Mostly the same code as GimpProcedure in app/pdb/.
Move the "run" function to GimpProcedure. Add API to GimpPlugIn to
list and create procedures, and always keep a list of the plug-ins
procedures around. Still only using the old params and return_vals.
The new way of doing plug-ins:
- subclass GimpPlugIn in your plug-in
- implement its query() and run() methods, run() will move to a
new GimpProcedure class soon
- instead of MAIN(), say GIMP_MAIN(YOUR_PLUG_IN_TYPE)
Instead of keeping around a GimpPlugInInfo struct, libgimp will
create an instance of your plug-in class, keep it around during
the plug-in's lifetime, and call its virtual functions.
... GimpRunProc function type.
In gimp_install_procedure(), make sure that @params and @return_vals are
processed as arrays, otherwise they are unusable.
As for the run procedure, make so that @return_vals and @n_return_vals
are considered as returned values. In Python binding for instance, that
makes these not parameters anymore, but actually returnable by the run
function.
With these changes, I made the first fully functional GI Python plug-in,
which just creates a new image and a display for this image. Still a lot
to improve clearly, but we are on the right track. :-)
This is an alternative way to set up a plug-in callbacks, apart from
setting directly the PlugInInfo struct properties.
The reason is that setting directly the Gimp*Proc properties crashes the
plug-in, when done through the Python GI binding.
It is most likely a bug in Pygobject, unless we need the proper
annotation (which I haven't found yet). See:
https://gitlab.gnome.org/GNOME/pygobject/issues/24#note_564968
Setting these callbacks from the C code works fine though, hence this
new API. It is to be noted that the '(scope async)' is the most
important part in this function annotations. Without these annotations,
the function pointers become invalid at the end of the set_callbacks()
call, hence the plug-in crashes when they are actually called.
Unfortunally I am also notified by ebassi that using (scope) at all (any
of the 3 possible values) is just wrong. An API change will be
necessary. For the time being, I leave this like this, for the sake of
testing further, but we'll need to improve things.
It doesn't really help yet with the problem I encountered allowing to
set and run these Python callbacks from the C code (cf. pygobject#347),
but at least let's improve a bit the documentation.
With GObject introspection, this allows to properly use this function,
otherwise it sees the argv argument as a string (and not an array of
string), which cannot be used properly.
For instance, with Python binding, you can just call it like this:
> Gimp.main (info, sys.argv)
At first I thought these could be different namespaces, but actually
GObject Introspection parses files and can only use (AFAICS) the
namespace actually used in our C function, which is always `gimp_` (and
not `gimpbase_` or whatever.
So make the introspection at the root level, and it will include all
libgimp* libraries in one namespace, same as the C lib anyway. For now
only libgimp and libgimpbase as I am still testing.
Also I move the introspectable sources in their own file in order to
share the list between the library building Makefile and the GI
makefile because I can't find how to pass over variables otherwise.
So far only libgimpbase is introspected. It just works though (I could
test that I could just run a plug-in which could access libgimpbase API
without any problem).
The g-ir-scanner call outputs a lot of warning but I won't care for
these right now.
The `introspection.m4` is taken straight from GEGL tree.
Add an assigned-threads variable to the dashboard's misc group,
showing the number of worker threads which currently have work
assigned. See commit
gegl@fc532f06773bd292c86abee31ac18f62e26d39df.
and make them build again without any wrong symbols.
This does not enable the devel-docs build yet, I'll fix
module by module before enabling the build again.
When translating a layer group, avoid separately updating the
original area of the child layers before translating them (as per
the fix to issue #3484), as this results in quadratic time
complexity w.r.t. to the maximal subgroup nesting level. Instead,
simply defer the updating of the group's offset until *after*
translating the child layers, so that their original area isn't
clipped by the parent, while their new area is still properly
updated even if the parent's size changes (see comment in code).
Michael Natterer