Contributing just involves sending a pull request. You will probably be more successful with your contribution if you visit #sway-devel on Libera Chat upfront and discuss your plans.
Note: rules are made to be broken. Adjust or ignore any/all of these as you see fit, but be prepared to justify it to your peers.
If you already have your own pull request habits, feel free to use them. If you don't, however, allow me to make a suggestion: feature branches pulled from upstream. Try this:
- Fork wlroots
git clone [email protected]:<username>/wlroots.git && cd wlrootsgit remote add upstream https://github.com/swaywm/wlroots
You only need to do this once. You're never going to use your fork's master branch. Instead, when you start working on a feature, do this:
git fetch upstreamgit checkout -b add-so-and-so-feature upstream/master- Add and commit your changes
git push -u origin add-so-and-so-feature- Make a pull request from your feature branch
When you submit your pull request, your commit log should do most of the talking when it comes to describing your changes and their motivation. In addition to this, your pull request's comments will ideally include a test plan that the reviewers can use to (1) demonstrate the problem on master, if applicable and (2) verify that the problem no longer exists with your changes applied (or that your new features work correctly). Document all of the edge cases you're aware of so we can adequately test them - then verify the test plan yourself before submitting.
Unlike many projects using GitHub and GitLab, wlroots has a linear, "recipe" style history. This means that every commit should be small, digestible, stand-alone, and functional. Rather than a purely chronological commit history like this:
doc: final docs for view transforms
fix tests when disabled, redo broken doc formatting
better transformed-view iteration (thanks Hannah!)
try to catch more cases in tests
tests: add new spline test
fix compilation on splines
doc: notes on reticulating splines
compositor: add spline reticulation for view transforms
We aim to have a clean history which only reflects the final state, broken up into functional groupings:
compositor: add spline reticulation for view transforms
compositor: new iterator for view transforms
tests: add view-transform correctness tests
doc: fix formatting for view transforms
This ensures that the final patch series only contains the final state, without the changes and missteps taken along the development process. A linear history eases reviewing, cherry-picking and reverting changes.
If you aren't comfortable with manipulating the Git history, have a look at git-rebase.io.
Please strive to write good commit messages. Here's some guidelines to follow:
The first line should be limited to 50 characters and should be a sentence that completes the thought [When applied, this commit will...] "Implement cmd_move" or "Improve performance of arrange_windows on ARM" or similar.
The subsequent lines should be separated from the subject line by a single
blank line, and include optional details. In this you can give justification
for the change, reference Github issues, or explain some of the subtler
details of your patch. This is important because when someone finds a line of
code they don't understand later, they can use the git blame command to find
out what the author was thinking when they wrote it. It's also easier to review
your pull requests if they're separated into logical commits that have good
commit messages and justify themselves in the extended commit description.
As a good rule of thumb, anything you might put into the pull request description on Github is probably fair game for going into the extended commit message as well.
See How to Write a Git Commit Message for more details.
When your changes are submitted for review, one or more core committers will look over them. Smaller changes might be merged with little fanfare, but larger changes will typically see review from several people. Be prepared to receive some feedback - you may be asked to make changes to your work. Our code review process is:
- Triage the pull request. Do the commit messages make sense? Is a test plan necessary and/or present? Add anyone as reviewers that you think should be there (using the relevant GitHub feature, if you have the permissions, or with an @mention if necessary).
- Review the code. Look for code style violations, naming convention violations, buffer overflows, memory leaks, logic errors, non-portable code (including GNU-isms), etc. For significant changes to the public API, loop in a couple more people for discussion.
- Execute the test plan, if present.
- Merge the pull request when all reviewers approve.
- File follow-up tickets if appropriate.
wlroots is written in C with a style similar to the kernel style, but with a few notable differences.
Try to keep your code conforming to C11 and POSIX as much as possible, and do not use GNU extensions.
Brackets always go on the same line, including in functions. Always include brackets for if/while/for, even if it's a single statement.
void function(void) {
if (condition1) {
do_thing1();
}
if (condition2) {
do_thing2();
} else {
do_thing3();
}
}Indentations are a single tab.
For long lines that need to be broken, the continuation line should be indented with an additional tab. If the line being broken is opening a new block (functions, if, while, etc.), the continuation line should be indented with two tabs, so they can't be misread as being part of the block.
really_long_function(argument1, argument2, ...,
argument3, argument4);
if (condition1 && condition2 && ...
condition3 && condition4) {
do_thing();
}Try to break the line in the place which you think is the most appropriate.
Try to keep your lines under 80 columns, but you can go up to 100 if it improves readability. Don't break lines indiscriminately, try to find nice breaking points so your code is easy to read.
Global function and type names should be prefixed with wlr_submodule_ (e.g.
struct wlr_output, wlr_output_set_cursor). For static functions and
types local to a file, the names chosen aren't as important. Local function
names shouldn't have a wlr_ prefix.
For include guards, use the header's filename relative to include. Uppercase all of the characters, and replace any invalid characters with an underscore.
For functions that are responsible for constructing and destructing an object, they should be written as a pair of one of two forms:
init/finish: These initialize/deinitialize a type, but are NOT responsible for allocating it. They should accept a pointer to some pre-allocated memory (e.g. a member of a struct).create/destroy: These also initialize/deinitialize, but will return a pointer to amalloced chunk of memory, and willfreeit indestroy.
