Re: built-in types

[James Henstridge <james daa com au>]

Adrian Feiguin wrote:

> I wonder if somebody can shed some light on the way the BOXED data types
> are generated. I understand that maketypes.awk is used, are there any
> guidelines? Thank you!
For your own code, I wouldn't recommend copying the maketypes.awk stuff.

First of all, you should ask yourself if the type you are talking about 
fits the model of a boxed type.  It should be safe to copy like a value 
(or support reference counting as a substitute for copying).  If your 
type fits this model okay, then here is what you should do:

  1. implement copy and free functions.  The copy function takes one
     argument (the one to copy/ref), and returns the copy.  The free
     function takes one argument (the one to free/unref), and has no
     return value.
  2. Write a get_type() function for the type, something like this:

       GType
       my_boxed_get_type(void)
       {
           static GType boxed_type;

           if (!boxed_type) {
               boxed_type = g_boxed_register_static("MyBoxed",
       my_boxed_copy, my_boxed_free);
           }
           return boxed_type;
       }

Then just put a prototype for the get_type function in your headers, and 
a macro to call it:

   #define MY_TYPE_BOXED (my_boxed_get_type())

You can then use MY_TYPE_BOXED as the typecode for the boxed type in 
signal and property definitions.  You don't need to make the copy/free 
functions public -- they can be accessed through the GBoxed APIs:

   copy = g_boxed_copy(MY_TYPE_BOXED, original);
   g_boxed_free(MY_TYPE_BOXED, copy);

(Language bindings may use the above API to handle arbitrary boxed types).

If your type doesn't behave this way, it is also possible to register it 
as a subtype of G_TYPE_POINTER.  However, you should make your type 
usable as a boxed or GObject if possible.  Don't register as a pointer 
type just because you don't want the value to be copied during signal 
emission -- you can use the static scope flag for the signal argument in 
this case instead.

James.