll damage records, each with its own scrolling region. This creates very complex scenarios.
2021-05-07 20:21:27 +02:00
clock_gettime(CLOCK_MONOTONIC, &start_double_buffering);
reapply_old_damage(term, buf, term->render.last_buf);
clock_gettime(CLOCK_MONOTONIC, &stop_double_buffering);
}
render: wip: re-apply last frame’s damage when forced to double buffer When we are forced to swap between two buffers, re-apply the old frame’s damage to the current buffer, before applying the current frame’s damage. First, while applying this frame’s scroll damage, copy it to the buffer’s scroll damage list (so that we can access it via term->render.last_buf). Also, when iterating and rendering the grid, build a pixman region of the damaged regions. This is currently done on a per-row basis. This is also stored in the buffer. Now, when being forced to double buffer, first iterate the old buffer’s damage, and re-apply it to the current buffer. Then, composite the old buffer on top of the current buffer, using the old frame’s damage region as clip region. This effectively copies everything that was rendered to the last frame. Remember, this is on a per-row basis. Then we go on and render the frame as usual. Note that it would be _really_ nice if we could subtract the current frame’s damage region from the clip region (no point in copying areas we’re going to overwrite anyway). Unfortunately, that’s harder than it looks; the current frame’s damage region is only valid *after* this frame’s scroll damage have been applied, while the last frame’s damage region is only valid *before* it’s been applied. Translating one to the other isn’t easy, since scroll damage isn’t just about counting lines - there may be multiple scroll damage records, each with its own scrolling region. This creates very complex scenarios.
2021-05-07 20:21:27 +02:00
if (term->render.last_buf != NULL) {
shm_unref(term->render.last_buf);
term->render.last_buf = NULL;
}
render: wip: re-apply last frame’s damage when forced to double buffer When we are forced to swap between two buffers, re-apply the old frame’s damage to the current buffer, before applying the current frame’s damage. First, while applying this frame’s scroll damage, copy it to the buffer’s scroll damage list (so that we can access it via term->render.last_buf). Also, when iterating and rendering the grid, build a pixman region of the damaged regions. This is currently done on a per-row basis. This is also stored in the buffer. Now, when being forced to double buffer, first iterate the old buffer’s damage, and re-apply it to the current buffer. Then, composite the old buffer on top of the current buffer, using the old frame’s damage region as clip region. This effectively copies everything that was rendered to the last frame. Remember, this is on a per-row basis. Then we go on and render the frame as usual. Note that it would be _really_ nice if we could subtract the current frame’s damage region from the clip region (no point in copying areas we’re going to overwrite anyway). Unfortunately, that’s harder than it looks; the current frame’s damage region is only valid *after* this frame’s scroll damage have been applied, while the last frame’s damage region is only valid *before* it’s been applied. Translating one to the other isn’t easy, since scroll damage isn’t just about counting lines - there may be multiple scroll damage records, each with its own scrolling region. This creates very complex scenarios.
2021-05-07 20:21:27 +02:00
term->render.last_buf = buf;
shm_addref(buf);
buf->age = 0;
render: wip: re-apply last frame’s damage when forced to double buffer When we are forced to swap between two buffers, re-apply the old frame’s damage to the current buffer, before applying the current frame’s damage. First, while applying this frame’s scroll damage, copy it to the buffer’s scroll damage list (so that we can access it via term->render.last_buf). Also, when iterating and rendering the grid, build a pixman region of the damaged regions. This is currently done on a per-row basis. This is also stored in the buffer. Now, when being forced to double buffer, first iterate the old buffer’s damage, and re-apply it to the current buffer. Then, composite the old buffer on top of the current buffer, using the old frame’s damage region as clip region. This effectively copies everything that was rendered to the last frame. Remember, this is on a per-row basis. Then we go on and render the frame as usual. Note that it would be _really_ nice if we could subtract the current frame’s damage region from the clip region (no point in copying areas we’re going to overwrite anyway). Unfortunately, that’s harder than it looks; the current frame’s damage region is only valid *after* this frame’s scroll damage have been applied, while the last frame’s damage region is only valid *before* it’s been applied. Translating one to the other isn’t easy, since scroll damage isn’t just about counting lines - there may be multiple scroll damage records, each with its own scrolling region. This creates very complex scenarios.
2021-05-07 20:21:27 +02:00
tll_foreach(term->grid->scroll_damage, it) {
switch (it->item.type) {
case DAMAGE_SCROLL:
if (term->grid->view == term->grid->offset)
grid_render_scroll(term, buf, &it->item);
break;
case DAMAGE_SCROLL_REVERSE:
if (term->grid->view == term->grid->offset)
grid_render_scroll_reverse(term, buf, &it->item);
break;
case DAMAGE_SCROLL_IN_VIEW:
grid_render_scroll(term, buf, &it->item);
break;
case DAMAGE_SCROLL_REVERSE_IN_VIEW:
grid_render_scroll_reverse(term, buf, &it->item);
break;
}
tll_remove(term->grid->scroll_damage, it);
}
/*
* Ensure selected cells have their 'selected' bit set. This is
* normally "automatically" true - the bit is set when the
* selection is made.
*
* However, if the cell is updated (printed to) while the
* selection is active, the 'selected' bit is cleared. Checking
* for this and re-setting the bit in term_print() is too
* expensive performance wise.
*
* Instead, we synchronize the selection bits here and now. This
* makes the performance impact linear to the number of selected
* cells rather than to the number of updated cells.
*
* (note that selection_dirty_cells() will not set the dirty flag
* on cells where the 'selected' bit is already set)
*/
selection_dirty_cells(term);
/* Translate offset-relative row to view-relative, unless cursor
* is hidden, then we just set it to -1 */
struct coord cursor = {-1, -1};
if (!term->hide_cursor) {
cursor = term->grid->cursor.point;
cursor.row += term->grid->offset;
cursor.row -= term->grid->view;
cursor.row &= term->grid->num_rows - 1;
}
2019-07-29 20:13:26 +02:00
if (term->conf->tweak.overflowing_glyphs) {
/*
* Pre-pass to dirty cells affected by overflowing glyphs.
*
* Given any two pair of cells where the first cell is
* overflowing into the second, *both* cells must be
* re-rendered if any one of them is dirty.
*
* Thus, given a string of overflowing glyphs, with a single
* dirty cell in the middle, we need to re-render the entire
* string.
*/
for (int r = 0; r < term->rows; r++) {
struct row *row = grid_row_in_view(term->grid, r);
if (!row->dirty)
continue;
/* Loop row from left to right, looking for dirty cells */
for (struct cell *cell = &row->cells[0];
cell < &row->cells[term->cols];
cell++)
{
if (cell->attrs.clean)
continue;
/*
* Cell is dirty, go back and dirty previous cells, if
* they are overflowing.
*
* As soon as we see a non-overflowing cell we can
* stop, since it isnt affecting the string of
* overflowing glyphs that follows it.
*
* As soon as we see a dirty cell, we can stop, since
* that means weve already handled it (remember the
* outer loop goes from left to right).
*/
for (struct cell *c = cell - 1; c >= &row->cells[0]; c--) {
if (c->attrs.confined)
break;
if (!c->attrs.clean)
break;
c->attrs.clean = false;
}
/*
* Now move forward, dirtying all cells until we hit a
* non-overflowing cell.
*
* Note that the first non-overflowing cell must be
* re-rendered as well, but any cell *after* that is
* unaffected by the string of overflowing glyphs
* were dealing with right now.
*
* For performance, this iterates the *outer* loops
* cell pointer - no point in re-checking all these
* glyphs again, in the outer loop.
