Skip to content

Instantly share code, notes, and snippets.

Last active May 5, 2023 16:26
What would you like to do?
Tiny Wayland compositor
# wayland-scanner is a tool which generates C headers and rigging for Wayland
# protocols, which are specified in XML. wlroots requires you to rig these up
# to your build system yourself and provide them in the include path.
wayland-scanner server-header \
$(WAYLAND_PROTOCOLS)/stable/xdg-shell/xdg-shell.xml $@
xdg-shell-protocol.c: xdg-shell-protocol.h
wayland-scanner private-code \
$(WAYLAND_PROTOCOLS)/stable/xdg-shell/xdg-shell.xml $@
tinywl: tinywl.c xdg-shell-protocol.h xdg-shell-protocol.c
$(CC) $(CFLAGS) \
-g -Werror -I. \
$(shell pkg-config --cflags --libs wlroots) \
$(shell pkg-config --cflags --libs wayland-server) \
$(shell pkg-config --cflags --libs xkbcommon) \
-o $@ $<
rm -f tinywl xdg-shell-protocol.h xdg-shell-protocol.c
.PHONY: clean
* TinyWL: A tiny(ish) Wayland compositor.
* This is the annotated source. The canonical source is available here:
* It is recommended that you read this file from the bottom-up. The entrypoint
* is at the bottom, and following the logic through there may be easier than
* starting at the top. You are also encouraged to consult wayland-server-util.h
* if necessary, I will not be explaining e.g. wl_list, wl_container_of, etc.
* This is based on wlroots[0], a library which makes building Wayland
* compositors easier. Without it, this could easily be 10,000+ lines of code.
* [0]
* Background:
* In the X11 world, X11 window managers have it easy. The majority of input
* device handling, input dispatching, rendering, output resource management,
* and so on, is all handled by the X11 server. The window manager simply tells
* it where to place the windows, who gets focus, etc.
* Our Wayland compositor has to do _all_ of this. This file is responsible for:
* - Everything an X11 WM would do
* - Negotiating window allocation with clients
* - Rendering windows to the display
* - Allocating output resources and managing a frame loop
* - Allocating input devices and dispatching input events
* - Focusing the appropriate windows in the appropriate circumstances
* - Multihead support
* wlroots is a very batteries-not-included library. It's the most flexible and
* comprehensive Wayland compositor library, but it requires a lot of work to
* get something working. Hence, this file is 582 SLOC and TinyWM[1] is only 50.
* In exchange for the extra work, you're able to make a huge variety of Wayland
* compositors and other applications based on wlroots.
* [1]
* Notable omissions from TinyWL:
* - HiDPI support
* - Any kind of configuration, e.g. output layout
* - Any protocol other than xdg-shell (e.g. layer-shell, for
* panels/taskbars/etc; or Xwayland, for proxied X11 windows)
* - Optional protocols, e.g. screen capture, primary selection, virtual
* keyboard, etc. Most of these are plug-and-play with wlroots, but they're
* omitted for brevity.
* - Damage tracking, which tracks which parts of the screen are changing and
* minimizes redraws accordingly.
* You can run this by simply running `./tinywl`. If you run it from within an
* existing X11 or Wayland session, wlroots will open a virtual output as a
* window in your existing session. If you run it at a tty, it'll run with the
* KMS+DRM backend. Use e.g. `./tinywl -s termite` to start with a terminal
* running, which will be necessary to run anything interesting. Note that
* weston-terminal will not work; it uses xdg-shell v6, which is supported by
* wlroots but not implemented in TinyWL. The wlroots xdg-shell v6 interface is
* very similar to the xdg-shell interface, implementing it may be a fun
* exercise for the reader.
* Press Alt+ESC to quit, and Alt+F1 to cycle between windows. Most Wayland
* clients provide their own window decorations and means of moving the window
* around; no TinyWL keybinding is provided for this purpose.
* For additional resources, please review my blog posts on Wayland compositor
* development:
* More coming soon!