A destruction function should always be able to accept a NULL pointer or a zeroed value and exit cleanly; this simplifies error handling a lot.
For functions not returning a value, they should return a (stdbool.h) bool to indicate whether they succeeded or not.
Try to keep the use of macros to a minimum, especially if a function can do the
job. If you do need to use them, try to keep them close to where they're being
used and #undef them after.
struct wlr_backend *wlr_backend_autocreate(struct wl_display *display) {
struct wlr_backend *backend;
if (getenv("WAYLAND_DISPLAY") || getenv("_WAYLAND_DISPLAY")) {
backend = attempt_wl_backend(display);
if (backend) {
return backend;
}
}
const char *x11_display = getenv("DISPLAY");
if (x11_display) {
return wlr_x11_backend_create(display, x11_display);
}
// Attempt DRM+libinput
struct wlr_session *session = wlr_session_create(display);
if (!session) {
wlr_log(WLR_ERROR, "Failed to start a DRM session");
return NULL;
}
int gpu = wlr_session_find_gpu(session);
if (gpu == -1) {
wlr_log(WLR_ERROR, "Failed to open DRM device");
goto error_session;
}
backend = wlr_multi_backend_create(session);
if (!backend) {
goto error_gpu;
}
struct wlr_backend *libinput = wlr_libinput_backend_create(display, session);
if (!libinput) {
goto error_multi;
}
struct wlr_backend *drm = wlr_drm_backend_create(display, session, gpu);
if (!drm) {
goto error_libinput;
}
wlr_multi_backend_add(backend, libinput);
wlr_multi_backend_add(backend, drm);
return backend;
error_libinput:
wlr_backend_destroy(libinput);
error_multi:
wlr_backend_destroy(backend);
error_gpu:
wlr_session_close_file(session, gpu);
error_session:
wlr_session_destroy(session);
return NULL;
}Each protocol generally lives in a file with the same name, usually containing
at least one struct for each interface in the protocol. For instance,
xdg_shell lives in types/wlr_xdg_shell.h and has a wlr_xdg_surface struct.
Global interfaces generally have public constructors and destructors. Their
struct has a field holding the wl_global itself, a destroy signal and a
wl_display destroy listener. Example:
struct wlr_compositor {
struct wl_global *global;
…
struct wl_listener display_destroy;
struct {
struct wl_signal new_surface;
struct wl_signal destroy;
} events;
};When the destructor is called, it should emit the destroy signal, remove the
display destroy listener, destroy the wl_global and then destroy the struct.
The destructor can assume all clients and resources have been already
destroyed.
Resources are the representation of Wayland objects on the compositor side. They
generally have an associated struct, called the object struct, stored in their
user_data field.
Object structs can be retrieved from resources via wl_resource_get_data. To
prevent bad casts, a safe helper function checking the type of the resource is
used:
static const struct wl_surface_interface surface_impl;
struct wlr_surface *wlr_surface_from_resource(struct wl_resource *resource) {
assert(wl_resource_instance_of(resource, &wl_surface_interface,
&surface_impl));
return wl_resource_get_user_data(resource);
}If a pointer to a wl_resource is stored, a resource destroy handler needs to
be registered to clean it up. libwayland will automatically destroy resources
in an arbitrary order when a client is disconnected, the compositor must handle
this correctly.
Object structs should only be destroyed when their resource is destroyed, ie.
in the resource destroy handler (set with wl_resource_set_implementation).
- If the object has a destructor request: the request handler should just call
wl_resource_destroyand do nothing else. The compositor must not destroy resources on its own outside the destructor request handler. - If the protocol specifies that an object is destroyed when an event is sent:
it's the only case where the compositor is allowed to send the event and then
call
wl_resource_destroy. An example of this iswl_callback.
Some resources can become inert in situations described in the protocol or when the compositor decides to get rid of them. All requests made to inert resources should be ignored, except the destructor. This is achieved by:
- When the resource becomes inert: destroy the object struct and call
wl_resource_set_user_data(resource, NULL). Do not destroy the resource. - For each request made to a resource that can be inert: add a NULL check to ignore the request if the resource is inert.
- When the client calls the destructor request on the resource: call
wl_resource_destroy(resource)as usual. - When the resource is destroyed, if the resource isn't inert, destroy the object struct.
Example:
// Handles the destroy request
static void subsurface_handle_destroy(struct wl_client *client,
struct wl_resource *resource) {
wl_resource_destroy(resource);
}
// Handles a regular request
static void subsurface_set_position(struct wl_client *client,
struct wl_resource *resource, int32_t x, int32_t y) {
struct wlr_subsurface *subsurface = subsurface_from_resource(resource);
if (subsurface == NULL) {
return;
}
…
}
// Destroys the wlr_subsurface struct
static void subsurface_destroy(struct wlr_subsurface *subsurface) {
if (subsurface == NULL) {
return;
}
…
wl_resource_set_user_data(subsurface->resource, NULL);
free(subsurface);
}
// Resource destroy listener
static void subsurface_handle_resource_destroy(struct wl_resource *resource) {
struct wlr_subsurface *subsurface = subsurface_from_resource(resource);
subsurface_destroy(subsurface);
}
// Makes the resource inert
static void subsurface_handle_surface_destroy(struct wl_listener *listener,
void *data) {
struct wlr_subsurface *subsurface =
wl_container_of(listener, subsurface, surface_destroy);
subsurface_destroy(subsurface);
}