*/
for (; cell < &row->cells[term->cols]; cell++) {
cell->attrs.clean = false;
if (cell->attrs.confined)
break;
}
}
}
}
pixman_region32_t damage;
pixman_region32_init(&damage);
render_sixel_images(term, buf->pix[0], &damage, &cursor);
if (term->render.workers.count > 0) {
mtx_lock(&term->render.workers.lock);
term->render.workers.buf = buf;
for (size_t i = 0; i < term->render.workers.count; i++)
sem_post(&term->render.workers.start);
2019-07-29 20:13:26 +02:00
xassert(tll_length(term->render.workers.queue) == 0);
}
for (int r = 0; r < term->rows; r++) {
struct row *row = grid_row_in_view(term->grid, r);
if (!row->dirty)
continue;
row->dirty = false;
if (term->render.workers.count > 0)
tll_push_back(term->render.workers.queue, r);
else {
/* TODO: damage region */
int cursor_col = cursor.row == r ? cursor.col : -1;
render_row(term, buf->pix[0], &damage, row, r, cursor_col);
}
}
/* Signal workers the frame is done */
if (term->render.workers.count > 0) {
for (size_t i = 0; i < term->render.workers.count; i++)
tll_push_back(term->render.workers.queue, -1);
2019-07-29 20:13:26 +02:00
mtx_unlock(&term->render.workers.lock);
for (size_t i = 0; i < term->render.workers.count; i++)
sem_wait(&term->render.workers.done);
term->render.workers.buf = NULL;
}
2019-07-29 20:13:26 +02:00
for (size_t i = 0; i < term->render.workers.count; i++)
pixman_region32_union(&damage, &damage, &buf->dirty[i + 1]);
pixman_region32_union(&buf->dirty[0], &buf->dirty[0], &damage);
{
int box_count = 0;
pixman_box32_t *boxes = pixman_region32_rectangles(&damage, &box_count);
for (size_t i = 0; i < box_count; i++) {
wl_surface_damage_buffer(
term->window->surface.surf,
boxes[i].x1, boxes[i].y1,
boxes[i].x2 - boxes[i].x1, boxes[i].y2 - boxes[i].y1);
}
}
pixman_region32_fini(&damage);
render: implement ‘flash’ and search mode’s ‘dimming’ with a sub-surface Search mode and ‘flash’ (OSC-555) both achieves similar visual effects: flash tints the entire window yellow, and search mode dims it (except the search match). But, they do so in completely different ways. Search mode is detected in render_cell(), and the colors are then dimmed there. Flash is implemented by blending a yellow, semi-transparent color on top of the rendered grid. This patch replaces those two implementations with a single one. We add a new sub-surface, called the ‘overlay’. In normal mode, it’s unmapped. When either search mode, or flash, is enabled, we enable it, and fill it with a semi-transparent color. Yellow for ‘flash’, and “black” (i.e. no color) for search mode. The compositor then blends it with the grid. Hopefully on the GPU, meaning it’ll be faster than if we blend in software. There are more performance benefits however. By using a separate surface, we can do much better damage tracking. The normal grid rendering code no longer have to care about neither search mode, nor flash. Thus, we get rid of a couple of ‘if’ statements in render_cell(), which is nice. But more importantly, we can drop full grid repaints in a couple of circumstances: * Entering/exiting search mode * Every frame while flash is active Now, when rendering the search mode overlay, we do want to do some damage tracking, also of the overlay. This, since search mode doesn’t dim the *entire* window. The search match is *not* dimmed. This is implemented by punching a hole in the overlay sub-surface. That is, we make part of it *fully* transparent. The basic idea is to set a clip region that excludes the search match, and then dim the rest of the overlay. It’s slightly more complicated than that however, if we want to reuse the last frame’s overlay buffer (i.e we don’t want to re-render the *entire* overlay every frame). In short, we need to: * Clear (punch hole) in areas that are part of this frame’s search match, but not the last frame’s (since those parts are _already_ cleared). * Dim the areas that were part of the last frame’s search match, but aren’t anymore (the rest of the overlay should already be dimmed). To do this, we save the last frame’s “holes” (as a pixman region). Then, when rendering the next frame, we first calculate the new frame’s “holes” region. The region to clear is “this frame’s holes minus last frame’s holes” The region to dim is “last frame’s holes minus this frames holes”. Finally, we compute the bounding box of all modified cells by taking the union of the two diff regions mentioned above. This allows us to limit the buffer damage sent to the compositor.
2022-04-16 17:49:46 +02:00
render_overlay(term);
render_ime_preedit(term, buf);
render_scrollback_position(term);
if (term->conf->tweak.render_timer != RENDER_TIMER_NONE) {
struct timespec end_time;
clock_gettime(CLOCK_MONOTONIC, &end_time);
struct timespec render_time;
timespec_sub(&end_time, &start_time, &render_time);
struct timespec double_buffering_time;
timespec_sub(&stop_double_buffering, &start_double_buffering, &double_buffering_time);
struct timespec total_render_time;
timespec_add(&render_time, &double_buffering_time, &total_render_time);
switch (term->conf->tweak.render_timer) {
case RENDER_TIMER_LOG:
case RENDER_TIMER_BOTH:
LOG_INFO(
"frame rendered in %lds %9ldns "
"(%lds %9ldns rendering, %lds %9ldns double buffering)",
(long)total_render_time.tv_sec,
total_render_time.tv_nsec,
(long)render_time.tv_sec,
render_time.tv_nsec,
(long)double_buffering_time.tv_sec,
double_buffering_time.tv_nsec);
break;
case RENDER_TIMER_OSD:
case RENDER_TIMER_NONE:
break;
}
switch (term->conf->tweak.render_timer) {
case RENDER_TIMER_OSD:
case RENDER_TIMER_BOTH:
render_render_timer(term, total_render_time);
break;
case RENDER_TIMER_LOG:
case RENDER_TIMER_NONE:
break;
}
}
xassert(term->grid->offset >= 0 && term->grid->offset < term->grid->num_rows);
xassert(term->grid->view >= 0 && term->grid->view < term->grid->num_rows);
xassert(term->window->frame_callback == NULL);
term->window->frame_callback = wl_surface_frame(term->window->surface.surf);
wl_callback_add_listener(term->window->frame_callback, &frame_listener, term);
wayl_win_scale(term->window, buf);
if (term->wl->presentation != NULL && term->conf->presentation_timings) {
struct timespec commit_time;
clock_gettime(term->wl->presentation_clock_id, &commit_time);
struct wp_presentation_feedback *feedback = wp_presentation_feedback(
term->wl->presentation, term->window->surface.surf);
if (feedback == NULL) {
LOG_WARN("failed to create presentation feedback");
} else {
struct presentation_context *ctx = xmalloc(sizeof(*ctx));
*ctx = (struct presentation_context){
.term = term,
.input.tv_sec = term->render.input_time.tv_sec,
.input.tv_usec = term->render.input_time.tv_nsec / 1000,
.commit.tv_sec = commit_time.tv_sec,
.commit.tv_usec = commit_time.tv_nsec / 1000,
};
wp_presentation_feedback_add_listener(
feedback, &presentation_feedback_listener, ctx);
term->render.input_time.tv_sec = 0;
term->render.input_time.tv_nsec = 0;
}
}
if (term->conf->tweak.damage_whole_window) {
wl_surface_damage_buffer(
term->window->surface.surf, 0, 0, INT32_MAX, INT32_MAX);
}
wl_surface_attach(term->window->surface.surf, buf->wl_buf, 0, 0);
wl_surface_commit(term->window->surface.surf);
}
static void
render_search_box(struct terminal *term)
{
xassert(term->window->search.sub != NULL);
/*
* We treat the search box pretty much like a row of cells. That
* is, a glyph is either 1 or 2 (or more) cells wide.
*
* The search length, and cursor (position) is in
* *characters*, not cells. This means we need to translate from
2020-12-05 23:34:27 +01:00
* character count to cell count when calculating the length of
* the search box, where in the search string we should start
* rendering etc.