#define _POSIX_C_SOURCE 200112L
#include <getopt.h>
#include <stdbool.h>
#include <stdlib.h>
#include <stdio.h>
#include <time.h>
#include <unistd.h>
#include <wayland-server.h>
#include <wlr/backend.h>
#include <wlr/render/wlr_renderer.h>
#include <wlr/types/wlr_cursor.h>
#include <wlr/types/wlr_compositor.h>
#include <wlr/types/wlr_data_device.h>
#include <wlr/types/wlr_input_device.h>
#include <wlr/types/wlr_keyboard.h>
#include <wlr/types/wlr_linux_dmabuf_v1.h>
#include <wlr/types/wlr_matrix.h>
#include <wlr/types/wlr_output.h>
#include <wlr/types/wlr_output_layout.h>
#include <wlr/types/wlr_pointer.h>
#include <wlr/types/wlr_seat.h>
#include <wlr/types/wlr_xcursor_manager.h>
#include <wlr/types/wlr_xdg_shell.h>
#include <wlr/util/log.h>
#include <xkbcommon/xkbcommon.h>
/* For brevity's sake, the struct members are annotated where they are used. */
enum tinywl_cursor_mode {
struct tinywl_server {
struct wl_display *wl_display;
struct wlr_backend *backend;
struct wlr_renderer *renderer;
struct wlr_xdg_shell *xdg_shell;
struct wl_listener new_xdg_surface;
struct wl_list views;
struct wlr_cursor *cursor;
struct wlr_xcursor_manager *cursor_mgr;
struct wl_listener cursor_motion;
struct wl_listener cursor_motion_absolute;
struct wl_listener cursor_button;
struct wl_listener cursor_axis;
struct wlr_seat *seat;
struct wl_listener new_input;
struct wl_listener request_cursor;
struct wl_list keyboards;
enum tinywl_cursor_mode cursor_mode;
struct tinywl_view *grabbed_view;
double grab_x, grab_y;
int grab_width, grab_height;
uint32_t resize_edges;
struct wlr_output_layout *output_layout;
struct wl_list outputs;
struct wl_listener new_output;
struct tinywl_output {
struct wl_list link;
struct tinywl_server *server;
struct wlr_output *wlr_output;
struct wl_listener frame;
struct tinywl_view {
struct wl_list link;
struct tinywl_server *server;
struct wlr_xdg_surface *xdg_surface;
struct wl_listener map;
struct wl_listener unmap;
struct wl_listener destroy;
struct wl_listener request_move;
struct wl_listener request_resize;
bool mapped;
int x, y;
struct tinywl_keyboard {
struct wl_list link;
struct tinywl_server *server;
struct wlr_input_device *device;
struct wl_listener modifiers;
struct wl_listener key;
struct tinywl_pointer {
struct wl_list link;
struct tinywl_server *server;
struct wlr_input_device *device;
static void focus_view(struct tinywl_view *view, struct wlr_surface *surface) {
/* Note: this function only deals with keyboard focus. */
if (view == NULL) {
struct tinywl_server *server = view->server;
struct wlr_seat *seat = server->seat;
struct wlr_surface *prev_surface = seat->keyboard_state.focused_surface;
if (prev_surface == surface) {
/* Don't re-focus an already focused surface. */
if (prev_surface) {
* Deactivate the previously focused surface. This lets the client know
* it no longer has focus and the client will repaint accordingly, e.g.
* stop displaying a caret.
struct wlr_xdg_surface *previous = wlr_xdg_surface_from_wlr_surface(
wlr_xdg_toplevel_set_activated(previous, false);
struct wlr_keyboard *keyboard = wlr_seat_get_keyboard(seat);
/* Move the view to the front */
wl_list_insert(&server->views, &view->link);
/* Activate the new surface */
wlr_xdg_toplevel_set_activated(view->xdg_surface, true);
* Tell the seat to have the keyboard enter this surface. wlroots will keep
* track of this and automatically send key events to the appropriate
* clients without additional work on your part.
wlr_seat_keyboard_notify_enter(seat, view->xdg_surface->surface,
keyboard->keycodes, keyboard->num_keycodes, &keyboard->modifiers);
static void keyboard_handle_modifiers(
struct wl_listener *listener, void *data) {
/* This event is raised when a modifier key, such as shift or alt, is
* pressed. We simply communicate this to the client. */
struct tinywl_keyboard *keyboard =
wl_container_of(listener, keyboard, modifiers);
* A seat can only have one keyboard, but this is a limitation of the
* Wayland protocol. We assign all connected keyboards to the same seat.
* You can swap it like this and wlr_seat handles this transparently.
wlr_seat_set_keyboard(keyboard->server->seat, keyboard->device);
/* Send modifiers to the client. */
static bool handle_keybinding(struct tinywl_server *server, xkb_keysym_t sym) {
* Here we handle compositor keybindings. This is when the compositor is
* processing keys, rather than passing them on to the client for its own
* processing.
