labwc/src/edges.c

642 lines
18 KiB
C

// SPDX-License-Identifier: GPL-2.0-only
#include <assert.h>
#include <limits.h>
#include <pixman.h>
#include <wlr/util/edges.h>
#include <wlr/util/box.h>
#include "common/border.h"
#include "common/box.h"
#include "common/macros.h"
#include "config/rcxml.h"
#include "edges.h"
#include "labwc.h"
#include "view.h"
#include "node.h"
static void
edges_for_target_geometry(struct border *edges, struct view *view,
struct wlr_box target)
{
struct border border = ssd_get_margin(view->ssd);
/* Use the effective height to properly handle shaded views */
int eff_height = view->shaded ? 0 : target.height;
edges->left = target.x - border.left - rc.gap;
edges->top = target.y - border.top - rc.gap;
edges->right = target.x + target.width + border.right + rc.gap;
edges->bottom = target.y + eff_height + border.bottom + rc.gap;
}
void
edges_initialize(struct border *edges)
{
assert(edges);
edges->top = INT_MIN;
edges->right = INT_MAX;
edges->bottom = INT_MAX;
edges->left = INT_MIN;
}
static inline struct edge
build_edge(struct border region, enum wlr_edges direction, int pad)
{
struct edge edge = { 0 };
switch (direction) {
case WLR_EDGE_LEFT:
edge.offset = clipped_sub(region.left, pad);
edge.min = region.top;
edge.max = region.bottom;
break;
case WLR_EDGE_RIGHT:
edge.offset = clipped_add(region.right, pad);
edge.min = region.top;
edge.max = region.bottom;
break;
case WLR_EDGE_TOP:
edge.offset = clipped_sub(region.top, pad);
edge.min = region.left;
edge.max = region.right;
break;
case WLR_EDGE_BOTTOM:
edge.offset = clipped_add(region.bottom, pad);
edge.min = region.left;
edge.max = region.right;
break;
case WLR_EDGE_NONE:
/* Should never be reached */
wlr_log(WLR_ERROR, "invalid direction");
abort();
}
return edge;
}
static inline bool
is_lesser(enum wlr_edges direction)
{
return direction == WLR_EDGE_LEFT || direction == WLR_EDGE_TOP;
}
static inline struct edge
build_visible_edge(struct border region, enum wlr_edges direction,
int pad, uint32_t edges_visible)
{
struct edge edge = build_edge(region, direction, pad);
if (!(edges_visible & direction)) {
edge.offset = is_lesser(direction) ? INT_MIN : INT_MAX;
}
return edge;
}
static void
validate_single_region_edge(int *valid_edge,
struct border view, struct border target,
struct border region, edge_validator_t validator,
enum wlr_edges direction, uint32_t edges_visible)
{
/*
* When a view snaps to another while moving to its target, it can do
* so in two ways: a view edge can snap to an "opposing" edge of the
* region (left <-> right, top <-> bottom) or to an "aligned" edge
* (left <-> left, right <-> right, top <-> top, bottom <-> bottom).
*
* When a view hits the opposing edge of a region, it should be
* separated by any configured gap and will resist *entry* into the
* region; when a view hits the aligned edge, it should not be
* separated by a gap and will resist *departure* from the region. The
* view and its target already include necessary padding to reflect the
* gap. The region does not. To make sure the "aligned" edges are
* properly aligned with respect to the configured gap, add padding to
* the region borders for aligned edges only.
