Below are some basic guidelines on coding style, source file layout and merge requests that hopefully makes it easier for you to land your code.
For build instructions see the README.md
For internal API documentation as well as notes for application developers see here.
Before filing a pull request run the tests:
meson test -C _build --print-errorlogs
Use descriptive commit messages, see
https://wiki.gnome.org/Git/CommitMessages
and check
https://wiki.openstack.org/wiki/GitCommitMessages
for good examples. The commits in a merge request should have "recipe" style history rather than being a work log. See here for an explanation of the difference. The advantage is that the code stays bisectable and individual bits can be cherry-picked or reverted.
When submitting a merge request consider checking these first:
- Does the code use the below coding patterns?
- Is the commit history in recipe style (see above)?
- Do the commit messages reference the bugs they fix. If so,
- Use
Helps:
if the commit partially addresses a bug or contributes to a solution. - UseCloses:
if the commit fully resolves the bug. This allows the release script to detect & mention it inNEWS
file. - Does the code crash or introduce new
CRITICAL
orWARNING
messages in the log or when run form the console. If so, fix these first? - Is the new code covered by any tests? This could be a unit test, an added screenshot test, a tool to exercise new DBus API (see e.g. tools/check-mount-operation.
- Are property assignments to default values removed from UI files? (See
gtk-builder-tool simplify file.ui
)
If any of the above criteria aren't met yet it's still fine (and encouraged) to open a merge request marked as draft. Please indicate why you consider it draft in this case. As Phosh is used on a wide range of devices and distributions please indicate in what scenarios you tested your code.
We're mostly using libhandy's Coding Style.
These are the differences:
- We're not picky about GTK style function argument indentation, that is having multiple arguments on one line is also o.k.
- For signal handler callbacks we prefer for the
on_<signalname>
naming pattern e.g.on_clicked
since this helps to keep the namespace clean. See below. - Since we're not a library we usually use
G_DEFINE_TYPE
instead ofG_DEFINE_TYPE_WITH_PRIVATE
(except when we need a deriveable type) since it makes the rest of the code more compact.
We use one file per GObject. It should be named like the GObject without
the phosh prefix, lowercase and '_' replaced by '-'. So a hypothetical
PhoshThing
would go to src/thing.c
. If there are likely name
clashes add the phosh-
prefix (e.g. phosh-wayland.c
). The
individual C files should be structured as (top to bottom of file):
-
License boilerplate
/* * Copyright (C) year copyright holder * * SPDX-License-Identifier: GPL-3-or-later * Author: you <[email protected]> */
-
A log domain, usually the filename with
phosh-
prefix#define G_LOG_DOMAIN "phosh-thing"
-
#include
s: Phosh ones go first, then glib/gtk, then generic C headers. These blocks are separated by newline and each sorted alphabetically:#define G_LOG_DOMAIN "phosh-settings" #include "phosh-config.h" #include "settings.h" #include "shell.h" #include <gio/gdesktopappinfo.h> #include <glib/glib.h> #include <math.h>
This helps to detect missing headers in includes.
-
docstring: If you have trouble to describe the class concisely, then it might be an indication that it should be split into multiple classes.
/** * PhoshYourThing: * * Short, single line, summary * * A longer description with details that can be * multiline. */
-
property enum
enum { PROP_0, PROP_FOO, PROP_BAR, LAST_PROP }; static GParamSpec *props[LAST_PROP];
-
signal enum
enum { FOO_HAPPENED, BAR_TRIGGERED, N_SIGNALS }; static guint signals[N_SIGNALS];
-
type definitions
typedef struct _PhoshThing { GObject parent; ... } PhoshThing; G_DEFINE_TYPE (PhoshThing, phosh_thing, G_TYPE_OBJECT)
-
private methods and callbacks (these can also go at convenient places above
phosh_thing_constructed ()
) -
phosh_thing_set_property ()
. Set properties. If setting a property requires more than a single line prefer adding a setter method, e.g. for thefoo
property, the setter method would bephosh_thing_set_foo ()
. This is almost always the case as we preferG_PARAM_EXPLICIT_NOTIFY
(see below).static void phosh_thing_property (GObject *object, guint property_id, const GValue *value, GParamSpec *pspec) { PhoshThing *self = PHOSH_TING (object); switch (property_id) { case PROP_FOO: phosh_thing_set_foo (self, g_value_get_string (value)); break; … default: G_OBJECT_WARN_INVALID_PROPERTY_ID (object, property_id, pspec); break; } }
-
phosh_thing_get_property ()
-
phosh_thing_constructed ()
: Finish object constructions. Usually only needed if you need the values of multiple properties passed at object construction time. -
phosh_thing_dispose ()
: Usually only needed when you need to break reference cycles. Otherwise preferfinalize
. Asdispose
can be run multiple times useg_clear_*
to avoid freeing resources multiple times:static void phosh_thing_dispose (GObject *object) { PhoshThing *self = PHOSH_THING (object); g_cancellable_cancel (self->cancel); g_clear_object (&self->cancel); g_clear_object (&self->bar); g_clear_pointer (&self->foo, g_free); G_OBJECT_CLASS (phosh_thing_parent_class)->dispose (object); }
-
phosh_thing_finalize ()
: Free allocated resources. -
phosh_thing_class_init ()
: Define properties and signals. For widget templates bind child widgets and signal handlers. -
phosh_thing_init ()
: Initialize defaults for member variables here. -
phosh_thing_new ()
: A convenience wrapper aroundg_object_new ()
. Don't do further object initialization here but rather do that inphosh_thing_init ()
,phosh_thing_constructed ()
or individual property setters. This ensures that objects can be constructed either via this constructor org_object_new ()
. -
Public methods, all starting with the object name (i.e.