*/
2020-12-06 12:18:46 +01:00
#if defined(FOOT_IME_ENABLED) && FOOT_IME_ENABLED
/* TODO: do we want to/need to handle multi-seat? */
struct seat *ime_seat = NULL;
tll_foreach(term->wl->seats, it) {
if (it->item.kbd_focus == term) {
ime_seat = &it->item;
break;
}
}
2020-12-06 12:18:46 +01:00
size_t text_len = term->search.len;
if (ime_seat != NULL && ime_seat->ime.preedit.text != NULL)
text_len += c32len(ime_seat->ime.preedit.text);
2020-12-06 12:18:46 +01:00
char32_t *text = xmalloc((text_len + 1) * sizeof(char32_t));
text[0] = U'\0';
/* Copy everything up to the cursor */
c32ncpy(text, term->search.buf, term->search.cursor);
text[term->search.cursor] = U'\0';
/* Insert pre-edit text at cursor */
if (ime_seat != NULL && ime_seat->ime.preedit.text != NULL)
c32cat(text, ime_seat->ime.preedit.text);
/* And finally everything after the cursor */
c32ncat(text, &term->search.buf[term->search.cursor],
term->search.len - term->search.cursor);
2020-12-06 12:18:46 +01:00
#else
const char32_t *text = term->search.buf;
const size_t text_len = term->search.len;
2020-12-06 12:18:46 +01:00
#endif
/* Calculate the width of each character */
int widths[text_len + 1];
for (size_t i = 0; i < text_len; i++)
widths[i] = max(0, c32width(text[i]));
widths[text_len] = 0;
const size_t total_cells = c32swidth(text, text_len);
const size_t wanted_visible_cells = max(20, total_cells);
const float scale = term->scale;
xassert(scale >= 1.);
const size_t margin = (size_t)roundf(3 * scale);
size_t width = term->width - 2 * margin;
size_t height = min(
term->height - 2 * margin,
margin + 1 * term->cell_height + margin);
width = roundf(scale * ceilf((term->width - 2 * margin) / scale));
height = roundf(scale * ceilf(height / scale));
size_t visible_width = min(
term->width - 2 * margin,
margin + wanted_visible_cells * term->cell_width + margin);
const size_t visible_cells = (visible_width - 2 * margin) / term->cell_width;
size_t glyph_offset = term->render.search_glyph_offset;
struct buffer_chain *chain = term->render.chains.search;
struct buffer *buf = shm_get_buffer(chain, width, height);
pixman_region32_t clip;
pixman_region32_init_rect(&clip, 0, 0, width, height);
pixman_image_set_clip_region32(buf->pix[0], &clip);
pixman_region32_fini(&clip);
#define WINDOW_X(x) (margin + x)
#define WINDOW_Y(y) (term->height - margin - height + y)
const bool is_match = term->search.match_len == text_len;
const bool custom_colors = is_match
? term->conf->colors.use_custom.search_box_match
: term->conf->colors.use_custom.search_box_no_match;
/* Background - yellow on empty/match, red on mismatch (default) */
const pixman_color_t color = color_hex_to_pixman(
is_match
? (custom_colors
? term->conf->colors.search_box.match.bg
: term->colors.table[3])
: (custom_colors
? term->conf->colors.search_box.no_match.bg
: term->colors.table[1]));
pixman_image_fill_rectangles(
PIXMAN_OP_SRC, buf->pix[0], &color,
1, &(pixman_rectangle16_t){width - visible_width, 0, visible_width, height});
pixman_color_t transparent = color_hex_to_pixman_with_alpha(0, 0);
pixman_image_fill_rectangles(
PIXMAN_OP_SRC, buf->pix[0], &transparent,
1, &(pixman_rectangle16_t){0, 0, width - visible_width, height});
struct fcft_font *font = term->fonts[0];
2020-12-06 12:18:46 +01:00
const int x_left = width - visible_width + margin;
const int x_ofs = term->font_x_ofs;
2020-12-06 12:18:46 +01:00
int x = x_left;
int y = margin;
pixman_color_t fg = color_hex_to_pixman(
custom_colors
? (is_match
? term->conf->colors.search_box.match.fg
: term->conf->colors.search_box.no_match.fg)
: term->colors.table[0]);
/* Move offset we start rendering at, to ensure the cursor is visible */
for (size_t i = 0, cell_idx = 0; i <= term->search.cursor; cell_idx += widths[i], i++) {
if (i != term->search.cursor)
continue;
2020-12-06 12:18:46 +01:00
#if (FOOT_IME_ENABLED) && FOOT_IME_ENABLED
if (ime_seat != NULL && ime_seat->ime.preedit.cells != NULL) {
if (ime_seat->ime.preedit.cursor.start == ime_seat->ime.preedit.cursor.end) {
/* All IME's I've seen so far keeps the cursor at
* index 0, so ensure the *end* of the pre-edit string
* is visible */
cell_idx += ime_seat->ime.preedit.count;
} else {
/* Try to predict in which direction we'll shift the text */
if (cell_idx + ime_seat->ime.preedit.cursor.start > glyph_offset)
cell_idx += ime_seat->ime.preedit.cursor.end;
else
cell_idx += ime_seat->ime.preedit.cursor.start;
}
}
2020-12-06 12:18:46 +01:00
#endif
if (cell_idx < glyph_offset) {
/* Shift to the *left*, making *this* character the
* *first* visible one */
term->render.search_glyph_offset = glyph_offset = cell_idx;
}
else if (cell_idx > glyph_offset + visible_cells) {
/* Shift to the *right*, making *this* character the
* *last* visible one */
term->render.search_glyph_offset = glyph_offset =
cell_idx - min(cell_idx, visible_cells);
}
/* Adjust offset if there is free space available */
if (total_cells - glyph_offset < visible_cells) {
term->render.search_glyph_offset = glyph_offset =
total_cells - min(total_cells, visible_cells);
}
break;
}
/* Ensure offset is at a character boundary */
for (size_t i = 0, cell_idx = 0; i <= text_len; cell_idx += widths[i], i++) {
if (cell_idx >= glyph_offset) {
term->render.search_glyph_offset = glyph_offset = cell_idx;
break;
}
}
/*
* Render the search string, starting at glyph_offset. Note that
* glyph_offset is in cells, not characters
*/
for (size_t i = 0,
cell_idx = 0,
width = widths[i],
next_cell_idx = width;
i < text_len;
i++,
cell_idx = next_cell_idx,
width = widths[i],
next_cell_idx += width)
{
/* Convert subsurface coordinates to window coordinates*/
/* Render cursor */
if (i == term->search.cursor) {
#if defined(FOOT_IME_ENABLED) && FOOT_IME_ENABLED
bool have_preedit =
ime_seat != NULL && ime_seat->ime.preedit.cells != NULL;
bool hidden =
ime_seat != NULL && ime_seat->ime.preedit.cursor.hidden;
if (have_preedit && !hidden) {
/* Cursor may be outside the visible area:
* cell_idx-glyph_offset can be negative */
int cells_left = visible_cells - max(
(ssize_t)(cell_idx - glyph_offset), 0);
/* If cursor is outside the visible area, we need to
* adjust our rectangle's position */
int start = ime_seat->ime.preedit.cursor.start
+ min((ssize_t)(cell_idx - glyph_offset), 0);
int end = ime_seat->ime.preedit.cursor.end
+ min((ssize_t)(cell_idx - glyph_offset), 0);
if (start == end) {
int count = min(ime_seat->ime.preedit.count, cells_left);
/* Underline the entire (visible part of) pre-edit text */
draw_underline(term, buf->pix[0], font, &fg, x, y, count);
/* Bar-styled cursor, if in the visible area */
if (start >= 0 && start <= visible_cells) {
draw_beam_cursor(
term, buf->pix[0], font, &fg,
x + start * term->cell_width, y);
}
term_ime_set_cursor_rect(term,
WINDOW_X(x + start * term->cell_width), WINDOW_Y(y),
1, term->cell_height);
} else {
/* Underline everything before and after the cursor */
int count1 = min(start, cells_left);
int count2 = max(
min(ime_seat->ime.preedit.count - ime_seat->ime.preedit.cursor.end,
cells_left - end),
0);
draw_underline(term, buf->pix[0], font, &fg, x, y, count1);
draw_underline(term, buf->pix[0], font, &fg, x + end * term->cell_width, y, count2);
/* TODO: how do we handle a partially hidden rectangle? */
if (start >= 0 && end <= visible_cells) {
draw_unfocused_block(
term, buf->pix[0], &fg, x + start * term->cell_width, y, end - start);
}
term_ime_set_cursor_rect(term,
WINDOW_X(x + start * term->cell_width), WINDOW_Y(y),
term->cell_width * (end - start), term->cell_height);
}
} else if (!have_preedit)
2020-12-06 12:18:46 +01:00
#endif
{
/* Cursor *should* be in the visible area */
xassert(cell_idx >= glyph_offset);
xassert(cell_idx <= glyph_offset + visible_cells);
draw_beam_cursor(term, buf->pix[0], font, &fg, x, y);
term_ime_set_cursor_rect(
term, WINDOW_X(x), WINDOW_Y(y), 1, term->cell_height);
}
}
if (next_cell_idx >= glyph_offset && next_cell_idx - glyph_offset > visible_cells) {
/* We're now beyond the visible area - nothing more to render */
break;
}
if (cell_idx < glyph_offset) {
/* We haven't yet reached the visible part of the string */
cell_idx = next_cell_idx;
continue;
}
const struct fcft_glyph *glyph = fcft_rasterize_char_utf32(
font, text[i], term->font_subpixel);
if (glyph == NULL) {
cell_idx = next_cell_idx;
continue;
}
if (unlikely(pixman_image_get_format(glyph->pix) == PIXMAN_a8r8g8b8)) {
/* Glyph surface is a pre-rendered image (typically a color emoji...) */
pixman_image_composite32(
PIXMAN_OP_OVER, glyph->pix, NULL, buf->pix[0], 0, 0, 0, 0,
x + x_ofs + glyph->x, y + term->font_baseline - glyph->y,
glyph->width, glyph->height);
} else {
int combining_ofs = width == 0
? (glyph->x < 0
? width * term->cell_width
: (width - 1) * term->cell_width)
: 0; /* Not a zero-width character - no additional offset */
pixman_image_t *src = pixman_image_create_solid_fill(&fg);
pixman_image_composite32(
PIXMAN_OP_OVER, src, glyph->pix, buf->pix[0], 0, 0, 0, 0,
x + x_ofs + combining_ofs + glyph->x,
y + term->font_baseline - glyph->y,
glyph->width, glyph->height);
pixman_image_unref(src);
}
x += width * term->cell_width;
cell_idx = next_cell_idx;
}
2020-12-06 12:18:46 +01:00
#if defined(FOOT_IME_ENABLED) && FOOT_IME_ENABLED
if (ime_seat != NULL && ime_seat->ime.preedit.cells != NULL)
/* Already rendered */;
else
2020-12-06 12:18:46 +01:00
#endif
if (term->search.cursor >= term->search.len) {
draw_beam_cursor(term, buf->pix[0], font, &fg, x, y);
term_ime_set_cursor_rect(
term, WINDOW_X(x), WINDOW_Y(y), 1, term->cell_height);
}
quirk_weston_subsurface_desync_on(term->window->search.sub);
/* TODO: this is only necessary on a window resize */
wl_subsurface_set_position(
term->window->search.sub,
roundf(margin / scale),
roundf(max(0, (int32_t)term->height - height - margin) / scale));
wayl_surface_scale(term->window, &term->window->search.surface, buf, scale);
wl_surface_attach(term->window->search.surface.surf, buf->wl_buf, 0, 0);
wl_surface_damage_buffer(term->window->search.surface.surf, 0, 0, width, height);
struct wl_region *region = wl_compositor_create_region(term->wl->compositor);
if (region != NULL) {
wl_region_add(region, width - visible_width, 0, visible_width, height);
wl_surface_set_opaque_region(term->window->search.surface.surf, region);
wl_region_destroy(region);
}
wl_surface_commit(term->window->search.surface.surf);
quirk_weston_subsurface_desync_off(term->window->search.sub);
2020-12-06 12:18:46 +01:00
#if defined(FOOT_IME_ENABLED) && FOOT_IME_ENABLED
free(text);
#endif
#undef WINDOW_X
#undef WINDOW_Y
}
static void
render_urls(struct terminal *term)
{
struct wl_window *win = term->window;
xassert(tll_length(win->urls) > 0);
const float scale = term->scale;
const int x_margin = (int)roundf(2 * scale);
const int y_margin = (int)roundf(1 * scale);
/* Calculate view start, counted from the *current* scrollback start */
const int scrollback_end
= (term->grid->offset + term->rows) & (term->grid->num_rows - 1);
const int view_start
= (term->grid->view
- scrollback_end
+ term->grid->num_rows) & (term->grid->num_rows - 1);
const int view_end = view_start + term->rows - 1;
const bool show_url = term->urls_show_uri_on_jump_label;
/*
* There can potentially be a lot of URLs.