*/
enum wlr_edges opposing = WLR_EDGE_NONE;
switch (direction) {
case WLR_EDGE_TOP:
opposing = WLR_EDGE_BOTTOM;
break;
case WLR_EDGE_BOTTOM:
opposing = WLR_EDGE_TOP;
break;
case WLR_EDGE_LEFT:
opposing = WLR_EDGE_RIGHT;
break;
case WLR_EDGE_RIGHT:
opposing = WLR_EDGE_LEFT;
break;
case WLR_EDGE_NONE:
/* Should never be reached */
assert(false);
return;
}
validator(valid_edge,
build_edge(view, direction, 0),
build_edge(target, direction, 0),
build_visible_edge(region, opposing, 0, edges_visible),
build_visible_edge(region, direction, rc.gap, edges_visible),
is_lesser(direction));
}
static void
validate_edges(struct border *valid_edges,
struct border view, struct border target,
struct border region, uint32_t edges_visible,
edge_validator_t validator)
{
/* Check for edges encountered during movement of left edge */
validate_single_region_edge(&valid_edges->left,
view, target, region, validator, WLR_EDGE_LEFT, edges_visible);
/* Check for edges encountered during movement of right edge */
validate_single_region_edge(&valid_edges->right,
view, target, region, validator, WLR_EDGE_RIGHT, edges_visible);
/* Check for edges encountered during movement of top edge */
validate_single_region_edge(&valid_edges->top,
view, target, region, validator, WLR_EDGE_TOP, edges_visible);
/* Check for edges encountered during movement of bottom edge */
validate_single_region_edge(&valid_edges->bottom,
view, target, region, validator, WLR_EDGE_BOTTOM, edges_visible);
}
static void
validate_single_output_edge(int *valid_edge,
struct border view, struct border target,
struct border region, edge_validator_t validator,
enum wlr_edges direction)
{
static struct border unbounded = {
.top = INT_MIN,
.right = INT_MAX,
.bottom = INT_MAX,
.left = INT_MIN,
};
validator(valid_edge,
build_edge(view, direction, 0),
build_edge(target, direction, 0),
build_edge(region, direction, 0),
build_edge(unbounded, direction, 0), is_lesser(direction));
}
static void
validate_output_edges(struct border *valid_edges,
struct border view, struct border target,
struct wlr_box usable, edge_validator_t validator)
{
/*
* When a view snaps to an output that contains it, it can be
* transformed into either of two equivalent problems:
*
* 1. The output region can be treated as if it were bounded by four
* half-planes, one sharing each edge of the view and extending
* infinitely *away* from the output. The moving view should then be
* tested as it encounters the "opposing" edge of each external region.
*
* 2. The output region can be treated as if it were composed of four
* half-planes, one sharing each edge of the view and extending
* infinitely to *overlap* the output. The moving view should then be
* tested as it encounters the "aligned" edge of each overlapping
* region.
*
* Either one of these problems can be realized by four calls to
* validate_edges with suitably defined half-plane regions, but most of
* the work in those validations will just be comparing invalid
* infinite edges.
*
* To save a bit of effort, just choose Problem 1 and directly validate
* only the non-infinite edges.
*/
struct border output = {
.top = usable.y,
.right = usable.x + usable.width,
.bottom = usable.y + usable.height,
.left = usable.x,
};
/* Left edge encounters a half-infinite region to the left of the output */
validate_single_output_edge(&valid_edges->left,
view, target, output, validator, WLR_EDGE_LEFT);
/* Right edge encounters a half-infinite region to the right of the output */
validate_single_output_edge(&valid_edges->right,
view, target, output, validator, WLR_EDGE_RIGHT);
/* Top edge encounters a half-infinite region above the output */
validate_single_output_edge(&valid_edges->top,
view, target, output, validator, WLR_EDGE_TOP);
/* Bottom edge encounters a half-infinite region below the output */
validate_single_output_edge(&valid_edges->bottom,
view, target, output, validator, WLR_EDGE_BOTTOM);
}
/* Test if parts of the current view is covered by the remaining space in the region */
static void
subtract_view_from_space(struct view *view, pixman_region32_t *available)
{
struct wlr_box view_size = ssd_max_extents(view);
pixman_box32_t view_rect = {
.x1 = view_size.x,
.x2 = view_size.x + view_size.width,
.y1 = view_size.y,
.y2 = view_size.y + view_size.height
};
pixman_region_overlap_t overlap =
pixman_region32_contains_rectangle(available, &view_rect);
switch (overlap) {
case PIXMAN_REGION_IN:
view->edges_visible = WLR_EDGE_TOP | WLR_EDGE_RIGHT
| WLR_EDGE_BOTTOM | WLR_EDGE_LEFT;
break;
case PIXMAN_REGION_OUT:
view->edges_visible = 0;
return;
case PIXMAN_REGION_PART:
; /* works around "a label can only be part of a statement" */
pixman_region32_t intersection;
pixman_region32_init(&intersection);
pixman_region32_intersect_rect(&intersection, available,
view_size.x, view_size.y,
view_size.width, view_size.height);
int nrects;
const pixman_box32_t *rects =
pixman_region32_rectangles(&intersection, &nrects);
view->edges_visible = 0;
for (int i = 0; i < nrects; i++) {
if (rects[i].x1 == view_rect.x1) {
view->edges_visible |= WLR_EDGE_LEFT;
}
if (rects[i].y1 == view_rect.y1) {
view->edges_visible |= WLR_EDGE_TOP;
}
if (rects[i].x2 == view_rect.x2) {
view->edges_visible |= WLR_EDGE_RIGHT;
}
if (rects[i].y2 == view_rect.y2) {
view->edges_visible |= WLR_EDGE_BOTTOM;
}
}
pixman_region32_fini(&intersection);
break;
}
/* Subtract the view geometry from the available region for the next check */
pixman_region32_t view_region;
pixman_region32_init_rects(&view_region, &view_rect, 1);
pixman_region32_subtract(available, available, &view_region);
pixman_region32_fini(&view_region);
}
static void
subtract_node_tree(struct wlr_scene_tree *tree, pixman_region32_t *available,
struct view *ignored_view)
{
struct view *view;
struct wlr_scene_node *node;
struct node_descriptor *node_desc;
wl_list_for_each_reverse(node, &tree->children, link) {
if (!node->enabled) {
/*
* This skips everything that is not being
* rendered, including minimized / unmapped
* windows and workspaces other than the
* current one.