* This function assumes Alt is held down.
switch (sym) {
case XKB_KEY_Escape:
case XKB_KEY_F1:
/* Cycle to the next view */
if (wl_list_length(&server->views) < 2) {
struct tinywl_view *current_view = wl_container_of(
server->, current_view, link);
struct tinywl_view *next_view = wl_container_of(
current_view->, next_view, link);
focus_view(next_view, next_view->xdg_surface->surface);
/* Move the previous view to the end of the list */
wl_list_insert(server->views.prev, &current_view->link);
return false;
return true;
static void keyboard_handle_key(
struct wl_listener *listener, void *data) {
/* This event is raised when a key is pressed or released. */
struct tinywl_keyboard *keyboard =
wl_container_of(listener, keyboard, key);
struct tinywl_server *server = keyboard->server;
struct wlr_event_keyboard_key *event = data;
struct wlr_seat *seat = server->seat;
/* Translate libinput keycode -> xkbcommon */
uint32_t keycode = event->keycode + 8;
/* Get a list of keysyms based on the keymap for this keyboard */
const xkb_keysym_t *syms;
int nsyms = xkb_state_key_get_syms(
keyboard->device->keyboard->xkb_state, keycode, &syms);
bool handled = false;
uint32_t modifiers = wlr_keyboard_get_modifiers(keyboard->device->keyboard);
if ((modifiers & WLR_MODIFIER_ALT) && event->state == WLR_BUTTON_PRESSED) {
/* If alt is held down and this button was _pressed_, we attempt to
* process it as a compositor keybinding. */
for (int i = 0; i < nsyms; i++) {
handled = handle_keybinding(server, syms[i]);
if (!handled) {
/* Otherwise, we pass it along to the client. */
wlr_seat_set_keyboard(seat, keyboard->device);
wlr_seat_keyboard_notify_key(seat, event->time_msec,
event->keycode, event->state);
static void server_new_keyboard(struct tinywl_server *server,
struct wlr_input_device *device) {
struct tinywl_keyboard *keyboard =
calloc(1, sizeof(struct tinywl_keyboard));
keyboard->server = server;
keyboard->device = device;
/* We need to prepare an XKB keymap and assign it to the keyboard. This
* assumes the defaults (e.g. layout = "us"). */
struct xkb_rule_names rules = { 0 };
struct xkb_context *context = xkb_context_new(XKB_CONTEXT_NO_FLAGS);
struct xkb_keymap *keymap = xkb_map_new_from_names(context, &rules,
wlr_keyboard_set_keymap(device->keyboard, keymap);
wlr_keyboard_set_repeat_info(device->keyboard, 25, 600);
/* Here we set up listeners for keyboard events. */
keyboard->modifiers.notify = keyboard_handle_modifiers;
wl_signal_add(&device->keyboard->events.modifiers, &keyboard->modifiers);
keyboard->key.notify = keyboard_handle_key;
wl_signal_add(&device->keyboard->events.key, &keyboard->key);
wlr_seat_set_keyboard(server->seat, device);
/* And add the keyboard to our list of keyboards */
wl_list_insert(&server->keyboards, &keyboard->link);
static void server_new_pointer(struct tinywl_server *server,
struct wlr_input_device *device) {
/* We don't do anything special with pointers. All of our pointer handling
* is proxied through wlr_cursor. On another compositor, you might take this
* opportunity to do libinput configuration on the device to set
* acceleration, etc. */
wlr_cursor_attach_input_device(server->cursor, device);
static void server_new_input(struct wl_listener *listener, void *data) {
/* This event is raised by the backend when a new input device becomes
* available. */
struct tinywl_server *server =
wl_container_of(listener, server, new_input);
struct wlr_input_device *device = data;
switch (device->type) {
server_new_keyboard(server, device);
server_new_pointer(server, device);
/* We need to let the wlr_seat know what our capabilities are, which is
* communiciated to the client. In TinyWL we always have a cursor, even if
* there are no pointer devices, so we always include that capability. */
if (!wl_list_empty(&server->keyboards)) {
wlr_seat_set_capabilities(server->seat, caps);
static void seat_request_cursor(struct wl_listener *listener, void *data) {
/* This event is rasied by the seat when a client provides a cursor image */
struct tinywl_server *server = wl_container_of(
listener, server, request_cursor);
struct wlr_seat_pointer_request_set_cursor_event *event = data;
struct wlr_seat_client *focused_client =
/* This can be sent by any client, so we check to make sure this one is
* actually has pointer focus first. */
if (focused_client == event->seat_client) {
/* Once we've vetted the client, we can tell the cursor to use the
* provided surface as the cursor image. It will set the hardware cursor
* on the output that it's currently on and continue to do so as the
* cursor moves between outputs. */
wlr_cursor_set_surface(server->cursor, event->surface,
event->hotspot_x, event->hotspot_y);
static bool view_at(struct tinywl_view *view,
double lx, double ly, struct wlr_surface **surface,
double *sx, double *sy) {
* XDG toplevels may have nested surfaces, such as popup windows for context
* menus or tooltips. This function tests if any of those are underneath the
* coordinates lx and ly (in output Layout Coordinates). If so, it sets the
* surface pointer to that wlr_surface and the sx and sy coordinates to the
* coordinates relative to that surface's top-left corner.
