sway/sway/desktop/transaction.c
Rouven Czerwinski e4909ab4a3 transaction: destroying nodes aren't hidden
Commit 37d7bc6998 ("transaction: Only wait for ack from visible
views") introduced a check which uses view_is_visible() to check if a view
is still visible on the screen. However view_is_visible() will early
return in case the node is in the destroying state. This is incorrect
for transactions, since a destroying view which is visible will trigger
configure events for other clients. This bug was visible when repeatedly
opening and closing two views side by side, since we ignore the
destroying node we get a frame where the still open view is shown with
the old configure values and the rest is the desktop background. The
next frame is than correct again.

Fix this by considering destroying views as visible, we correctly wait
for them and send the configure events to other views in time, fixing
the background flicker.

Fixes #6473
2022-01-18 19:42:15 +01:00

553 lines
16 KiB
C

#define _POSIX_C_SOURCE 200809L
#include <stdbool.h>
#include <stdlib.h>
#include <string.h>
#include <time.h>
#include <wlr/types/wlr_buffer.h>
#include "sway/config.h"
#include "sway/desktop.h"
#include "sway/desktop/idle_inhibit_v1.h"
#include "sway/desktop/transaction.h"
#include "sway/input/cursor.h"
#include "sway/input/input-manager.h"
#include "sway/output.h"
#include "sway/tree/container.h"
#include "sway/tree/node.h"
#include "sway/tree/view.h"
#include "sway/tree/workspace.h"
#include "list.h"
#include "log.h"
struct sway_transaction {
struct wl_event_source *timer;
list_t *instructions; // struct sway_transaction_instruction *
size_t num_waiting;
size_t num_configures;
struct timespec commit_time;
};
struct sway_transaction_instruction {
struct sway_transaction *transaction;
struct sway_node *node;
union {
struct sway_output_state output_state;
struct sway_workspace_state workspace_state;
struct sway_container_state container_state;
};
uint32_t serial;
bool server_request;
bool waiting;
};
static struct sway_transaction *transaction_create(void) {
struct sway_transaction *transaction =
calloc(1, sizeof(struct sway_transaction));
if (!sway_assert(transaction, "Unable to allocate transaction")) {
return NULL;
}
transaction->instructions = create_list();
return transaction;
}
static void transaction_destroy(struct sway_transaction *transaction) {
// Free instructions
for (int i = 0; i < transaction->instructions->length; ++i) {
struct sway_transaction_instruction *instruction =
transaction->instructions->items[i];
struct sway_node *node = instruction->node;
node->ntxnrefs--;
if (node->instruction == instruction) {
node->instruction = NULL;
}
if (node->destroying && node->ntxnrefs == 0) {
switch (node->type) {
case N_ROOT:
sway_assert(false, "Never reached");
break;
case N_OUTPUT:
output_destroy(node->sway_output);
break;
case N_WORKSPACE:
workspace_destroy(node->sway_workspace);
break;
case N_CONTAINER:
container_destroy(node->sway_container);
break;
}
}
free(instruction);
}
list_free(transaction->instructions);
if (transaction->timer) {
wl_event_source_remove(transaction->timer);
}
free(transaction);
}
static void copy_output_state(struct sway_output *output,
struct sway_transaction_instruction *instruction) {
struct sway_output_state *state = &instruction->output_state;
if (state->workspaces) {
state->workspaces->length = 0;
} else {
state->workspaces = create_list();
}
list_cat(state->workspaces, output->workspaces);
state->active_workspace = output_get_active_workspace(output);
}
static void copy_workspace_state(struct sway_workspace *ws,
struct sway_transaction_instruction *instruction) {
struct sway_workspace_state *state = &instruction->workspace_state;
state->fullscreen = ws->fullscreen;
state->x = ws->x;
state->y = ws->y;
state->width = ws->width;
state->height = ws->height;
state->layout = ws->layout;
state->output = ws->output;
if (state->floating) {
state->floating->length = 0;
} else {
state->floating = create_list();
}
if (state->tiling) {
state->tiling->length = 0;
} else {
state->tiling = create_list();
}
list_cat(state->floating, ws->floating);
list_cat(state->tiling, ws->tiling);
struct sway_seat *seat = input_manager_current_seat();
state->focused = seat_get_focus(seat) == &ws->node;
// Set focused_inactive_child to the direct tiling child
struct sway_container *focus = seat_get_focus_inactive_tiling(seat, ws);
if (focus) {
while (focus->pending.parent) {
focus = focus->pending.parent;
}
}
state->focused_inactive_child = focus;
}
static void copy_container_state(struct sway_container *container,
struct sway_transaction_instruction *instruction) {
struct sway_container_state *state = &instruction->container_state;
if (state->children) {
list_free(state->children);
}
memcpy(state, &container->pending, sizeof(struct sway_container_state));
if (!container->view) {
// We store a copy of the child list to avoid having it mutated after
// we copy the state.