*
* Since each URL is a separate sub-surface, and requires its own
* SHM buffer, we may be allocating a lot of buffers.
*
* SHM buffers normally have their own, private SHM buffer
* pool. Each pool is mmapped, and thus allocates *at least*
* 4K. Since URL labels are typically small, we end up using an
* excessive amount of both virtual and physical memory.
*
* For this reason, we instead use shm_get_many(), which uses a
* single, shared pool for all buffers.
*
* To be able to use it, we need to have all the *all* the buffer
* dimensions up front.
*
* Thus, the first iteration through the URLs do the heavy
* lifting: builds the label contents and calculates both its
* position and size. But instead of rendering the label
* immediately, we store the calculated data, and then do a second
* pass, where we first get all our buffers, and then render to
* them.
*/
/* Positioning data + label contents */
struct {
const struct wl_url *url;
char32_t *text;
int x;
int y;
} info[tll_length(win->urls)];
/* For shm_get_many() */
int widths[tll_length(win->urls)];
int heights[tll_length(win->urls)];
size_t render_count = 0;
tll_foreach(win->urls, it) {
const struct url *url = it->item.url;
const char32_t *key = url->key;
const size_t entered_key_len = c32len(term->url_keys);
if (key == NULL) {
/* TODO: if we decide to use the .text field, we cannot
* just skip the entire jump label like this */
continue;
}
struct wl_surface *surf = it->item.surf.surface.surf;
struct wl_subsurface *sub_surf = it->item.surf.sub;
if (surf == NULL || sub_surf == NULL)
continue;
bool hide = false;
const struct coord *pos = &url->range.start;
const int _row
= (pos->row
- scrollback_end
+ term->grid->num_rows) & (term->grid->num_rows - 1);
if (_row < view_start || _row > view_end)
hide = true;
if (c32len(key) <= entered_key_len)
hide = true;
if (c32ncasecmp(term->url_keys, key, entered_key_len) != 0)
hide = true;
if (hide) {
wl_surface_attach(surf, NULL, 0, 0);
wl_surface_commit(surf);
continue;
}
int col = pos->col;
int row = pos->row - term->grid->view;
while (row < 0)
row += term->grid->num_rows;
row &= (term->grid->num_rows - 1);
/* Position label slightly above and to the left */
int x = col * term->cell_width - 15 * term->cell_width / 10;
int y = row * term->cell_height - 5 * term->cell_height / 10;
/* Dont position it outside our window */
if (x < -term->margins.left)
x = -term->margins.left;
if (y < -term->margins.top)
y = -term->margins.top;
/* Maximum width of label, in pixels */
const int max_width =
term->width - term->margins.left - term->margins.right - x;
const int max_cols = max_width / term->cell_width;
const size_t key_len = c32len(key);
size_t url_len = mbstoc32(NULL, url->url, 0);
if (url_len == (size_t)-1)
url_len = 0;
char32_t url_wchars[url_len + 1];
mbstoc32(url_wchars, url->url, url_len + 1);
/* Format label, not yet subject to any size limitations */
size_t chars = key_len + (show_url ? (2 + url_len) : 0);
char32_t label[chars + 1];
label[chars] = U'\0';
if (show_url) {
c32cpy(label, key);
c32cat(label, U": ");
c32cat(label, url_wchars);
} else
c32ncpy(label, key, chars);
/* Upper case the key characters */
for (size_t i = 0; i < c32len(key); i++)
label[i] = toc32upper(label[i]);
/* Blank already entered key characters */
for (size_t i = 0; i < entered_key_len; i++)
label[i] = U' ';
/*
* Dont extend outside our window
*
* Truncate label so that it doesnt extend outside our
* window.
*
* Do it in a way such that we dont cut the label in the
* middle of a double-width character.
