Allow implementing a GAction interface

[andy128k]

Closes #1636

[andy128k]

Another issue I have is that GAction implies that an implementor had some properties.

(process:527349): GLib-GObject-CRITICAL **: 20:51:28.369: Object class ExampleRenamedAction doesn't implement property 'state-type' from interface 'GAction'

(process:527349): GLib-GObject-CRITICAL **: 20:51:28.369: Object class ExampleRenamedAction doesn't implement property 'state' from interface 'GAction'

(process:527349): GLib-GObject-CRITICAL **: 20:51:28.369: Object class ExampleRenamedAction doesn't implement property 'parameter-type' from interface 'GAction'

(process:527349): GLib-GObject-CRITICAL **: 20:51:28.369: Object class ExampleRenamedAction doesn't implement property 'enabled' from interface 'GAction'

I have no clue what would be an idiomatic way to use g_object_class_override_property in Rust.

[bilelmoussaoui]

Another issue I have is that GAction implies that an implementor had some properties.

(process:527349): GLib-GObject-CRITICAL **: 20:51:28.369: Object class ExampleRenamedAction doesn't implement property 'state-type' from interface 'GAction'

(process:527349): GLib-GObject-CRITICAL **: 20:51:28.369: Object class ExampleRenamedAction doesn't implement property 'state' from interface 'GAction'

(process:527349): GLib-GObject-CRITICAL **: 20:51:28.369: Object class ExampleRenamedAction doesn't implement property 'parameter-type' from interface 'GAction'

(process:527349): GLib-GObject-CRITICAL **: 20:51:28.369: Object class ExampleRenamedAction doesn't implement property 'enabled' from interface 'GAction'

I have no clue what would be an idiomatic way to use g_object_class_override_property in Rust.

GtkEditable provides a method to automatically add the properties for you along with the get_property and set_property. Either the example implementing a custom Action would have to do that manually or you could add helpers that would do it automatically for you, see the EditableImpl in the gtk4-rs repo.

[bilelmoussaoui]

The properties macro allows overriding properties by the way

[andy128k]

I gave it a try but received an error

(process:718176): GLib-GObject-CRITICAL **: 22:59:07.010: When installing property: type 'ExampleRenamedAction' already has a property named 'name'

I guess it's because of the sequence of initialization. When I define a class and implement an interface then class installs properties first and interface fails to install them.

[sdroege]

The way how you do it now should work or not?

[andy128k]

It does not work. I guess It works as expected in the cases like

class A implements IFace

class B extends A implements IFace {
   override property x of IFace
}

where initialization sequence is: A, IFace, B. So, when B is initializing IFace is already there and could be overriden.

but does not work for

class B implements IFace {
   override property x of IFace
}

where initialization sequence is: B, IFace. So, B has nothing to override yet.

[andy128k]

@bilelmoussaoui Thanks for pointing a direction! Unfortunately Gio does not provide helpers similar to gtk_editable_*_delegate_*, so I needed to replicate C code myself.

---

I am not sure if I fully understand the subject but this line seems wrong. Gtk's doc uses NUM_PROPERTIES as a second parameter which is usually not an amount of properties but amount + 1 (props are numbered from 1). gtk_editable_install_properties does not add 1.

[sdroege]

That's a nice hack for the properties :)

[sdroege]

If you pass in Self::type_() here then you don't need to do all the parent type walking inside from_type() and always have directly the correct type

[andy128k]

That's what I've found in GtkEditable's code.

[sdroege]

@bilelmoussaoui Why does it do things so complicated, do you remember? :)

[sdroege]

This should probably be moved into the set_qdata() call. Now you'd get inconsistent behaviour if the function returns Some on the first call and None afterwards

[andy128k]

It's not directly forbidden by Gio's docs, but I think it could be insane if the type of a parameter would change during the action's lifetime.

A similar assumption is made in ActionGroup.

[sdroege]

My personal approach here would be to either only call the function once to don't give the user a chance to do it wrong, or otherwise at least to add an assertion :)

[andy128k]

I've changed this and other places (name, parameter_type and state_type) in the way that the function is called once.