state->children = create_list();
list_cat(state->children, container->pending.children);
} else {
state->children = NULL;
}
struct sway_seat *seat = input_manager_current_seat();
state->focused = seat_get_focus(seat) == &container->node;
if (!container->view) {
struct sway_node *focus =
seat_get_active_tiling_child(seat, &container->node);
state->focused_inactive_child = focus ? focus->sway_container : NULL;
}
}
static void transaction_add_node(struct sway_transaction *transaction,
struct sway_node *node, bool server_request) {
struct sway_transaction_instruction *instruction = NULL;
// Check if we have an instruction for this node already, in which case we
// update that instead of creating a new one.
if (node->ntxnrefs > 0) {
for (int idx = 0; idx < transaction->instructions->length; idx++) {
struct sway_transaction_instruction *other =
transaction->instructions->items[idx];
if (other->node == node) {
instruction = other;
break;
}
}
}
if (!instruction) {
instruction = calloc(1, sizeof(struct sway_transaction_instruction));
if (!sway_assert(instruction, "Unable to allocate instruction")) {
return;
}
instruction->transaction = transaction;
instruction->node = node;
instruction->server_request = server_request;
list_add(transaction->instructions, instruction);
node->ntxnrefs++;
} else if (server_request) {
instruction->server_request = true;
}
switch (node->type) {
case N_ROOT:
break;
case N_OUTPUT:
copy_output_state(node->sway_output, instruction);
break;
case N_WORKSPACE:
copy_workspace_state(node->sway_workspace, instruction);
break;
case N_CONTAINER:
copy_container_state(node->sway_container, instruction);
break;
}
}
static void apply_output_state(struct sway_output *output,
struct sway_output_state *state) {
output_damage_whole(output);
list_free(output->current.workspaces);
memcpy(&output->current, state, sizeof(struct sway_output_state));
output_damage_whole(output);
}
static void apply_workspace_state(struct sway_workspace *ws,
struct sway_workspace_state *state) {
output_damage_whole(ws->current.output);
list_free(ws->current.floating);
list_free(ws->current.tiling);
memcpy(&ws->current, state, sizeof(struct sway_workspace_state));
output_damage_whole(ws->current.output);
}
static void apply_container_state(struct sway_container *container,
struct sway_container_state *state) {
struct sway_view *view = container->view;
// Damage the old location
desktop_damage_whole_container(container);
if (view && !wl_list_empty(&view->saved_buffers)) {
struct sway_saved_buffer *saved_buf;
wl_list_for_each(saved_buf, &view->saved_buffers, link) {
struct wlr_box box = {
.x = saved_buf->x - view->saved_geometry.x,
.y = saved_buf->y - view->saved_geometry.y,
.width = saved_buf->width,
.height = saved_buf->height,
};
desktop_damage_box(&box);
}
}
// There are separate children lists for each instruction state, the
// container's current state and the container's pending state
// (ie. con->children). The list itself needs to be freed here.
// Any child containers which are being deleted will be cleaned up in
// transaction_destroy().
list_free(container->current.children);
memcpy(&container->current, state, sizeof(struct sway_container_state));
if (view && !wl_list_empty(&view->saved_buffers)) {
if (!container->node.destroying || container->node.ntxnrefs == 1) {
view_remove_saved_buffer(view);
}
}
// If the view hasn't responded to the configure, center it within
// the container. This is important for fullscreen views which
// refuse to resize to the size of the output.
if (view && view->surface) {
view_center_surface(view);
}
// Damage the new location
desktop_damage_whole_container(container);
if (view && view->surface) {
struct wlr_surface *surface = view->surface;
struct wlr_box box = {
.x = container->current.content_x - view->geometry.x,
.y = container->current.content_y - view->geometry.y,
.width = surface->current.width,
.height = surface->current.height,
};
desktop_damage_box(&box);
}
if (!container->node.destroying) {
container_discover_outputs(container);
}
}
/**
* Apply a transaction to the "current" state of the tree.