*/
int cols = 0;
for (size_t i = 0; i <= c32len(label); i++) {
int _cols = c32swidth(label, i);
if (_cols == (size_t)-1)
continue;
if (_cols >= max_cols) {
if (i > 0)
label[i - 1] = U'';
label[i] = U'\0';
cols = max_cols;
break;
}
cols = _cols;
}
if (cols == 0)
continue;
int width = x_margin + cols * term->cell_width + x_margin;
int height = y_margin + term->cell_height + y_margin;
width = roundf(scale * ceilf(width / scale));
height = roundf(scale * ceilf(height / scale));
info[render_count].url = &it->item;
info[render_count].text = xc32dup(label);
info[render_count].x = x;
info[render_count].y = y;
widths[render_count] = width;
heights[render_count] = height;
render_count++;
}
struct buffer_chain *chain = term->render.chains.url;
struct buffer *bufs[render_count];
shm_get_many(chain, render_count, widths, heights, bufs);
uint32_t fg = term->conf->colors.use_custom.jump_label
? term->conf->colors.jump_label.fg
: term->colors.table[0];
uint32_t bg = term->conf->colors.use_custom.jump_label
? term->conf->colors.jump_label.bg
: term->colors.table[3];
for (size_t i = 0; i < render_count; i++) {
const struct wayl_sub_surface *sub_surf = &info[i].url->surf;
const char32_t *label = info[i].text;
const int x = info[i].x;
const int y = info[i].y;
xassert(sub_surf->surface.surf != NULL);
xassert(sub_surf->sub != NULL);
wl_subsurface_set_position(
sub_surf->sub,
roundf((term->margins.left + x) / scale),
roundf((term->margins.top + y) / scale));
render_osd(
term, sub_surf, term->fonts[0], bufs[i], label,
fg, 0xffu << 24 | bg, x_margin);
free(info[i].text);
}
}
static void
render_update_title(struct terminal *term)
{
static const size_t max_len = 2048;
const char *title = term->window_title != NULL ? term->window_title : "foot";
char *copy = NULL;
if (strlen(title) > max_len) {
copy = xstrndup(title, max_len);
title = copy;
}
xdg_toplevel_set_title(term->window->xdg_toplevel, title);
free(copy);
}
static void
frame_callback(void *data, struct wl_callback *wl_callback, uint32_t callback_data)
{
struct terminal *term = data;
xassert(term->window->frame_callback == wl_callback);
wl_callback_destroy(wl_callback);
term->window->frame_callback = NULL;
bool grid = term->render.pending.grid;
bool csd = term->render.pending.csd;
bool search = term->is_searching && term->render.pending.search;
bool urls = urls_mode_is_active(term) > 0 && term->render.pending.urls;
term->render.pending.grid = false;
term->render.pending.csd = false;
term->render.pending.search = false;
term->render.pending.urls = false;
struct grid *original_grid = term->grid;
if (urls_mode_is_active(term)) {
xassert(term->url_grid_snapshot != NULL);
term->grid = term->url_grid_snapshot;
}
if (csd && term->window->csd_mode == CSD_YES) {
quirk_weston_csd_on(term);
render_csd(term);
quirk_weston_csd_off(term);
}
if (search)
render_search_box(term);
if (urls)
render_urls(term);
if ((grid && !term->delayed_render_timer.is_armed) || (csd | search | urls))
grid_render(term);
tll_foreach(term->wl->seats, it) {
if (it->item.ime_focus == term)
ime_update_cursor_rect(&it->item);
}
term->grid = original_grid;
}
static void
tiocswinsz(struct terminal *term)
{
if (term->ptmx >= 0) {
if (ioctl(term->ptmx, (unsigned int)TIOCSWINSZ,
&(struct winsize){
.ws_row = term->rows,
.ws_col = term->cols,
.ws_xpixel = term->cols * term->cell_width,
.ws_ypixel = term->rows * term->cell_height}) < 0)
{
LOG_ERRNO("TIOCSWINSZ");
}
}
}
static void
delayed_reflow_of_normal_grid(struct terminal *term)
{
if (term->interactive_resizing.grid == NULL)
return;
xassert(term->interactive_resizing.new_rows > 0);
struct coord *const tracking_points[] = {
&term->selection.coords.start,
&term->selection.coords.end,
};
/* Reflow the original (since before the resize was started) grid,
* to the *current* dimensions */
grid_resize_and_reflow(
term->interactive_resizing.grid,
term->interactive_resizing.new_rows, term->normal.num_cols,
term->interactive_resizing.old_screen_rows, term->rows,
term->selection.coords.end.row >= 0 ? ALEN(tracking_points) : 0,
tracking_points);
/* Replace the current, truncated, “normal” grid with the
* correctly reflowed one */
grid_free(&term->normal);
term->normal = *term->interactive_resizing.grid;
free(term->interactive_resizing.grid);
term->hide_cursor = term->interactive_resizing.old_hide_cursor;
/* Reset */
term->interactive_resizing.grid = NULL;
term->interactive_resizing.old_screen_rows = 0;
term->interactive_resizing.new_rows = 0;
term->interactive_resizing.old_hide_cursor = false;
/* Invalidate render pointers */
shm_unref(term->render.last_buf);
term->render.last_buf = NULL;
term->render.last_cursor.row = NULL;
tll_free(term->normal.scroll_damage);
sixel_reflow_grid(term, &term->normal);
if (term->grid == &term->normal) {
term_damage_view(term);
render_refresh(term);
}
term_ptmx_resume(term);
}
static bool
fdm_tiocswinsz(struct fdm *fdm, int fd, int events, void *data)
{
struct terminal *term = data;
if (events & EPOLLIN) {
tiocswinsz(term);
delayed_reflow_of_normal_grid(term);
}
if (term->window->resize_timeout_fd >= 0) {
fdm_del(fdm, term->window->resize_timeout_fd);
term->window->resize_timeout_fd = -1;
}
return true;
}
static void
send_dimensions_to_client(struct terminal *term)
{
struct wl_window *win = term->window;
if (!win->is_resizing || term->conf->resize_delay_ms == 0) {
/* Send new dimensions to client immediately */
tiocswinsz(term);
delayed_reflow_of_normal_grid(term);
/* And make sure to reset and deallocate a lingering timer */
if (win->resize_timeout_fd >= 0) {
fdm_del(term->fdm, win->resize_timeout_fd);
win->resize_timeout_fd = -1;
}
} else {
/* Send new dimensions to client “in a while” */
assert(win->is_resizing && term->conf->resize_delay_ms > 0);
int fd = win->resize_timeout_fd;
uint16_t delay_ms = term->conf->resize_delay_ms;
bool successfully_scheduled = false;
if (fd < 0) {
/* Lazy create timer fd */
fd = timerfd_create(CLOCK_MONOTONIC, TFD_CLOEXEC | TFD_NONBLOCK);
if (fd < 0)
LOG_ERRNO("failed to create TIOCSWINSZ timer");
else if (!fdm_add(term->fdm, fd, EPOLLIN, &fdm_tiocswinsz, term)) {
close(fd);
fd = -1;
}
win->resize_timeout_fd = fd;
}
if (fd >= 0) {
2021-01-17 16:38:24 +01:00
/* Reset timeout */
const struct itimerspec timeout = {
.it_value = {
.tv_sec = delay_ms / 1000,
.tv_nsec = (delay_ms % 1000) * 1000000,
},
};
if (timerfd_settime(fd, 0, &timeout, NULL) < 0) {
LOG_ERRNO("failed to arm TIOCSWINSZ timer");
fdm_del(term->fdm, fd);
win->resize_timeout_fd = -1;
} else
successfully_scheduled = true;
}
if (!successfully_scheduled) {
tiocswinsz(term);
delayed_reflow_of_normal_grid(term);
}
}
}
/* Move to terminal.c? */
static bool
maybe_resize(struct terminal *term, int width, int height, bool force)
{
if (term->shutdown.in_progress)
return false;
if (!term->window->is_configured)
return false;
if (term->cell_width == 0 && term->cell_height == 0)
return false;
terminal: break out scaling factor updating, and reduce number of calls to render_resize() Break out the logic that updates the terminal’s scaling factor value, from render_resize(), to a new function, term_update_scale(). This allows us to update the scaling factor without a full grid resize. We also change how we pick the scaling factor (when fractional scaling is not in use). Before, we’d use the highest scaling factor from all monitors we were mapped on. Now, we use the scaling factor from the monitor we were *last* mapped on. Then, add a boolean parameter to term_set_fonts(), and when false, *don’t* call render_resize_force(). Also change term_font_dpi_changed() to only return true if the font was changed in any way. Finally, rewrite update_term_for_output_change() to: * Call term_update_scale() before doing anything else * Call render_resize{,_force} *last*, and *only* if either the scale or the fonts were updated. This fixes several things: * A bug where we failed to update the fonts when fractional scaling was in use, and we guessed the initial scale/DPI wrong. The bug happened because updated the internal "preferred" scale value, and a later call to render_resize() updated the terminal’s scale value, but since that code path didn’t call term_font_dpi_changed() (and it shouldn’t), the fonts weren’t resized properly. * It ensures we only resize the grid *once* when the scaling factor, or DPI is changed. Before this, we’d resize it twice. And this happened when e.g. dragging the window between monitors.
2023-07-17 16:21:16 +02:00
const float scale = term->scale;
width = round(width * scale);
height = round(height * scale);
if (width == 0 && height == 0) {
/*
* The compositor is letting us choose the size
*
* If we have a "last" used size - use that. Otherwise, use
* the size from the user configuration.