double view_sx = lx - view->x;
double view_sy = ly - view->y;
struct wlr_surface_state *state = &view->xdg_surface->surface->current;
struct wlr_box box = {
.x = 0, .y = 0,
.width = state->width, .height = state->height,
double _sx, _sy;
struct wlr_surface *_surface = NULL;
_surface = wlr_xdg_surface_surface_at(
view->xdg_surface, view_sx, view_sy, &_sx, &_sy);
if (_surface != NULL) {
*sx = _sx;
*sy = _sy;
*surface = _surface;
return true;
return false;
static struct tinywl_view *desktop_view_at(
struct tinywl_server *server, double lx, double ly,
struct wlr_surface **surface, double *sx, double *sy) {
/* This iterates over all of our surfaces and attempts to find one under the
* cursor. This relies on server->views being ordered from top-to-bottom. */
struct tinywl_view *view;
wl_list_for_each(view, &server->views, link) {
if (view_at(view, lx, ly, surface, sx, sy)) {
return view;
return NULL;
static void process_cursor_move(struct tinywl_server *server, uint32_t time) {
/* Move the grabbed view to the new position. */
server->grabbed_view->x = server->cursor->x - server->grab_x;
server->grabbed_view->y = server->cursor->y - server->grab_y;
static void process_cursor_resize(struct tinywl_server *server, uint32_t time) {
* Resizing the grabbed view can be a little bit complicated, because we
* could be resizing from any corner or edge. This not only resizes the view
* on one or two axes, but can also move the view if you resize from the top
* or left edges (or top-left corner).
* Note that I took some shortcuts here. In a more fleshed-out compositor,
* you'd wait for the client to prepare a buffer at the new size, then
* commit any movement that was prepared.
struct tinywl_view *view = server->grabbed_view;
double dx = server->cursor->x - server->grab_x;
double dy = server->cursor->y - server->grab_y;
double x = view->x;
double y = view->y;
int width = server->grab_width;
int height = server->grab_height;
if (server->resize_edges & WLR_EDGE_TOP) {
y = server->grab_y + dy;
height -= dy;
if (height < 1) {
y += height;
} else if (server->resize_edges & WLR_EDGE_BOTTOM) {
height += dy;
if (server->resize_edges & WLR_EDGE_LEFT) {
x = server->grab_x + dx;
width -= dx;
if (width < 1) {
x += width;
} else if (server->resize_edges & WLR_EDGE_RIGHT) {
width += dx;
view->x = x;
view->y = y;
wlr_xdg_toplevel_set_size(view->xdg_surface, width, height);
static void process_cursor_motion(struct tinywl_server *server, uint32_t time) {
/* If the mode is non-passthrough, delegate to those functions. */
if (server->cursor_mode == TINYWL_CURSOR_MOVE) {
process_cursor_move(server, time);
} else if (server->cursor_mode == TINYWL_CURSOR_RESIZE) {
process_cursor_resize(server, time);
/* Otherwise, find the view under the pointer and send the event along. */
double sx, sy;
struct wlr_seat *seat = server->seat;
struct wlr_surface *surface = NULL;
struct tinywl_view *view = desktop_view_at(server,
server->cursor->x, server->cursor->y, &surface, &sx, &sy);
if (!view) {
/* If there's no view under the cursor, set the cursor image to a
* default. This is what makes the cursor image appear when you move it
* around the screen, not over any views. */
server->cursor_mgr, "left_ptr", server->cursor);
if (surface) {
bool focus_changed = seat->pointer_state.focused_surface != surface;
* "Enter" the surface if necessary. This lets the client know that the
* cursor has entered one of its surfaces.
* Note that this gives the surface "pointer focus", which is distinct
* from keyboard focus. You get pointer focus by moving the pointer over
* a window.