phosh_thing_
).
The reason public methods go at the bottom is that they have declarations in the header file and can thus be referenced from anywhere else in the source file.
For custom widgets set the css name using gtk_widget_class_set_css_name ()
.
There's no need set an (additional) style class in the ui file.
Good:
static void
phosh_lockscreen_class_init (PhoshLockscreenClass *klass)
{
…
gtk_widget_class_set_css_name (widget_class, "phosh-lockscreen");
…
}
Bad:
<template class="PhoshLockscreen" parent="…">
…
<style>
<class name="phosh-lockscreen"/>
</style>
…
</template>
Except for G_CONSTRUCT_ONLY
properties use G_PARAM_EXPLICIT_NOTIFY
and notify
about property changes only when the underlying variable changes value:
static void
on_present_changed (PhoshDockedInfo *self)
{
…
if (self->enabled == enabled)
return;
self->enabled = enabled;
g_object_notify_by_pspec (G_OBJECT (self), props[PROP_ENABLED]);
}
static void
phosh_docked_info_class_init (PhoshDockedInfoClass *klass)
{
…
/**
* PhoshDockedInfo:enabled:
*
* Whether docked mode is enabled
*/
props[PROP_PRESENT] =
g_param_spec_boolean ("present", "", "",
G_PARAM_READABLE |
G_PARAM_STATIC_STRINGS |
G_PARAM_EXPLICIT_NOTIFY);
…
}
This makes sure we minimize the notifications on changed property values.
Prefer a docstring over filling in the properties' nick
and blurb
. See
example above.
If the state of a property depends on the state of another one prefer
g_object_bind_property ()
to keep these in sync:
Good:
g_object_bind_property (self, "bar",
self->avatar, "loadable-icon",
G_BINDING_DEFAULT | G_BINDING_SYNC_CREATE);
This has the upside that the binding goes away automatically when either of the two objects gets disposed and also avoids bugs in manual syncing code.
For widgets you can construct the binding via the UI XML:
<object class="PhoshQuickSetting" id="foo_quick_setting">
<property name="sensitive" bind-source="wwaninfo" bind-property="present" bind-flags="sync-create"/>
…
</object>
There's callbacks for signals, async functions, and actions. We
usually have them all start with on_
to make it easy to spot
that these aren't just methods (and hence have restrictions regarding
their function arguments and return values).
-
Signal handlers are named
on_<signalname>
. E.g. the handler for aGListModel
sitems-changed
signal would be namedon_items_changed ()
. To avoid ambiguity one can add the emitter's name (on_devices_list_box_items_changed ()
). -
For
notify::
signal handler callbacks acting on property changes we use theon_<property>_changed ()
naming. E.g.on_nm_signal_strength_changed ()
. -
For
GAsyncReadyCallback
s we useon_<async-function>_ready()
, e.g. the callback fornm_client_new_async
that invokesnm_client_new_finish
would be namedon_nm_client_new_ready ()
. -
For actions we use
on_<action>_activated
, e.g. for ago-back
action we'd useon_go_back_activated ()
. -
In cases where the signal handler name should express what it does rather than what signal it connects to, we use a
_cb
suffix. This is often the case when we want to use the same signal handler to handle multiple signals. E.g. a callback that updates aGtkStackPage
when a signal happens would be namedupdate_stack_page_cb ()
.
Public (non static) functions must check the input arguments at the
top of the function. This makes it easy to reuse them in other parts
and makes API misuse easy to debug via G_DEBUG=fatal-criticals
. You
usually want to check argument types and if the arguments fulfill the
requirements (e.g. if they need to be non-NULL).
Good:
void
phosh_foo_set_name (PhoshFoo *self, const char *name)
{
GtkWidget *somewidget;
g_return_if_fail (PHOSH_IS_FOO (self));
g_return_if_fail (!name);
…
}