I hesitated a lot whether we should just drop the whole localization of
plug-ins' label and description (blurb) within the core. Actually the
commit messages I wrote a few days ago were moving towards this logic.
It really looks to me like plug-in localization can happen fully within
plug-in themselves. As far as I can see, the only advantage which the
current logic has theoretically is that if we needed, we have access to
both the original strings and their translations (e.g. it could be
useful for text search). Nevertheless I am not sure if we will ever make
use of this, and this is limited cases as all filters turned GEGL ops
don't have such ability anyway.
Nevertheless since previous contributors clearly put quite a lot of work
on this code of localizing the plug-in's label and description within
the main binary, I want to give myself a little more time to think and
study the whole thing because doing anything rash.
In the meantime, what changes is that by default now, a plug-in without
a local gettext catalog is simply not localized. In particular, the core
process doesn't try to localize it using the default catalog, a.k.a.
GETTEXT_PACKAGE"-std-plug-ins" ("gimp30-std-plug-ins"). It just doesn't
make sense and the worst which could happen would be to get unexpected
and wrong translations.
Now by default, plug-ins will try to find a catalog in their main
folder, named as this folder. If it fails to find it, a message is
printed to stderr and localization is disabled (rather than falling back
to a default catalog). It is up to plug-in developers to either install
a catalog, or implement set_i18n() to give the right catalog, folder, or
disable localization with gettext, as handled by libgimp.
Clearly the old logic for localizing plug-ins was "core plug-ins first":
* In particular, we were defaulting to the "gimp*-std-plugins" domain,
asking plug-ins to override it with libgimp function
gimp_plug_in_set_translation_domain(). Obviously any third-party
plug-in would have to call this function since their strings are most
likely not the same as GIMP core plug-ins.
* Moreover this was only for the localization of menu items, which means
we had to duplicate work anyway (simplified for core C plug-ins only
with the INIT_I18N macro).
* Also plug-ins had to manage the location of the Gettext catalogs,
which is simpler for core plug-ins with gimp_locale_directory(), but
more annoying for third-party plug-ins which can be installed
basically anywhere (assuming their message catalogs are in their
directory, not centrally installed on the system, especially for
non-UNIX-like packages, with relocatable GIMP, no central package
system and so on).
* Finally in this logic of centrally installed catalogs, we were
requesting that the domain name is unique, which makes sense in a
Linux world with well maintained packages to avoid name clashes, not
in a world with people making plug-ins without knowing too much what's
done by their neighbour.
So now the new logic is:
- By default, GIMP will search for a folder called locale/ on the same
level as the plug-in executable. The Gettext domain will be the name
of the executable folder (and it doesn't matter too much if it's not
unique in such configuration).
- This can be disabled by overriding set_i18n() to NULL or
reimplementing it. All our core plug-ins will do this since they will
continue to use the centrally installed "gimp*-std-plugins" domain (or
"gimp*-python" for Python plug-ins).
- When not disabled while the folder is not available, warning messages
will be outputted to stderr, so that plug-in developers can easily
detect what is missing and how to handle it (and how to easily support
internationalization of their plug-in).
- gimp_plug_in_set_translation_domain() is removed and a future commit
is going to get rid of _gimp_plug_in_domain_register() because we are
going to get rid of the core-side localization of menu items (with
logic explained in the upcoming commit).
Fixing these 2 warnings in the CI which end up fatal:
WARNING: Invalid fragment for 'Gimp.Parasite': it should be struct
Serializes the object properties of @config to a [class@Parasite].
^~~~~~~~~~~~~~~~
WARNING: Invalid fragment for 'GLib.MainLoop': it should be struct
it has a GUI and is hanging around in a [class@GLib.MainLoop], it must call
^~~~~~~~~~~~~~~~~~~~~
gtk-doc has been slowly dying for the past few years; with gi-docgen we
have a nice successor.
This also makes sure the C documentation also uses the GIR file, which
in turn means faster build times (since all the C code doesn't have to
be parsed and recompiled again), and has a clear dependency graph.
See the [gi-docgen tutorial] for more info on how the system works.
[gi-docgen tutorial]: https://gnome.pages.gitlab.gnome.org/gi-docgen/tutorial.html
We heavily rely on GError in libgimp to retrieve plug-in error messages.
In a lot of our code, we just use domain=0 for g_set_error*() functions
and alike, but this is actually forbidden and results in GLib warnings.
Some plug-ins instead create their own domain, other use G_FILE_ERROR
nearly everywhere, even in some cases where the choice is really
questionable. Since anyway this is mostly useful for passing the error
message through, it is much nicer to provide a generic domain
GIMP_PLUG_IN_ERROR, which can be used by all plug-ins when they don't
want to bother with the error domain.