*/
static void transaction_apply(struct sway_transaction *transaction) {
sway_log(SWAY_DEBUG, "Applying transaction %p", transaction);
if (debug.txn_timings) {
struct timespec now;
clock_gettime(CLOCK_MONOTONIC, &now);
struct timespec *commit = &transaction->commit_time;
float ms = (now.tv_sec - commit->tv_sec) * 1000 +
(now.tv_nsec - commit->tv_nsec) / 1000000.0;
sway_log(SWAY_DEBUG, "Transaction %p: %.1fms waiting "
"(%.1f frames if 60Hz)", transaction, ms, ms / (1000.0f / 60));
}
// Apply the instruction state to the node's current state
for (int i = 0; i < transaction->instructions->length; ++i) {
struct sway_transaction_instruction *instruction =
transaction->instructions->items[i];
struct sway_node *node = instruction->node;
switch (node->type) {
case N_ROOT:
break;
case N_OUTPUT:
apply_output_state(node->sway_output, &instruction->output_state);
break;
case N_WORKSPACE:
apply_workspace_state(node->sway_workspace,
&instruction->workspace_state);
break;
case N_CONTAINER:
apply_container_state(node->sway_container,
&instruction->container_state);
break;
}
node->instruction = NULL;
}
cursor_rebase_all();
}
static void transaction_commit_pending(void);
static void transaction_progress(void) {
if (!server.queued_transaction) {
return;
}
if (server.queued_transaction->num_waiting > 0) {
return;
}
transaction_apply(server.queued_transaction);
transaction_destroy(server.queued_transaction);
server.queued_transaction = NULL;
if (!server.pending_transaction) {
sway_idle_inhibit_v1_check_active(server.idle_inhibit_manager_v1);
return;
}
transaction_commit_pending();
}
static int handle_timeout(void *data) {
struct sway_transaction *transaction = data;
sway_log(SWAY_DEBUG, "Transaction %p timed out (%zi waiting)",
transaction, transaction->num_waiting);
transaction->num_waiting = 0;
transaction_progress();
return 0;
}
static bool should_configure(struct sway_node *node,
struct sway_transaction_instruction *instruction) {
if (!node_is_view(node)) {
return false;
}
if (node->destroying) {
return false;
}
if (!instruction->server_request) {
return false;
}
struct sway_container_state *cstate = &node->sway_container->current;
struct sway_container_state *istate = &instruction->container_state;
#if HAVE_XWAYLAND
// Xwayland views are position-aware and need to be reconfigured
// when their position changes.
if (node->sway_container->view->type == SWAY_VIEW_XWAYLAND) {
// Sway logical coordinates are doubles, but they get truncated to
// integers when sent to Xwayland through `xcb_configure_window`.
// X11 apps will not respond to duplicate configure requests (from their
// truncated point of view) and cause transactions to time out.
if ((int)cstate->content_x != (int)istate->content_x ||
(int)cstate->content_y != (int)istate->content_y) {
return true;
}
}
#endif
if (cstate->content_width == istate->content_width &&
cstate->content_height == istate->content_height) {
return false;
}
return true;
}
static void transaction_commit(struct sway_transaction *transaction) {
sway_log(SWAY_DEBUG, "Transaction %p committing with %i instructions",
transaction, transaction->instructions->length);
transaction->num_waiting = 0;
for (int i = 0; i < transaction->instructions->length; ++i) {
struct sway_transaction_instruction *instruction =
transaction->instructions->items[i];
struct sway_node *node = instruction->node;
bool hidden = node_is_view(node) && !node->destroying &&
!view_is_visible(node->sway_container->view);
if (should_configure(node, instruction)) {
instruction->serial = view_configure(node->sway_container->view,
instruction->container_state.content_x,
instruction->container_state.content_y,
instruction->container_state.content_width,
instruction->container_state.content_height);
if (!hidden) {
instruction->waiting = true;
++transaction->num_waiting;
}
// From here on we are rendering a saved buffer of the view, which
// means we can send a frame done event to make the client redraw it
// as soon as possible. Additionally, this is required if a view is
// mapping and its default geometry doesn't intersect an output.