wlr_seat_pointer_notify_enter(seat, surface, sx, sy);
if (!focus_changed) {
/* The enter event contains coordinates, so we only need to notify
* on motion if the focus did not change. */
wlr_seat_pointer_notify_motion(seat, time, sx, sy);
} else {
/* Clear pointer focus so future button events and such are not sent to
* the last client to have the cursor over it. */
static void server_cursor_motion(struct wl_listener *listener, void *data) {
/* This event is forwarded by the cursor when a pointer emits a _relative_
* pointer motion event (i.e. a delta) */
struct tinywl_server *server =
wl_container_of(listener, server, cursor_motion);
struct wlr_event_pointer_motion *event = data;
/* The cursor doesn't move unless we tell it to. The cursor automatically
* handles constraining the motion to the output layout, as well as any
* special configuration applied for the specific input device which
* generated the event. You can pass NULL for the device if you want to move
* the cursor around without any input. */
wlr_cursor_move(server->cursor, event->device,
event->delta_x, event->delta_y);
process_cursor_motion(server, event->time_msec);
static void server_cursor_motion_absolute(
struct wl_listener *listener, void *data) {
/* This event is forwarded by the cursor when a pointer emits an _absolute_
* motion event, from 0..1 on each axis. This happens, for example, when
* wlroots is running under a Wayland window rather than KMS+DRM, and you
* move the mouse over the window. You could enter the window from any edge,
* so we have to warp the mouse there. There is also some hardware which
* emits these events. */
struct tinywl_server *server =
wl_container_of(listener, server, cursor_motion_absolute);
struct wlr_event_pointer_motion_absolute *event = data;
wlr_cursor_warp_absolute(server->cursor, event->device, event->x, event->y);
process_cursor_motion(server, event->time_msec);
static void server_cursor_button(struct wl_listener *listener, void *data) {
/* This event is forwarded by the cursor when a pointer emits a button
* event. */
struct tinywl_server *server =
wl_container_of(listener, server, cursor_button);
struct wlr_event_pointer_button *event = data;
/* Notify the client with pointer focus that a button press has occured */
event->time_msec, event->button, event->state);
if (event->state == WLR_BUTTON_RELEASED) {
/* If you released any buttons, we exit interactive move/resize mode. */
server->cursor_mode = TINYWL_CURSOR_PASSTHROUGH;
} else {
/* Focus that client if the button was _pressed_ */
double sx, sy;
struct wlr_seat *seat = server->seat;
struct wlr_surface *surface;
struct tinywl_view *view = desktop_view_at(server,
server->cursor->x, server->cursor->y, &surface, &sx, &sy);
focus_view(view, surface);
static void server_cursor_axis(struct wl_listener *listener, void *data) {
/* This event is forwarded by the cursor when a pointer emits an axis event,
* for example when you move the scroll wheel. */
struct tinywl_server *server =
wl_container_of(listener, server, cursor_axis);
struct wlr_event_pointer_axis *event = data;
/* Notify the client with pointer focus of the axis event. */
event->time_msec, event->orientation, event->delta,
event->delta_discrete, event->source);
/* Used to move all of the data necessary to render a surface from the top-level
* frame handler to the per-surface render function. */
struct render_data {
struct wlr_output *output;
struct wlr_renderer *renderer;
struct tinywl_view *view;
struct timespec *when;
static void render_surface(struct wlr_surface *surface,
int sx, int sy, void *data) {
/* This function is called for every surface that needs to be rendered. */
struct render_data *rdata = data;
struct tinywl_view *view = rdata->view;
struct wlr_output *output = rdata->output;
/* We first obtain a wlr_texture, which is a GPU resource. wlroots
* automatically handles negotiating these with the client. The underlying
* resource could be an opaque handle passed from the client, or the client
* could have sent a pixel buffer which we copied to the GPU, or a few other
* means. You don't have to worry about this, wlroots takes care of it. */
struct wlr_texture *texture = wlr_surface_get_texture(surface);
if (texture == NULL) {
/* The view has a position in layout coordinates. If you have two displays,
* one next to the other, both 1080p, a view on the rightmost display might
* have layout coordinates of 2000,100. We need to translate that to
* output-local coordinates, or (2000 - 1920). */
double ox = 0, oy = 0;
view->server->output_layout, output, &ox, &oy);
ox += view->x + sx, oy += view->y + sy;
/* We also have to apply the scale factor for HiDPI outputs. This is only
* part of the puzzle, TinyWL does not fully support HiDPI. */
struct wlr_box box = {
.x = ox * output->scale,
.y = oy * output->scale,
.width = surface->current.width * output->scale,
.height = surface->current.height * output->scale,
* Those familiar with OpenGL are also familiar with the role of matricies
* in graphics programming. We need to prepare a matrix to render the view
* with. wlr_matrix_project_box is a helper which takes a box with a desired
* x, y coodrinates, width and height, and an output geometry, then
* prepares an orthographic projection and multiplies the necessary
* transforms to produce a model-view-projection matrix.
