Blob Blame History Raw

All code should follow the same formatting standards which are broadly based on the GNU style ( with some
additions. Briefly:

 - Tab indents are used and braces for blocks go on the same line as the block statement:

	if (x < foo (y, z)) {
		haha = bar[4] + 5;
	} else {
		while (z) {
			haha += foo (z, z);
		return abc (haha);

   Braces should be omitted for single-line blocks, but included for all blocks in a multi-part if statement which has blocks containing more than
   one line (as above).

 - Spaces should be present between function name and argument block, and after commas:

	foo (z, z)

 - In pointer types, the '*' is grouped with the variable name, not with the base type. 

	int *a;


	int* a;

   In cases where there is no variable name, for instance, return values, there should be a single space between the base type and the '*'.

   Type casts should have no space between the type and '*', but a space before the variable being cast:

	(gchar*) foobar;
	(gchar**) &foobar;

 - Function and variable names are lower_case_with_underscores, type names are CamelCase and macro names are UPPER_CASE_WITH_UNDERSCORES.

 - Comparisons to NULL, TRUE and FALSE should always be made explicit, for clarity.

 - Code should be wrapped at the 150th column, such that it doesn't require horizontal scrolling on a decent-sized display.
   Don't wrap at the 80th column.

Documentation comments

All public API functions should have inline documentation headers in the gtk-doc style. For more information about gtk-doc comments, see the gtk-doc
manual ( There are a few conventions above and beyond the standard gtk-doc formatting which
libgdata employs:

 - For API which returns allocated memory, the relevant free/unref function must be mentioned in the "Return value" part of the documentation:

	* Return value: a new #GDataEntry; unref with g_object_unref()

   If the function can also return NULL (on error, for example) or some other "default" value (-1, 0, etc.), format it as follows:

	* Return value: a new #GDataGDFeedLink, or %NULL; free with gdata_gd_feed_link_free()

   Note that if a function returns NULL as a result of a precondition or assertion failure, this should not be listed in the documentation. The
   precondition itself may be listed in the function documentation prose, but if it's the only reason for a function to return NULL, the "or %NULL"
   clause should be omitted.

 - When adding API, make sure to add a "Since" clause:

	* Since: 0.2.0

 - For object methods, the "self" parameter should be documented simply as "a #GObjectType":

	* @self: a #GDataQuery

 - For function parameters which can legitimately be set to NULL (or some other default value), list that as follows:

	* @updated_max: the new maximum update time, or %NULL

 - If numbers, such as -1, are mentioned in documentation as values to be passed around in code, they should be wrapped in a DocBook "code" tag
   (e.g. "<code class="literal">-1</code>"), so that they appear similarly to NULL in the documentation.

 - The documentation explaining the purpose of a property, its limitations, interpretation, etc., should be given in the gtk-doc comment for the
   GObject property itself, not in the documentation for its getter or setter. The documentation for the getter and setter should be stubs which
   refer to the property's documentation. The getter and setter documentation should, however, still include full details about whether NULL values
   are permissible for the function parameters, or are possible as the return value, for example.

Adding public API

 - Ensure it has proper guards against bad parameters:

	g_return_if_fail (GDATA_IS_QUERY (self));
	g_return_if_fail (foobar != NULL);

 - All public API must have a gtk-doc comment, and be added to the docs/reference/gdata-sections.txt file, to include it in the documentation.
   The documentation comment must have a "Since" clause (see "Documentation comments" section).

 - All public API must be listed in gdata/gdata-*.symbols.

 - Non-trivial API should have a test case added in the relevant test suite file in gdata/tests. Note that the "general" test suite file cannot make
   network requests in the course of running its test cases.

 - All GObject properties must have getter/setter functions.

 - All API which returns allocated memory must be tagged with G_GNUC_WARN_UNUSED_RESULT after its declaration, to safeguard against consumers of the
   API forgetting to use (and consequently free) the memory. This is unlikely, but adding the macro costs little and acts as a reminder in the API
   documentation to free the result.

 - All GObject *_get_type function declarations must be tagged with the G_GNUC_CONST macro, as well as any other applicable functions
   (see the gcc documentation:

 - All properties which could be considered to use an enumerated value should almost definitely use a string instead. See the documentation section
   on "Enumerable Properties" in the "GData Overview" section. Where appropriate, the values for these string properties should be made available
   as #defines.

 - New services should be implemented in libgdata itself, not by applications which use libgdata. See the documentation section on "New Services" in
   the "GData Overview" section.