Don't drop references we do not own. Turns out bindings can have
things referenced even after all procedure code has returned. Keep the
old code there in #if 0 and keep the debug warning for now, maybe we
can do something generic about this.
Mention and guarantee the order of procedure registration (see
previous thumbnail loader commit). Remove duplicate docs from the
GimpPlugIn class comments that are needed for introspection, it's too
cumbersome to keep two identical texts in sync, and the removed text
is used nowhere.
Break reference cycles between the objects and the procedures they
keep by moving procedure destruction to dispose() and calling
g_object_run_dispose() before unrefing PLUG_IN and PDB in gimp.c.
Also some formatting and "Since: 3.0" annotation .
which frees exactly what _gimp_value_array_to_gp_params() has
allocated, honors its "full_copy" parameter, and plugs the last
libgimp refactoring leaks I'm currently aware of.
because we were simply destroying the proxy hashes when the last
procedure is done. Now we run gimp_plug_in_destroy_proxies() with the
right flag from gimp_plug_in_pop_procedure() before destroying the
hashes and get the proper debug warnings.
Need to push the current procedure around the entire body of
gimp_plug_in_proc_run_internal() because
_gimp_gp_params_to_value_array() needs access to proxy objects now.
Turn all ID param specs into object param specs (e.g. GimpParamImageID
becomes GimpParamImage) and convert between IDs and objects in
gimpgpparams.c directly above the the wire protocol, so all of app/,
libgimp/ and plug-ins/ can deal directly with objects down to the
lowest level and not care about IDs.
Use the actual object param specs for procedure arguments and return
values again instead of a plain g_param_spec_object() and bring back
the none_ok parameter.
This implies changing the PDB type checking functions to work on pure
integers instead of IDs (one can't check whether object creation is
possible if performing that check requires the object to already
exist).
For example gimp_foo_is_valid() becomes gimp_foo_id_is_valid() and is
not involved in automatic object creation magic at the protocol
level. Added wrappers which still say gimp_foo_is_valid() and take the
respective objects.
Adapted all code, and it all becomes nicer and less convoluted, even
the generated PDB wrappers in app/ and libgimp/.
Add an additional ref count to all proxies handed out by
GimpPlugIn, and when destroying the proxies, warn about any
proxy that has been unrefed, or additionally refed.
This catches and warns about only one broken unref, any more will
simply crash like before.
Turn GimpPlugIn into the main factory for all proxies and keep the
main hash tables there. The hash tables keep the initial reference.
For each GimpProcedure::run(), have s "sub-factory" which hands out
proxies to the actual procedure code. Each run() has hash tables of
its own which hold additional references. When run() is done, get rid
of its hash tables and their references, *and* drop the main plug-in
reference counts from the global hashes if the proxies' refcount has
dropped to one.
which are GimpProcedure subclasses with API to register as load/save
handlers and their own kind of run functions that get their standard
arguments passed directly instead of packed into a GimpValueArray.
They also register their standard arguments themselves, which removes
quite some boilerplate from load/save plug-ins.
Remove gimpprocedure-private.[ch] because install() and uninstall()
are now virtual functions of GimpProcedure.
They only contain private functions and don't need to be installed or
included by gimp_pdb_headers.h.
The PDB generation part is done by adding a "lib_private" variable
that can be set on PDB groups which should not be public API; the rest
is manual Makefile fiddling.
gpointer is not introspectable, and we should always prefer boxed types
over generic pointer types (thanks nielsdg!).
This also fixes gjs (JavaScript) binding of our API which was dropping
our properties even though they didn't need to be handled in JS.
See https://gitlab.gnome.org/GNOME/gjs/issues/266
Add gimplegacy-private.h to keep it separate from gimp-private.h.
The legacy code could now be removed and GimpPlugIn-ported plug-ins
would not notice.
Move all old wire code to gimplegacy.c and add wire code to
GimpPlugIn, which now talks with the GIMP core all by itself.
Add some more ASSERT_NO_PLUG_IN_EXISTS assertions to gimplegacy.c and
fix new code that was still using legacy API.
and make the places who need it dispatch between it and
_gimp_read_expect_msg() depending on whether there is a GimpPlugIn
instance. Also clean up some includes.
Basically this commit makes sure that all return values that are marked
as "Returns:" also have a `(nullable)` annotation if it is mentioned on
the same line that NULL can also be returned.
This will prevent a few problems in GObject-introspection.
Start copying all the actual wire communication to GimpPlugIn, and
move the legacy versions to gimplegacy.c.
This implies having the entire protocol code twice, but without any
if(PLUG_IN) { plug_in_stuff(); } else { legacy_stuff(); }
At the moment it is a wild mixture of old and new, but when finished
the wire code in gimplegacy.c will be entirely separate from the wire
code in GimpPlugIn, which will make it easy to g_assert() that only
one API is used by a plug-in.
