Re: g_(s)list_data()

[Joel Becker <jlbec evilplan org>]

On Thu, May 24, 2001 at 05:08:18PM -0700, Derek Simkowiak wrote:
> 	I disagree that this is nice to read.  I'd prefer to just see the
> typecast -- at least then a new user will know what's going on (instead of
> needing to lookup yet another macro, especially one that doesn't exist
> in any of the current code).

	The macro can be learned.  I'm of the 'private accessor' school
of thought, so I like the non-casting version better.  This is certainly
a matter of taste.  Do you use

foo = g_new(Foo *, 1);

or

foo = (Foo *)g_malloc(1);

?  I personally prefer the former.

> 	It would need to follow the "object-like" Glib naming convention
> of
> 
> g_thing_action(GThing *thing, ... )

	That's fine.  I was following g_new(), but you are right with
this one.

> 	Although g_list_next() exists, it has been agreed that there is
> way too much code that directly accesses list->next to deprecate that kind
> of peeking.  In short, even if list->next was meant to be private, through
> real-world practice it has become public.  And since list->data is already
> public, there is no need for g_list_data().

	I know folks use list->next and list->data.  I am not, in any
way, claiming we can fix it in existing code.  But we can provide pretty
accessors for new code that some folks might like.

> 	As a frequent Glib user, if g_list_data() gets added, I will still
> use list->data in all my code.  Sorry  :)

	Go right ahead.  This is certainly a structure we can't change
due to compatibility with existing code.  But I think direct structure
access is semi-ugly outside of its library.  This is my preference (and
probably some chunk of folks as well, but certainly not everyone).

Joel


-- 

"Every new beginning comes from some other beginning's end."

			http://www.jlbec.org/
			jlbec evilplan org