*/
continue;
}
node_desc = node->data;
if (node_desc && node_desc->type == LAB_NODE_DESC_VIEW) {
view = node_view_from_node(node);
if (view != ignored_view) {
subtract_view_from_space(view, available);
}
} else if (node->type == WLR_SCENE_NODE_TREE) {
subtract_node_tree(wlr_scene_tree_from_node(node),
available, ignored_view);
}
}
}
void
edges_calculate_visibility(struct server *server, struct view *ignored_view)
{
/*
* The region stores the available output layout space
* and subtracts the window geometries in reverse rendering
* order, e.g. a window rendered on top is subtracted first.
*
* This allows to detect if a window is actually visible.
* If there is no overlap of its geometry and the remaining
* region it must be completely covered by other windows.
*
*/
pixman_region32_t region;
pixman_region32_init(&region);
/*
* Initialize the region with each individual output.
*
* If we were to use NULL for the reference output we
* would get a single combined wlr_box of the whole
* layout which could cover actual invisible areas
* in case the output resolutions differ.
*/
struct output *output;
struct wlr_box layout_box;
wl_list_for_each(output, &server->outputs, link) {
if (!output_is_usable(output)) {
continue;
}
wlr_output_layout_get_box(server->output_layout,
output->wlr_output, &layout_box);
pixman_region32_union_rect(&region, &region,
layout_box.x, layout_box.y, layout_box.width, layout_box.height);
}
subtract_node_tree(&server->scene->tree, &region, ignored_view);
pixman_region32_fini(&region);
}
void
edges_find_neighbors(struct border *nearest_edges, struct view *view,
struct wlr_box origin, struct wlr_box target,
struct output *output, edge_validator_t validator, bool ignore_hidden)
{
assert(view);
assert(validator);
assert(nearest_edges);
if (!output_is_usable(view->output)) {
wlr_log(WLR_DEBUG, "ignoring edge search for view on unusable output");
return;
}
struct border view_edges = { 0 };
struct border target_edges = { 0 };
edges_for_target_geometry(&view_edges, view, origin);
edges_for_target_geometry(&target_edges, view, target);
struct view *v;
for_each_view(v, &view->server->views, LAB_VIEW_CRITERIA_CURRENT_WORKSPACE) {
if (v == view || v->minimized || !output_is_usable(v->output)) {
continue;
}
uint32_t edges_visible = ignore_hidden ? v->edges_visible :
WLR_EDGE_TOP | WLR_EDGE_LEFT
| WLR_EDGE_BOTTOM | WLR_EDGE_RIGHT;
if (edges_visible == 0) {
continue;
}
if (output && output != v->output && !view_on_output(v, output)) {
continue;
}
/* Both view and v must share a common output */
if (view->output != v->output && !(view->outputs & v->outputs)) {
continue;
}
struct border border = ssd_get_margin(v->ssd);
struct border win_edges = {
.top = v->current.y - border.top,
.left = v->current.x - border.left,
.bottom = v->current.y + border.bottom
+ view_effective_height(v, /* use_pending */ false),
.right = v->current.x + v->current.width + border.right,
};
validate_edges(nearest_edges, view_edges,
target_edges, win_edges, edges_visible, validator);
}
}
void
edges_find_outputs(struct border *nearest_edges, struct view *view,
struct wlr_box origin, struct wlr_box target,
struct output *output, edge_validator_t validator)
{
assert(view);
assert(validator);
assert(nearest_edges);
if (!output_is_usable(view->output)) {
wlr_log(WLR_DEBUG,
"ignoring edge search for view on unusable output");
return;
}
struct border view_edges = { 0 };
struct border target_edges = { 0 };
edges_for_target_geometry(&view_edges, view, origin);
edges_for_target_geometry(&target_edges, view, target);
struct output *o;
wl_list_for_each(o, &view->server->outputs, link) {
if (!output_is_usable(o)) {
continue;
}
if (output && o != output) {
continue;
}
struct wlr_box usable =
output_usable_area_in_layout_coords(o);
if (!box_intersects(&origin, &usable)
&& !box_intersects(&target, &usable)) {
continue;
}
validate_output_edges(nearest_edges,
view_edges, target_edges, usable, validator);
}
}
static void
adjust_move_coords_1d(int *edge, int lesser, int lesser_offset,
int greater, int greater_offset, bool decreasing)
{
/* Default best candidate is not valid */
int best = INT_MAX;
if (BOUNDED_INT(lesser)) {