*/
if (term->stashed_width != 0 && term->stashed_height != 0) {
width = term->stashed_width;
height = term->stashed_height;
} else {
switch (term->conf->size.type) {
case CONF_SIZE_PX:
width = term->conf->size.width;
height = term->conf->size.height;
/* Take CSDs into account */
if (wayl_win_csd_titlebar_visible(term->window))
height -= term->conf->csd.title_height;
if (wayl_win_csd_borders_visible(term->window)) {
height -= 2 * term->conf->csd.border_width_visible;
width -= 2 * term->conf->csd.border_width_visible;
}
width *= scale;
height *= scale;
break;
case CONF_SIZE_CELLS:
width = term->conf->size.width * term->cell_width;
height = term->conf->size.height * term->cell_height;
width += 2 * term->conf->pad_x * scale;
height += 2 * term->conf->pad_y * scale;
/* Ensure width/height is a valid multiple of scale */
width = roundf(scale * roundf(width / scale));
height = roundf(scale * roundf(height / scale));
break;
}
}
}
/* Don't shrink grid too much */
const int min_cols = 2;
const int min_rows = 1;
/* Minimum window size (must be divisible by the scaling factor)*/
const int min_width = roundf(scale * ceilf((min_cols * term->cell_width) / scale));
const int min_height = roundf(scale * ceilf((min_rows * term->cell_height) / scale));
width = max(width, min_width);
height = max(height, min_height);
/* Padding */
const int max_pad_x = (width - min_width) / 2;
const int max_pad_y = (height - min_height) / 2;
const int pad_x = min(max_pad_x, scale * term->conf->pad_x);
const int pad_y = min(max_pad_y, scale * term->conf->pad_y);
if (!force && width == term->width && height == term->height && scale == term->scale)
return false;
/* Cancel an application initiated "Synchronized Update" */
term_disable_app_sync_updates(term);
/* Drop out of URL mode */
urls_reset(term);
terminal: break out scaling factor updating, and reduce number of calls to render_resize() Break out the logic that updates the terminal’s scaling factor value, from render_resize(), to a new function, term_update_scale(). This allows us to update the scaling factor without a full grid resize. We also change how we pick the scaling factor (when fractional scaling is not in use). Before, we’d use the highest scaling factor from all monitors we were mapped on. Now, we use the scaling factor from the monitor we were *last* mapped on. Then, add a boolean parameter to term_set_fonts(), and when false, *don’t* call render_resize_force(). Also change term_font_dpi_changed() to only return true if the font was changed in any way. Finally, rewrite update_term_for_output_change() to: * Call term_update_scale() before doing anything else * Call render_resize{,_force} *last*, and *only* if either the scale or the fonts were updated. This fixes several things: * A bug where we failed to update the fonts when fractional scaling was in use, and we guessed the initial scale/DPI wrong. The bug happened because updated the internal "preferred" scale value, and a later call to render_resize() updated the terminal’s scale value, but since that code path didn’t call term_font_dpi_changed() (and it shouldn’t), the fonts weren’t resized properly. * It ensures we only resize the grid *once* when the scaling factor, or DPI is changed. Before this, we’d resize it twice. And this happened when e.g. dragging the window between monitors.
2023-07-17 16:21:16 +02:00
LOG_DBG("resized: size=%dx%d (scale=%.2f)", width, height, term->scale);
term->width = width;
term->height = height;
const uint32_t scrollback_lines = term->render.scrollback_lines;
/* Screen rows/cols before resize */
int old_cols = term->cols;
int old_rows = term->rows;
/* Screen rows/cols after resize */
const int new_cols = (term->width - 2 * pad_x) / term->cell_width;
const int new_rows = (term->height - 2 * pad_y) / term->cell_height;
/* Grid rows/cols after resize */
const int new_normal_grid_rows = 1 << (32 - __builtin_clz(new_rows + scrollback_lines - 1));
const int new_alt_grid_rows = 1 << (32 - __builtin_clz(new_rows));
xassert(new_cols >= 1);
xassert(new_rows >= 1);
/* Margins */
const int grid_width = new_cols * term->cell_width;
const int grid_height = new_rows * term->cell_height;
const int total_x_pad = term->width - grid_width;
const int total_y_pad = term->height - grid_height;
2021-01-18 09:52:11 +01:00
if (term->conf->center && !term->window->is_resizing) {
term->margins.left = total_x_pad / 2;
term->margins.top = total_y_pad / 2;
} else {
term->margins.left = pad_x;
term->margins.top = pad_y;
}
term->margins.right = total_x_pad - term->margins.left;
term->margins.bottom = total_y_pad - term->margins.top;
xassert(term->margins.left >= pad_x);
xassert(term->margins.right >= pad_x);
xassert(term->margins.top >= pad_y);
xassert(term->margins.bottom >= pad_y);
if (new_cols == old_cols && new_rows == old_rows) {
LOG_DBG("grid layout unaffected; skipping reflow");
term->interactive_resizing.new_rows = new_normal_grid_rows;
goto damage_view;
render: initial support for text reflow The algorithm is as follows: Start at the beginning of the scrollback. That is, at the oldest emitted lines. This is done by taking the current offset, and adding the number of (old) screen rows, and then iterating until we find the first allocated line. Next, we iterate the entire old grid. At the beginning, we allocate a line for the new grid, and setup a global pointer for that line, and the current cell index. For each line in the old grid, iterate its cells. Copy the the cells over to the new line. Whenever the new line reaches its maximum number of columns, we line break it by increasing the current row index and allocating a new row (if necessary - we may be overwriting old scrollback if the new grid is smaller than the old grid). Whenever we reach the end of a line of the old grid, we insert a line break in the new grid's line too **if** the last cell in the old line was empty. If it was **not** empty, we **don't** line break the new line. Furthermore, empty cells in general need special consideration. A line ending with a string of empty cells doesn't have to be copied the new line. And more importantly, should **not** increase the new line's cell index (which may cause line breaks, which is incorrect). However, if a string of empty cells is followed by non empty cells, we need to copy all the preceding empty cells to the line too. When the entire scrollback history has been reflowed, we need to figure out the new grid's offset. This is done by trying to put the **last** emitted line at the bottom of the screen. I.e. the new offset is typically "last_line_idx - term->rows". However, we need to handle empty lines. So, after subtracting the number of screen rows, we _increase_ the offset until we see a non-empty line. This ensures we handle grid's that doesn't fill an entire screen. Finally, we need to re-position the cursor. This is done by trying to place the cursor **at** (_not_ after) the last emitted line. We keep the current cursor column as is (but possibly truncated, if the grid's width decreased).
2020-02-10 20:35:24 +01:00
}
/*
* Since text reflow is slow, dont do it *while* resizing. Only
* do it when done, or after pausing the resize for sufficiently
2023-10-03 14:11:55 +02:00
* long. We reuse the TIOCSWINSZ timer to handle this. See
* send_dimensions_to_client() and fdm_tiocswinsz().
*
* To be able to do the final reflow correctly, we need a copy of
* the original grid, before the resize started.
*/
if (term->window->is_resizing && term->conf->resize_delay_ms > 0) {
if (term->interactive_resizing.grid == NULL) {
term_ptmx_pause(term);
/* Stash the current normal grid, as-is, to be used when
* doing the final reflow */
term->interactive_resizing.old_screen_rows = term->rows;
term->interactive_resizing.old_cols = term->cols;
term->interactive_resizing.old_hide_cursor = term->hide_cursor;
term->interactive_resizing.grid = xmalloc(sizeof(*term->interactive_resizing.grid));
*term->interactive_resizing.grid = term->normal;
if (term->grid == &term->normal)
term->interactive_resizing.selection_coords = term->selection.coords;
} else {
/* Well replace the current temporary grid, with a new
* one (again based on the original grid) */
grid_free(&term->normal);
}
struct grid *orig = term->interactive_resizing.grid;
/*
* Copy the current viewport (of the original grid) to a new
* grid that will be used during the resize. For now, throw
* away sixels and OSC-8 URLs. Theyll be "restored" when we
* do the final reflow.
*
* Note that OSC-8 URLs are perfectly ok to throw away; they
* cannot be interacted with during the resize. And, even if
* url.osc8-underline=always, the underline attribute is
* part of the cell, not the URI struct (and thus our faked
* grid will still render OSC-8 links underlined).
*
* TODO:
* - sixels?
*/
struct grid g = {
.num_rows = 1 << (32 - __builtin_clz(term->interactive_resizing.old_screen_rows)),
.num_cols = term->interactive_resizing.old_cols,
.offset = 0,
.view = 0,
.cursor = orig->cursor,
.saved_cursor = orig->saved_cursor,
.rows = xcalloc(g.num_rows, sizeof(g.rows[0])),
.cur_row = NULL,
.scroll_damage = tll_init(),
.sixel_images = tll_init(),
.kitty_kbd = orig->kitty_kbd,
};
term->selection.coords.start.row -= orig->view;
term->selection.coords.end.row -= orig->view;
for (size_t i = 0, j = orig->view;
i < term->interactive_resizing.old_screen_rows;
i++, j = (j + 1) & (orig->num_rows - 1))
{
g.rows[i] = grid_row_alloc(g.num_cols, false);
memcpy(g.rows[i]->cells,
orig->rows[j]->cells,
g.num_cols * sizeof(g.rows[i]->cells[0]));
}
term->normal = g;
term->hide_cursor = true;
}
if (term->grid == &term->alt)
selection_cancel(term);
else {
/*
* Dont cancel, but make sure there arent any ongoing
* selections after the resize.
*/
tll_foreach(term->wl->seats, it) {
if (it->item.kbd_focus == term)
selection_finalize(&it->item, term, it->item.pointer.serial);
}
}
/*
* TODO: if we remove the selection_finalize() call above (i.e. if
* we start allowing selections to be ongoing across resizes), the
* selections pivot point coordinates *must* be added to the
* tracking points list.