* Naturally you can do this any way you like, for example to make a 3D
* compositor.
float matrix[9];
enum wl_output_transform transform =
wlr_matrix_project_box(matrix, &box, transform, 0,
/* This takes our matrix, the texture, and an alpha, and performs the actual
* rendering on the GPU. */
wlr_render_texture_with_matrix(rdata->renderer, texture, matrix, 1);
/* This lets the client know that we've displayed that frame and it can
* prepare another one now if it likes. */
wlr_surface_send_frame_done(surface, rdata->when);
static void output_frame(struct wl_listener *listener, void *data) {
/* This function is called every time an output is ready to display a frame,
* generally at the output's refresh rate (e.g. 60Hz). */
struct tinywl_output *output =
wl_container_of(listener, output, frame);
struct wlr_renderer *renderer = output->server->renderer;
struct timespec now;
clock_gettime(CLOCK_MONOTONIC, &now);
/* wlr_output_make_current makes the OpenGL context current. */
if (!wlr_output_make_current(output->wlr_output, NULL)) {
/* The "effective" resolution can change if you rotate your outputs. */
int width, height;
wlr_output_effective_resolution(output->wlr_output, &width, &height);
/* Begin the renderer (calls glViewport and some other GL sanity checks) */
wlr_renderer_begin(renderer, width, height);
float color[4] = {0.3, 0.3, 0.3, 1.0};
wlr_renderer_clear(renderer, color);
/* Each subsequent window we render is rendered on top of the last. Because
* our view list is ordered front-to-back, we iterate over it backwards. */
struct tinywl_view *view;
wl_list_for_each_reverse(view, &output->server->views, link) {
if (!view->mapped) {
/* An unmapped view should not be rendered. */
struct render_data rdata = {
.output = output->wlr_output,
.view = view,
.renderer = renderer,
.when = &now,
/* This calls our render_surface function for each surface among the
* xdg_surface's toplevel and popups. */
render_surface, &rdata);
/* Conclude rendering and swap the buffers, showing the final frame
* on-screen. */
wlr_output_swap_buffers(output->wlr_output, NULL, NULL);
static void server_new_output(struct wl_listener *listener, void *data) {
/* This event is rasied by the backend when a new output (aka a display or
* monitor) becomes available. */
struct tinywl_server *server =
wl_container_of(listener, server, new_output);
struct wlr_output *wlr_output = data;
/* Some backends don't have modes. DRM+KMS does, and we need to set a mode
* before we can use the output. The mode is a tuple of (width, height,
* refresh rate), and each monitor supports only a specific set of modes. We
* just pick the first, a more sophisticated compositor would let the user
* configure it or pick the mode the display advertises as preferred. */
if (!wl_list_empty(&wlr_output->modes)) {
struct wlr_output_mode *mode =
wl_container_of(wlr_output->modes.prev, mode, link);
wlr_output_set_mode(wlr_output, mode);
/* Allocates and configures our state for this output */
struct tinywl_output *output =
calloc(1, sizeof(struct tinywl_output));
output->wlr_output = wlr_output;
output->server = server;
/* Sets up a listener for the frame notify event. */
output->frame.notify = output_frame;
wl_signal_add(&wlr_output->events.frame, &output->frame);
wl_list_insert(&server->outputs, &output->link);
/* Adds this to the output layout. The add_auto function arranges outputs
* from left-to-right in the order they appear. A more sophisticated
* compositor would let the user configure the arrangement of outputs in the
* layout. */
wlr_output_layout_add_auto(server->output_layout, wlr_output);
/* Creating the global adds a wl_output global to the display, which Wayland
* clients can see to find out information about the output (such as
* DPI, scale factor, manufacturer, etc). */
static void xdg_surface_map(struct wl_listener *listener, void *data) {
/* Called when the surface is mapped, or ready to display on-screen. */
struct tinywl_view *view = wl_container_of(listener, view, map);
view->mapped = true;
focus_view(view, view->xdg_surface->surface);
static void xdg_surface_unmap(struct wl_listener *listener, void *data) {
/* Called when the surface is unmapped, and should no longer be shown. */
struct tinywl_view *view = wl_container_of(listener, view, unmap);
view->mapped = false;
static void xdg_surface_destroy(struct wl_listener *listener, void *data) {
/* Called when the surface is destroyed and should never be shown again. */
struct tinywl_view *view = wl_container_of(listener, view, destroy);
static void begin_interactive(struct tinywl_view *view,
enum tinywl_cursor_mode mode, uint32_t edges) {