/* A valid lesser edge is the always the first candidate */
best = clipped_add(lesser, lesser_offset);
}
if (BOUNDED_INT(greater)) {
/* Check if a valid greater edge is a better candidate */
best = edge_get_best(best,
clipped_sub(greater, greater_offset), decreasing);
}
if (BOUNDED_INT(best)) {
/* Replace the edge if a valid candidate was found */
*edge = best;
}
}
void
edges_adjust_move_coords(struct view *view, struct border edges,
int *x, int *y, bool use_pending)
{
assert(view);
struct border border = ssd_get_margin(view->ssd);
struct wlr_box *view_geom =
use_pending ? &view->pending : &view->current;
/* When moving, limit motion to the best valid, intervening edge */
if (view_geom->x != *x) {
int lshift = border.left + rc.gap;
int rshift = border.right + rc.gap + view_geom->width;
adjust_move_coords_1d(x, edges.left, lshift,
edges.right, rshift, *x < view_geom->x);
}
if (view_geom->y != *y) {
int tshift = border.top + rc.gap;
int bshift = border.bottom + rc.gap
+ view_effective_height(view, use_pending);
adjust_move_coords_1d(y, edges.top, tshift,
edges.bottom, bshift, *y < view_geom->y);
}
}
void
edges_adjust_resize_geom(struct view *view, struct border edges,
uint32_t resize_edges, struct wlr_box *geom, bool use_pending)
{
assert(view);
struct border border = ssd_get_margin(view->ssd);
struct wlr_box *view_geom =
use_pending ? &view->pending : &view->current;
/*
* When resizing along a given edge, limit the motion of that edge to
* any valid nearest edge in the corresponding direction.
*/
if (resize_edges & WLR_EDGE_LEFT) {
if (BOUNDED_INT(edges.left)) {
geom->x = edges.left + border.left + rc.gap;
geom->width = view_geom->width + view_geom->x - geom->x;
}
} else if (resize_edges & WLR_EDGE_RIGHT) {
if (BOUNDED_INT(edges.right)) {
geom->width = edges.right
- view_geom->x - border.right - rc.gap;
}
}
if (resize_edges & WLR_EDGE_TOP) {
if (BOUNDED_INT(edges.top)) {
geom->y = edges.top + border.top + rc.gap;
geom->height = view_geom->height + view_geom->y - geom->y;
}
} else if (resize_edges & WLR_EDGE_BOTTOM) {
if (BOUNDED_INT(edges.bottom)) {
geom->height = edges.bottom
- view_geom->y - border.bottom - rc.gap;
}
}
}
static double
linear_interp(int x, int x1, int y1, int x2, int y2)
{
/*
* For a line y = mx + b that passes through both (x1, y1) and
* (x2, y2), find and return the value y for a given point x.
*
* The point x does not need to fall in the range [x1, x2].
*/
/* No need to interpolate if line is horizontal */
int rise = y2 - y1;
if (rise == 0) {
return y2;
}
/* For degenerate line, just pick a midpoint */
int run = x2 - x1;
if (run == 0) {
return 0.5 * (y1 + y2);
}
/* Otherwise, linearly interpolate */
int dx = x - x1;
return y1 + dx * (rise / (double)run);
}
bool
edges_traverse_edge(struct edge current, struct edge target, struct edge obstacle)
{
/*
* Each edge structure defines a line segment that can be represented
* in a local coordinate system as starting at (offset, min) and
* finishing at (offset, max).
*
* The starting and ending points of the "current" edge trace
* respective lines
*
* 1. (current.offset, current.min) -> (target.offset, target.min)
* 2. (current.offset, current.max) -> (target.offset, target.max)
*
* as the segment transits from its current position to its target.
* Hence, motion of the entire edge from current to target will sweep a
* quadrilateral bounded by (locally) vertical lines at current.offset
* and target.offset as well as the segments (1) and (2) above.
*
* To test if the motion will encounter the obstacle edge, we need to
* test if any of the obstacle edge falls within this quadrilateral.
* Thus, we need to find the extent of the quadrilateral at the same
* offset as the obstacle: a segment with starting point
* (obstacle.offset, lo) and ending point (obstacle.offset, hi).
*/
double lo =
linear_interp(obstacle.offset,
current.offset, current.min, target.offset, target.min);
/* Motion misses when obstacle ends above start of quad segment */
if (obstacle.max <= lo) {
return false;
}
double hi =
linear_interp(obstacle.offset,
current.offset, current.max, target.offset, target.max);
/* Motion hits when obstacle starts above the end of quad segment */
return obstacle.min < hi;
}