Re: inline pixbufs

[Havoc Pennington <hp redhat com>]

Darin Adler <darin@eazel.com> writes:
> > +        if (length >= 0 &&
> > +            length < (height * rowstride + 13)) {
> > +                /* Not enough buffer to hold the remaining header
> > +                 * information plus the data.
> > +                 */
> > +                
> > +                return NULL;
> > +        }
> 
> Need to handle overflow here. If a garbled height or row stride is a very
> large value, then length might look good even though it is too short.
>

I'm thinking you mean height*rowstride overflows my int, if you don't
mean that then correct me.
 
> Could have a check here that has_alpha is consistent with n_channels. Does
> it work to have n_channels == 4 and has_alpha == false?
>

Well, the code works because n_channels gets ignored, but worth fixing
for clarity.
 
> One way to do it is:
> 
>            if (n_channels != (has_alpha ? 4 : 3))
>                    return NULL;
>

 
> Also, someone pedantic might suggest that you reject booleans that are
> anything other than 0 or 1, rather than treating all non-0 as 1, but that
> person would be way more pedantic than me :-)
> 

gdk_pixbuf_new() should canonicalize the incoming bool anyway, if it
doesn't. GTK normally does this so we can use equality tests on bools.

Thanks!

Havoc