 - New API must never be added in a stable micro release. API additions can only be made in a major or minor release; this is to prevent the LT version
   of one minor version's micro releases exceeding the LT version of the next minor version as almost happened between versions 0.6.3 and 0.7.0.
   See for information about libtool's versioning system. See also the
   “Versioning” section below.

 - Any async function which uses non-async-scope callbacks as well as the async ready callback should provide GDestroyNotify callbacks for destroying
   the user data for those callbacks. See for details.

 - Any new class’ class structure (e.g. GDataServiceClass) must have reserved slots to allow future API additions of virtual functions without
   breaking ABI.

Choosing function names

In general, use common sense. However, there are some specific cases where a standard is enforced:

 - For boolean getters (e.g. for FooBar:is-baz) use foo_bar_is_baz, rather than foo_bar_get_is_baz. Note that the property name should be "is-baz",
   rather than just "baz".

 - For boolean setters use foo_bar_set_is_baz, rather than foo_bar_set_baz.

 - In general, try to keep API named similarly to the actual GData API constructs. In certain cases, however, it might make more sense to rename pieces
   of API to be less cryptic (e.g. "timezone" instead of "ctz").

Deprecating public API

As libgdata moves towards API stability, old API should be deprecated rather than broken or removed entirely. The following should be ensured when
deprecating API:

 - G_GNUC_DEPRECATED_FOR is added to the API in the public header file.

 - A “Deprecated:” line is added to the gtk-doc comment for the API. This should mention what API replaces it, give a brief explanation of why the
   API has been deprecated, and finish with “(Since: [version].)” to list the version the API was deprecated.

 - “#ifndef LIBGDATA_DISABLE_DEPRECATED” should be wrapped around the API in the public header file.

 - All references to the API/uses of the API in libgdata code (including demos and the test suite) should be ported to use the replacement API instead.
   If this isn't possible, the deprecated function should be split into a static function which contains all the code, and the public symbol should
   become a simple wrapper of this static function. This allows the static function to be used inside libgdata without causing deprecation warnings.

 - Don't remove deprecated symbols from gdata-*.symbols.

 - Don't forget to also deprecate related symbols, such as the getter/setter for a property (or vice-versa).

Commit messages

libgdata does not use a ChangeLog; it is auto-generated from the git log when packaging a release. Commit messages should follow the GNOME commit
message guidelines (, with the exception that when a commit closes a bug, the short explanation of the commit
should simply be the bug's title, as copied from Bugzilla (e.g. "Bug 579885 – Add code examples to documentation"). The long explanation should then
be used to give details of the changes. If the bug's title is not relevant, it should be changed before committing the changes.

Unless the short explanation of a commit is a bug title, it should always be prefixed by a tag to describe the part of the library it touches, using
the following format "tag: Short explanation". The following tags are valid:

 - core: for the core code in the gdata directory, such as GDataEntry.

 - atom: for the Atom-namespaced code in the gdata/atom directory.

 - gcontact: for the gContact-namespaced code in the gdata/gcontact directory.

 - gd: for the GData-namespaced code in the gdata/gd directory.

 - media: for the Media RSS-namespaced code in the gdata/media directory.

 - build: for build changes and releases.

 - docs: for documentation changes which are not specific to a service, such as updates to the docs directory, NEWS, README, this file, etc.

 - tests: for changes to the test code in gdata/tests which are not specific to a service or namespace.

 - demos: for changes to the demo applications in the demos directory.

 - introspection: for introspection annotations and build changes.

 - calendar: for the Google Calendar code in gdata/services/calendar.

 - contacts: for the Google Contacts code in gdata/services/contacts.

 - documents: for the Google Documents code in gdata/services/documents.

 - picasaweb: for the PicasaWeb code in gdata/services/picasaweb.

 - youtube: for the YouTube code in gdata/services/youtube.

The only commits which should not have a tag are translation commits, touching only the po directory.


Starting with version 0.9.0, libgdata has adopted an even–odd/stable–unstable versioning policy, where odd minor version numbers are unstable releases,
released periodically (with increasing micro version numbers) and leading to a stable release with the next even minor version number. API breaks of
new API are allowed in micro releases with an odd minor version number, but not in micro releases with an even minor version number.

ABI and API backwards compatibility must be preserved unless the major version number is changed. libgdata is now ABI-stable.

It is encouraged to make a new micro release of an odd minor series after each large API addition or break.