Re: Maintaining the GTK+ and GDK abi

[Matthias Clasen <mclasen redhat com>]

Am Mittwoch, den 11.08.2004, 16:52 -0400 schrieb Morten Welinder:
> I like the idea, but I really, really dislike the implementation of
> the idea.
> 
> Currently, all gtk+ functions have been given an alias function in
> gtkalias.h, gdkalias.h, etc.
> 
> Problems:
> 
> * Any change to any .h file in gtk+ will cause a recompile of all
>   .c files.  That sort-of kills the idea of "make".
> 
> * Every .c files compiled must now parse an extra 11167-line .h
>   file.  The pre-processor must do gazillions of macro expansions.
> 
> * Debug information for every single of these alias functions are
>   now in the .so file taking up tons of space.  I'm sure there is
>   some gdb issues lurking in there.
> 
> * Every gtk+ function now has two definitions making life harder
>   when working with things like etags.
> 
> * We broke even C99's identifier length guarantee.  (Think 71 chars:
>   gtk_color_selection_set_change_palette_with_screen_hook__internal_alias.)
> 
> * When adding a new function there is suddenly a lot of things that
>   can go wrong.
> 
> I pointed this -- and a solution -- out in bug 145519, but that
> was just closed.  I don't understand this opposition to doing
> things right.
> 
> (And that is even ignoring the at least three bugs remaining in the
> implementation of the alias scheme when the bug was closed.)

Sure Morten, the current approach can be improved. Looking forward to
see some patches from you...