*/
/* Resize grids */
if (term->window->is_resizing && term->conf->resize_delay_ms > 0) {
/* Simple truncating resize, *while* an interactive resize is
* ongoing. */
xassert(term->interactive_resizing.grid != NULL);
xassert(new_normal_grid_rows > 0);
term->interactive_resizing.new_rows = new_normal_grid_rows;
grid_resize_without_reflow(
&term->normal, new_alt_grid_rows, new_cols,
term->interactive_resizing.old_screen_rows, new_rows);
} else {
/* Full text reflow */
int old_normal_rows = old_rows;
if (term->interactive_resizing.grid != NULL) {
/* Throw away the current, truncated, “normal” grid, and
* use the original grid instead (from before the resize
* started) */
grid_free(&term->normal);
term->normal = *term->interactive_resizing.grid;
free(term->interactive_resizing.grid);
term->hide_cursor = term->interactive_resizing.old_hide_cursor;
term->selection.coords = term->interactive_resizing.selection_coords;
old_normal_rows = term->interactive_resizing.old_screen_rows;
term->interactive_resizing.grid = NULL;
term->interactive_resizing.old_screen_rows = 0;
term->interactive_resizing.new_rows = 0;
term->interactive_resizing.old_hide_cursor = false;
term->interactive_resizing.selection_coords = (struct range){{-1, -1}, {-1, -1}};
term_ptmx_resume(term);
}
struct coord *const tracking_points[] = {
&term->selection.coords.start,
&term->selection.coords.end,
};
grid_resize_and_reflow(
&term->normal, new_normal_grid_rows, new_cols, old_normal_rows, new_rows,
term->selection.coords.end.row >= 0 ? ALEN(tracking_points) : 0,
tracking_points);
}
grid_resize_without_reflow(
&term->alt, new_alt_grid_rows, new_cols, old_rows, new_rows);
/* Reset tab stops */
tll_free(term->tab_stops);
for (int c = 0; c < new_cols; c += 8)
tll_push_back(term->tab_stops, c);
term->cols = new_cols;
term->rows = new_rows;
sixel_reflow(term);
terminal: break out scaling factor updating, and reduce number of calls to render_resize() Break out the logic that updates the terminal’s scaling factor value, from render_resize(), to a new function, term_update_scale(). This allows us to update the scaling factor without a full grid resize. We also change how we pick the scaling factor (when fractional scaling is not in use). Before, we’d use the highest scaling factor from all monitors we were mapped on. Now, we use the scaling factor from the monitor we were *last* mapped on. Then, add a boolean parameter to term_set_fonts(), and when false, *don’t* call render_resize_force(). Also change term_font_dpi_changed() to only return true if the font was changed in any way. Finally, rewrite update_term_for_output_change() to: * Call term_update_scale() before doing anything else * Call render_resize{,_force} *last*, and *only* if either the scale or the fonts were updated. This fixes several things: * A bug where we failed to update the fonts when fractional scaling was in use, and we guessed the initial scale/DPI wrong. The bug happened because updated the internal "preferred" scale value, and a later call to render_resize() updated the terminal’s scale value, but since that code path didn’t call term_font_dpi_changed() (and it shouldn’t), the fonts weren’t resized properly. * It ensures we only resize the grid *once* when the scaling factor, or DPI is changed. Before this, we’d resize it twice. And this happened when e.g. dragging the window between monitors.
2023-07-17 16:21:16 +02:00
LOG_DBG("resized: grid: cols=%d, rows=%d "
"(left-margin=%d, right-margin=%d, top-margin=%d, bottom-margin=%d)",
terminal: break out scaling factor updating, and reduce number of calls to render_resize() Break out the logic that updates the terminal’s scaling factor value, from render_resize(), to a new function, term_update_scale(). This allows us to update the scaling factor without a full grid resize. We also change how we pick the scaling factor (when fractional scaling is not in use). Before, we’d use the highest scaling factor from all monitors we were mapped on. Now, we use the scaling factor from the monitor we were *last* mapped on. Then, add a boolean parameter to term_set_fonts(), and when false, *don’t* call render_resize_force(). Also change term_font_dpi_changed() to only return true if the font was changed in any way. Finally, rewrite update_term_for_output_change() to: * Call term_update_scale() before doing anything else * Call render_resize{,_force} *last*, and *only* if either the scale or the fonts were updated. This fixes several things: * A bug where we failed to update the fonts when fractional scaling was in use, and we guessed the initial scale/DPI wrong. The bug happened because updated the internal "preferred" scale value, and a later call to render_resize() updated the terminal’s scale value, but since that code path didn’t call term_font_dpi_changed() (and it shouldn’t), the fonts weren’t resized properly. * It ensures we only resize the grid *once* when the scaling factor, or DPI is changed. Before this, we’d resize it twice. And this happened when e.g. dragging the window between monitors.
2023-07-17 16:21:16 +02:00
term->cols, term->rows,
term->margins.left, term->margins.right,
term->margins.top, term->margins.bottom);
if (term->scroll_region.start >= term->rows)
term->scroll_region.start = 0;
if (term->scroll_region.end > term->rows ||
term->scroll_region.end >= old_rows)
{
term->scroll_region.end = term->rows;
}
term->render.last_cursor.row = NULL;
damage_view:
/* Signal TIOCSWINSZ */
send_dimensions_to_client(term);
if (!term->window->is_maximized &&
!term->window->is_fullscreen &&
!term->window->is_tiled)
{
/* Stash current size, to enable us to restore it when we're
* being un-maximized/fullscreened/tiled */
term->stashed_width = term->width;
term->stashed_height = term->height;
}
{
const bool title_shown = wayl_win_csd_titlebar_visible(term->window);
const bool border_shown = wayl_win_csd_borders_visible(term->window);
const int title = title_shown
? roundf(term->conf->csd.title_height * scale)
: 0;
const int border = border_shown
? roundf(term->conf->csd.border_width_visible * scale)
: 0;
/* Must use surface logical coordinates (same calculations as
in get_csd_data(), but with different inputs) */
const int toplevel_min_width = roundf(border / scale) +
roundf(min_width / scale) +
roundf(border / scale);
const int toplevel_min_height = roundf(border / scale) +
roundf(title / scale) +
roundf(min_height / scale) +
roundf(border / scale);
const int toplevel_width = roundf(border / scale) +
roundf(term->width / scale) +
roundf(border / scale);
const int toplevel_height = roundf(border / scale) +
roundf(title / scale) +
roundf(term->height / scale) +
roundf(border / scale);
const int x = roundf(-border / scale);
const int y = roundf(-title / scale) - roundf(border / scale);
xdg_toplevel_set_min_size(
term->window->xdg_toplevel,
toplevel_min_width, toplevel_min_height);
xdg_surface_set_window_geometry(
term->window->xdg_surface,
x, y, toplevel_width, toplevel_height);
2020-03-03 18:29:46 +01:00
}
tll_free(term->normal.scroll_damage);
tll_free(term->alt.scroll_damage);
shm_unref(term->render.last_buf);
term->render.last_buf = NULL;
term_damage_view(term);
render_refresh_csd(term);
render_refresh_search(term);
render_refresh(term);
return true;
}
bool
render_resize(struct terminal *term, int width, int height)
{
return maybe_resize(term, width, height, false);
}
bool
render_resize_force(struct terminal *term, int width, int height)
{
return maybe_resize(term, width, height, true);
}
static void xcursor_callback(
void *data, struct wl_callback *wl_callback, uint32_t callback_data);
static const struct wl_callback_listener xcursor_listener = {
.done = &xcursor_callback,
};
bool
render_xcursor_is_valid(const struct seat *seat, const char *cursor)
{
if (cursor == NULL)
return false;
return wl_cursor_theme_get_cursor(seat->pointer.theme, cursor) != NULL;
}
static void
render_xcursor_update(struct seat *seat)
{
/* If called from a frame callback, we may no longer have mouse focus */
if (!seat->mouse_focus)
return;
xassert(seat->pointer.shape != CURSOR_SHAPE_NONE);
if (seat->pointer.shape == CURSOR_SHAPE_HIDDEN) {
/* Hide cursor */
LOG_DBG("hiding cursor using client-side NULL-surface");
wl_surface_attach(seat->pointer.surface.surf, NULL, 0, 0);
wl_pointer_set_cursor(
seat->wl_pointer, seat->pointer.serial, seat->pointer.surface.surf,
0, 0);
wl_surface_commit(seat->pointer.surface.surf);
return;
}
xassert(seat->pointer.cursor != NULL);
const enum cursor_shape shape = seat->pointer.shape;
const char *const xcursor = seat->pointer.last_custom_xcursor;
if (seat->pointer.shape_device != NULL) {
xassert(shape != CURSOR_SHAPE_CUSTOM || xcursor != NULL);
const enum wp_cursor_shape_device_v1_shape custom_shape =
(shape == CURSOR_SHAPE_CUSTOM && xcursor != NULL
? cursor_string_to_server_shape(xcursor)
: 0);
if (shape != CURSOR_SHAPE_CUSTOM || custom_shape != 0) {
xassert(custom_shape == 0 || shape == CURSOR_SHAPE_CUSTOM);
const enum wp_cursor_shape_device_v1_shape wp_shape = custom_shape != 0
? custom_shape
: cursor_shape_to_server_shape(shape);
LOG_DBG("setting %scursor shape using cursor-shape-v1",
custom_shape != 0 ? "custom " : "");
wp_cursor_shape_device_v1_set_shape(
seat->pointer.shape_device,
seat->pointer.serial,
wp_shape);
return;
}
}
LOG_DBG("setting %scursor shape using a client-side cursor surface",
terminal: break out scaling factor updating, and reduce number of calls to render_resize() Break out the logic that updates the terminal’s scaling factor value, from render_resize(), to a new function, term_update_scale(). This allows us to update the scaling factor without a full grid resize. We also change how we pick the scaling factor (when fractional scaling is not in use). Before, we’d use the highest scaling factor from all monitors we were mapped on. Now, we use the scaling factor from the monitor we were *last* mapped on. Then, add a boolean parameter to term_set_fonts(), and when false, *don’t* call render_resize_force(). Also change term_font_dpi_changed() to only return true if the font was changed in any way. Finally, rewrite update_term_for_output_change() to: * Call term_update_scale() before doing anything else * Call render_resize{,_force} *last*, and *only* if either the scale or the fonts were updated. This fixes several things: * A bug where we failed to update the fonts when fractional scaling was in use, and we guessed the initial scale/DPI wrong. The bug happened because updated the internal "preferred" scale value, and a later call to render_resize() updated the terminal’s scale value, but since that code path didn’t call term_font_dpi_changed() (and it shouldn’t), the fonts weren’t resized properly. * It ensures we only resize the grid *once* when the scaling factor, or DPI is changed. Before this, we’d resize it twice. And this happened when e.g. dragging the window between monitors.
