Re: Widget descriptors as unions

[Florian Pelz <pelzflorian pelzflorian de>]

On 08/29/2015 10:53 AM, LRN wrote:
> On 29.08.2015 11:46, Florian Pelz wrote:
> > On 08/29/2015 10:39 AM, Jean-Marie Delapierre wrote:
> > > I have explained (I hope in a clear way, but I begin to doubt about it)
> > > that the unions I suggest only own pointers, not child objects as you
> > > say in your answer.
> >
> > I think it is clear, but you still need to declare one such union for
> > each class.
>
> Union-pointers sound like a neat trick.
>
> Can't you generate headers with these union-pointer declarations
> automatically, using gobject introspection?

I don't like requiring an additional tool for building. It also makes
the inner workings less transparent than using only the C high-level
assembler with macros. I don't think you can do the declaration
conveniently without additional tools, i.e. just a C preprocessor and
compiler.

> I'm also not sure what are the ABI implications. Are unions compile-time
> only? I guess you'd only change a pointer-union if the class hierarchy of
> its corresponding widget class changes, which sounds like a good excuse to
> recompile anyway...

ABI-wise, unions don't make a difference. But: If you use a library that
uses a class that inherits from a GTK class and the GTK class starts
implementing a new interface, you'd need to recompile the library's
header files.