Re: gtk_tree_model_get end of Line marker

[David NeÄas <yeti physics muni cz>]

On Sun, Oct 07, 2012 at 05:43:52PM +0200, Arne Pagel wrote:
> because of my lazy programming style I sometimes forget or overwirte
> the comma before the -1  end of line marker for the gtk_tree_model_get
> or gtk_tree_store_set function.  This error can usually not be
> detected by the c-compilers, so that I had to spend some really
> unnecessary time to fix this typo.
>
> Now I defined a makro for this, and in the case I forget the comma
> again at least I get a compiler error:
>
> #define TREE_MODEL_EOL (int)-1
>
> gtk_tree_model_get(model,iter,STUFF,&stuff,TREE_MODEL_EOL);
>
> I think something like this would be integrated officially and also
> the all the examples would be updated, at lot of headache of many
> people could be avoided.

Errors related to the -1 sentinel could be avoided by much more
straightforward means: defining gtk_tree_model_get() with column id and
value pointer reversed, i.e. first comes the value pointer (presumably
non-NULL), then comes the column id.  Then gtk_tree_model_get() could
just carry __attribute__((sentinel)) a.k.a. G_GNUC_NULL_TERMINATED.  But
the Gtk+ devs decided otherwise.

It would be nice (also in other cases) if gcc had an attribute for
arbitrary constant-value sentinels.  Unfortunately, it has not.  The
corresponding bug

    http://gcc.gnu.org/bugzilla/show_bug.cgi?id=28319

seems to be stale.  I'm not sure about LLVM but it does not seem to
support this either.

While I've never forget the comma before -1 I have forgetten the -1
entirely a few times so something that prevents all sentinels errors
would be quite useful.

Yeti