/* This function sets up an interactive move or resize operation, where the
* compositor stops propegating pointer events to clients and instead
* consumes them itself, to move or resize windows. */
struct tinywl_server *server = view->server;
struct wlr_surface *focused_surface =
if (view->xdg_surface->surface != focused_surface) {
/* Deny move/resize requests from unfocused clients. */
server->grabbed_view = view;
server->cursor_mode = mode;
struct wlr_box geo_box;
wlr_xdg_surface_get_geometry(view->xdg_surface, &geo_box);
if (mode == TINYWL_CURSOR_MOVE) {
server->grab_x = server->cursor->x - view->x;
server->grab_y = server->cursor->y - view->y;
} else {
server->grab_x = server->cursor->x + geo_box.x;
server->grab_y = server->cursor->y + geo_box.y;
server->grab_width = geo_box.width;
server->grab_height = geo_box.height;
server->resize_edges = edges;
static void xdg_toplevel_request_move(
struct wl_listener *listener, void *data) {
/* This event is raised when a client would like to begin an interactive
* move, typically because the user clicked on their client-side
* decorations. Note that a more sophisticated compositor should check the
* provied serial against a list of button press serials sent to this
* client, to prevent the client from requesting this whenver they want. */
struct tinywl_view *view = wl_container_of(listener, view, request_move);
begin_interactive(view, TINYWL_CURSOR_MOVE, 0);
static void xdg_toplevel_request_resize(
struct wl_listener *listener, void *data) {
/* This event is raised when a client would like to begin an interactive
* resize, typically because the user clicked on their client-side
* decorations. Note that a more sophisticated compositor should check the
* provied serial against a list of button press serials sent to this
* client, to prevent the client from requesting this whenver they want. */
struct wlr_xdg_toplevel_resize_event *event = data;
struct tinywl_view *view = wl_container_of(listener, view, request_resize);
begin_interactive(view, TINYWL_CURSOR_RESIZE, event->edges);
static void server_new_xdg_surface(struct wl_listener *listener, void *data) {
/* This event is raised when wlr_xdg_shell receives a new xdg surface from a
* client, either a toplevel (application window) or popup. */
struct tinywl_server *server =
wl_container_of(listener, server, new_xdg_surface);
struct wlr_xdg_surface *xdg_surface = data;
if (xdg_surface->role != WLR_XDG_SURFACE_ROLE_TOPLEVEL) {
/* Allocate a tinywl_view for this surface */
struct tinywl_view *view =
calloc(1, sizeof(struct tinywl_view));
view->server = server;
view->xdg_surface = xdg_surface;
/* Listen to the various events it can emit */
view->map.notify = xdg_surface_map;
wl_signal_add(&xdg_surface->, &view->map);
view->unmap.notify = xdg_surface_unmap;
wl_signal_add(&xdg_surface->events.unmap, &view->unmap);
view->destroy.notify = xdg_surface_destroy;
wl_signal_add(&xdg_surface->events.destroy, &view->destroy);
/* cotd */
struct wlr_xdg_toplevel *toplevel = xdg_surface->toplevel;
view->request_move.notify = xdg_toplevel_request_move;
wl_signal_add(&toplevel->events.request_move, &view->request_move);
view->request_resize.notify = xdg_toplevel_request_resize;
wl_signal_add(&toplevel->events.request_resize, &view->request_resize);
/* Add it to the list of views. */
wl_list_insert(&server->views, &view->link);
int main(int argc, char *argv[]) {
wlr_log_init(WLR_DEBUG, NULL);
char *startup_cmd = NULL;
int c;
while ((c = getopt(argc, argv, "s:h")) != -1) {
switch (c) {
case 's':
startup_cmd = optarg;
printf("Usage: %s [-s startup command]\n", argv[0]);
return 0;
if (optind < argc) {
printf("Usage: %s [-s startup command]\n", argv[0]);
return 0;
struct tinywl_server server;
/* The Wayland display is managed by libwayland. It handles accepting
* clients from the Unix socket, manging Wayland globals, and so on. */
server.wl_display = wl_display_create();
/* The backend is a wlroots feature which abstracts the underlying input and
* output hardware. The autocreate option will choose the most suitable
* backend based on the current environment, such as opening an X11 window
* if an X11 server is running. The NULL argument here optionally allows you
* to pass in a custom renderer if wlr_renderer doesn't meet your needs. The
* backend uses the renderer, for example, to fall back to software cursors
* if the backend does not support hardware cursors (some older GPUs
* don't). */
server.backend = wlr_backend_autocreate(server.wl_display, NULL);
/* If we don't provide a renderer, autocreate makes a GLES2 renderer for us.