struct timespec now;
clock_gettime(CLOCK_MONOTONIC, &now);
wlr_surface_send_frame_done(
node->sway_container->view->surface, &now);
}
if (!hidden && node_is_view(node) &&
wl_list_empty(&node->sway_container->view->saved_buffers)) {
view_save_buffer(node->sway_container->view);
memcpy(&node->sway_container->view->saved_geometry,
&node->sway_container->view->geometry,
sizeof(struct wlr_box));
}
node->instruction = instruction;
}
transaction->num_configures = transaction->num_waiting;
if (debug.txn_timings) {
clock_gettime(CLOCK_MONOTONIC, &transaction->commit_time);
}
if (debug.noatomic) {
transaction->num_waiting = 0;
} else if (debug.txn_wait) {
// Force the transaction to time out even if all views are ready.
// We do this by inflating the waiting counter.
transaction->num_waiting += 1000000;
}
if (transaction->num_waiting) {
// Set up a timer which the views must respond within
transaction->timer = wl_event_loop_add_timer(server.wl_event_loop,
handle_timeout, transaction);
if (transaction->timer) {
wl_event_source_timer_update(transaction->timer,
server.txn_timeout_ms);
} else {
sway_log_errno(SWAY_ERROR, "Unable to create transaction timer "
"(some imperfect frames might be rendered)");
transaction->num_waiting = 0;
}
}
}
static void transaction_commit_pending(void) {
if (server.queued_transaction) {
return;
}
struct sway_transaction *transaction = server.pending_transaction;
server.pending_transaction = NULL;
server.queued_transaction = transaction;
transaction_commit(transaction);
transaction_progress();
}
static void set_instruction_ready(
struct sway_transaction_instruction *instruction) {
struct sway_transaction *transaction = instruction->transaction;
if (debug.txn_timings) {
struct timespec now;
clock_gettime(CLOCK_MONOTONIC, &now);
struct timespec *start = &transaction->commit_time;
float ms = (now.tv_sec - start->tv_sec) * 1000 +
(now.tv_nsec - start->tv_nsec) / 1000000.0;
sway_log(SWAY_DEBUG, "Transaction %p: %zi/%zi ready in %.1fms (%s)",
transaction,
transaction->num_configures - transaction->num_waiting + 1,
transaction->num_configures, ms,
instruction->node->sway_container->title);
}
// If the transaction has timed out then its num_waiting will be 0 already.
if (instruction->waiting && transaction->num_waiting > 0 &&
--transaction->num_waiting == 0) {
sway_log(SWAY_DEBUG, "Transaction %p is ready", transaction);
wl_event_source_timer_update(transaction->timer, 0);
}
instruction->node->instruction = NULL;
transaction_progress();
}
void transaction_notify_view_ready_by_serial(struct sway_view *view,
uint32_t serial) {
struct sway_transaction_instruction *instruction =
view->container->node.instruction;
if (instruction != NULL && instruction->serial == serial) {
set_instruction_ready(instruction);
}
}
void transaction_notify_view_ready_by_geometry(struct sway_view *view,
double x, double y, int width, int height) {
struct sway_transaction_instruction *instruction =
view->container->node.instruction;
if (instruction != NULL &&
(int)instruction->container_state.content_x == (int)x &&
(int)instruction->container_state.content_y == (int)y &&
instruction->container_state.content_width == width &&
instruction->container_state.content_height == height) {
set_instruction_ready(instruction);
}
}
static void _transaction_commit_dirty(bool server_request) {
if (!server.dirty_nodes->length) {
return;
}
if (!server.pending_transaction) {
server.pending_transaction = transaction_create();
if (!server.pending_transaction) {
return;
}
}
for (int i = 0; i < server.dirty_nodes->length; ++i) {
struct sway_node *node = server.dirty_nodes->items[i];
transaction_add_node(server.pending_transaction, node, server_request);
node->dirty = false;
}
server.dirty_nodes->length = 0;
transaction_commit_pending();
}
void transaction_commit_dirty(void) {
_transaction_commit_dirty(true);
}
void transaction_commit_dirty_client(void) {
_transaction_commit_dirty(false);
}