2023-07-17 16:21:16 +02:00
seat->pointer.shape == CURSOR_SHAPE_CUSTOM ? "custom " : "");
const float scale = seat->pointer.scale;
struct wl_cursor_image *image = seat->pointer.cursor->images[0];
struct wl_buffer *buf = wl_cursor_image_get_buffer(image);
wayl_surface_scale_explicit_width_height(
seat->mouse_focus->window,
&seat->pointer.surface, image->width, image->height, scale);
wl_surface_attach(seat->pointer.surface.surf, buf, 0, 0);
wl_pointer_set_cursor(
seat->wl_pointer, seat->pointer.serial,
seat->pointer.surface.surf,
image->hotspot_x / scale, image->hotspot_y / scale);
wl_surface_damage_buffer(
seat->pointer.surface.surf, 0, 0, INT32_MAX, INT32_MAX);
xassert(seat->pointer.xcursor_callback == NULL);
seat->pointer.xcursor_callback = wl_surface_frame(seat->pointer.surface.surf);
wl_callback_add_listener(seat->pointer.xcursor_callback, &xcursor_listener, seat);
wl_surface_commit(seat->pointer.surface.surf);
}
static void
xcursor_callback(void *data, struct wl_callback *wl_callback, uint32_t callback_data)
{
struct seat *seat = data;
xassert(seat->pointer.xcursor_callback == wl_callback);
wl_callback_destroy(wl_callback);
seat->pointer.xcursor_callback = NULL;
if (seat->pointer.xcursor_pending) {
render_xcursor_update(seat);
seat->pointer.xcursor_pending = false;
}
}
static void
fdm_hook_refresh_pending_terminals(struct fdm *fdm, void *data)
{
struct renderer *renderer = data;
struct wayland *wayl = renderer->wayl;
tll_foreach(renderer->wayl->terms, it) {
struct terminal *term = it->item;
if (unlikely(term->shutdown.in_progress || !term->window->is_configured))
continue;
bool grid = term->render.refresh.grid;
bool csd = term->render.refresh.csd;
bool search = term->is_searching && term->render.refresh.search;
bool urls = urls_mode_is_active(term) && term->render.refresh.urls;
if (!(grid | csd | search | urls))
2020-12-06 12:18:46 +01:00
continue;
if (term->render.app_sync_updates.enabled && !(csd | search | urls))
2020-12-06 12:18:46 +01:00
continue;
term->render.refresh.grid = false;
term->render.refresh.csd = false;
term->render.refresh.search = false;
term->render.refresh.urls = false;
if (term->window->frame_callback == NULL) {
struct grid *original_grid = term->grid;
if (urls_mode_is_active(term)) {
xassert(term->url_grid_snapshot != NULL);
term->grid = term->url_grid_snapshot;
}
if (csd && term->window->csd_mode == CSD_YES) {
quirk_weston_csd_on(term);
render_csd(term);
quirk_weston_csd_off(term);
}
if (search)
render_search_box(term);
if (urls)
render_urls(term);
if (grid | csd | search | urls)
grid_render(term);
tll_foreach(term->wl->seats, it) {
if (it->item.ime_focus == term)
ime_update_cursor_rect(&it->item);
}
term->grid = original_grid;
} else {
/* Tells the frame callback to render again */
term->render.pending.grid |= grid;
term->render.pending.csd |= csd;
term->render.pending.search |= search;
term->render.pending.urls |= urls;
}
}
tll_foreach(wayl->seats, it) {
if (it->item.pointer.xcursor_pending) {
if (it->item.pointer.xcursor_callback == NULL) {
render_xcursor_update(&it->item);
it->item.pointer.xcursor_pending = false;
} else {
/* Frame callback will call render_xcursor_update() */
}
}
}
}
void
render_refresh_title(struct terminal *term)
{
if (term->render.title.is_armed)
return;
struct timespec now;
if (clock_gettime(CLOCK_MONOTONIC, &now) < 0)
return;
struct timespec diff;
timespec_sub(&now, &term->render.title.last_update, &diff);
if (diff.tv_sec == 0 && diff.tv_nsec < 8333 * 1000) {
const struct itimerspec timeout = {
.it_value = {.tv_nsec = 8333 * 1000 - diff.tv_nsec},
};
timerfd_settime(term->render.title.timer_fd, 0, &timeout, NULL);
} else {
term->render.title.last_update = now;
render_update_title(term);
}
render_refresh_csd(term);
}
2019-07-24 20:09:49 +02:00
void
render_refresh(struct terminal *term)
{
term->render.refresh.grid = true;
}
void
render_refresh_csd(struct terminal *term)
{
if (term->window->csd_mode == CSD_YES)
term->render.refresh.csd = true;
}
void
render_refresh_search(struct terminal *term)
{
if (term->is_searching)
term->render.refresh.search = true;
2019-07-24 20:09:49 +02:00
}
void
render_refresh_urls(struct terminal *term)
{
if (urls_mode_is_active(term))
term->render.refresh.urls = true;
}
bool
render_xcursor_set(struct seat *seat, struct terminal *term,
enum cursor_shape shape)
{
if (seat->pointer.theme == NULL)
return false;
if (seat->mouse_focus == NULL) {
seat->pointer.shape = CURSOR_SHAPE_NONE;
return true;
}
if (seat->mouse_focus != term) {
/* This terminal doesn't have mouse focus */
return true;
}
if (seat->pointer.shape == shape &&
!(shape == CURSOR_SHAPE_CUSTOM &&
!streq(seat->pointer.last_custom_xcursor,
term->mouse_user_cursor)))
{
return true;
}
/* TODO: skip this when using server-side cursors */
if (shape != CURSOR_SHAPE_HIDDEN) {
const char *const xcursor = shape == CURSOR_SHAPE_CUSTOM
? term->mouse_user_cursor
: cursor_shape_to_string(shape);
const char *const fallback =
cursor_shape_to_string(CURSOR_SHAPE_TEXT_FALLBACK);
seat->pointer.cursor = wl_cursor_theme_get_cursor(
seat->pointer.theme, xcursor);
if (seat->pointer.cursor == NULL) {
seat->pointer.cursor = wl_cursor_theme_get_cursor(
seat->pointer.theme, fallback);
if (seat->pointer.cursor == NULL) {
LOG_ERR("failed to load xcursor pointer "
"'%s', and fallback '%s'", xcursor, fallback);
return false;
}
}
if (shape == CURSOR_SHAPE_CUSTOM) {
free(seat->pointer.last_custom_xcursor);
seat->pointer.last_custom_xcursor = xstrdup(term->mouse_user_cursor);
}
} else {
seat->pointer.cursor = NULL;
free(seat->pointer.last_custom_xcursor);
seat->pointer.last_custom_xcursor = NULL;
}
/* FDM hook takes care of actual rendering */
seat->pointer.shape = shape;
seat->pointer.xcursor_pending = true;
return true;
}