* The renderer is responsible for defining the various pixel formats it
* supports for shared memory, this configures that for clients. */
server.renderer = wlr_backend_get_renderer(server.backend);
wlr_renderer_init_wl_display(server.renderer, server.wl_display);
/* This creates some hands-off wlroots interfaces. The compositor is
* necessary for clients to allocate surfaces, dmabuf allows them to use
* opaque GPU handles for buffers to avoid copying pixels on the CPU, and
* the data device manager handles the clipboard. Each of these wlroots
* interfaces has room for you to dig your fingers in and play with their
* behavior if you want. */
wlr_compositor_create(server.wl_display, server.renderer);
wlr_linux_dmabuf_v1_create(server.wl_display, server.renderer);
/* Creates an output layout, which a wlroots utility for working with an
* arrangement of screens in a physical layout. */
server.output_layout = wlr_output_layout_create();
/* Configure a listener to be notified when new outputs are available on the
* backend. */
server.new_output.notify = server_new_output;
wl_signal_add(&server.backend->events.new_output, &server.new_output);
/* Set up our list of views and the xdg-shell. The xdg-shell is a Wayland
* protocol which is used for application windows. For more detail on
* shells, refer to my article:
server.xdg_shell = wlr_xdg_shell_create(server.wl_display);
server.new_xdg_surface.notify = server_new_xdg_surface;
* Creates a cursor, which is a wlroots utility for tracking the cursor
* image shown on screen.
server.cursor = wlr_cursor_create();
wlr_cursor_attach_output_layout(server.cursor, server.output_layout);
/* Creates an xcursor manager, another wlroots utility which loads up
* Xcursor themes to source cursor images from and makes sure that cursor
* images are available at all scale factors on the screen (necessary for
* HiDPI support). We add a cursor theme at scale factor 1 to begin with. */
server.cursor_mgr = wlr_xcursor_manager_create(NULL, 24);
wlr_xcursor_manager_load(server.cursor_mgr, 1);
* wlr_cursor *only* displays an image on screen. It does not move around
* when the pointer moves. However, we can attach input devices to it, and
* it will generate aggregate events for all of them. In these events, we
* can choose how we want to process them, forwarding them to clients and
* moving the cursor around. More detail on this process is described in my
* input handling blog post:
* And more comments are sprinkled throughout the notify functions above.
server.cursor_motion.notify = server_cursor_motion;
wl_signal_add(&server.cursor->events.motion, &server.cursor_motion);
server.cursor_motion_absolute.notify = server_cursor_motion_absolute;
server.cursor_button.notify = server_cursor_button;
wl_signal_add(&server.cursor->events.button, &server.cursor_button);
server.cursor_axis.notify = server_cursor_axis;
wl_signal_add(&server.cursor->events.axis, &server.cursor_axis);
* Configures a seat, which is a single "seat" at which a user sits and
* operates the computer. This conceptually includes up to one keyboard,
* pointer, touch, and drawing tablet device. We also rig up a listener to
* let us know when new input devices are available on the backend.
server.new_input.notify = server_new_input;
wl_signal_add(&server.backend->events.new_input, &server.new_input); = wlr_seat_create(server.wl_display, "seat0");
server.request_cursor.notify = seat_request_cursor;
/* Add a Unix socket to the Wayland display. */
const char *socket = wl_display_add_socket_auto(server.wl_display);
if (!socket) {
return 1;
/* Start the backend. This will enumerate outputs and inputs, become the DRM
* master, etc */
if (!wlr_backend_start(server.backend)) {
return 1;
/* Set the WAYLAND_DISPLAY environment variable to our socket and run the
* startup command if requested. */
setenv("WAYLAND_DISPLAY", socket, true);
if (startup_cmd) {
if (fork() == 0) {
execl("/bin/sh", "/bin/sh", "-c", startup_cmd, (void *)NULL);
/* Run the Wayland event loop. This does not return until you exit the
* compositor. Starting the backend rigged up all of the necessary event
* loop configuration to listen to libinput events, DRM events, generate
* frame events at the refresh rate, and so on. */
/* Once wl_display_run returns, we shut down the server. */
return 0;
Copy link

maximaximal commented Jul 22, 2019

Very nice example. To run it on the newest version, the code requires two small changes because of commit output: refactor frame submission API . After this, the compositor is working again. Diff:

--- original.c  2019-07-22 11:09:29.823850525 +0200
+++ tinywl.c    2019-07-22 11:10:12.233850968 +0200
@@ -666,7 +666,7 @@
        clock_gettime(CLOCK_MONOTONIC, &now);
        /* wlr_output_make_current makes the OpenGL context current. */
-       if (!wlr_output_make_current(output->wlr_output, NULL)) {
+       if (!wlr_output_attach_render(output->wlr_output, NULL)) {
        /* The "effective" resolution can change if you rotate your outputs. */
@@ -701,7 +701,7 @@
        /* Conclude rendering and swap the buffers, showing the final frame
         * on-screen. */
-       wlr_output_swap_buffers(output->wlr_output, NULL, NULL);
+       wlr_output_commit(output->wlr_output);

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment