Re: oversensible assertion

[Owen Taylor <otaylor redhat com>]

Tim Janik <timj gtk org> writes:

> On Fri, 23 Aug 2002, Owen Taylor wrote:
> 
> > Tim Janik <timj gtk org> writes:

> > > maybe that assertion should be (length < 1 || text != NULL) ?
> > 
> > I don't think so. There are two cases:
> > 
> >  A) length < 0; text is nul terminated string.
> >  B) length >= 0; text is an array of characters of length 'length'.
> > 
> > NULL is valid for B, not for A). If the canvas wants to support
> > 
> >   set (item, "text", NULL);
> > 
> > To mean "", it needs to do it explicitely.
> 
> i don't see the point here, there's really no good reason to demand
> extra code from the canvas and every other pango_layout_set_text()
> user at this point *if* you already support text==NULL
> with length=0.

If you support, a length argument, you *have* to support text==NULL,
length==0. That's because g_malloc (gchar, 0); gives you NULL.
But the string that is represented by (NULL, 0) is "". The empty
string. We've been very clear throughout the years that "" is
distinct from NULL.

> > (Just changing 
> > pango_layout_set_text() wouldn't make this stuff work in
> > general:
> >  
> >   if (len < 0)
> >     len = strlen (text);
> 
> i didn't claim that this is going to work, i'm well aware of
> strlen() not handling NULL, but this is completely besides the point.
> in your API, you chosed to allow text=NULL and not demand text to
> be a valid 0-terminated string if length=0. at that point, it's
> simply ridiculous for length<0 to not mean "i don't know the size,
> find out yourself" but instead "i don't know the size, however i checked
> that text is not NULL since i'm not passing length==0".

length == -1, doesn't mean "I don't know the length", it means
"the text is a NUL terminated string".
 
> > Is certainly present in lots and lots of places;
> 
> bad code seen elsewhere is not an excuse, and for the record,
> i come across this rather seldomly (if at all, it's usually
> surrounded by other code telling me the author either has a
> global "don't check for NULL" policy, or didn't know his
> pointers anyway).

I've written this code in quite a few places, and I'd
defend at as being 100% correct.

> > and in fact,
> > it wouldn't make it work in particular, since you'd just
> > transfer the problem to g_utf8_validate().)
> 
> really? wow, i actually expected this to be an issue as
> simple as length = !text && length < 1 ? 0 : length; since
> you already support length==0 && text==NULL, i'm sorry
> i didn't grasp the problem in it's full complexity.

Yep, you aren't grasping the problem, I think. :-)

While this may sound pedantic:

 - I actually don't have a real problem with making
   pango_layout_set_text() support NULL for text
   as an API addition. NULL frequently means "reset
   to initial state", and we do support 
   gtk_label_set_text (label, NULL) g_string_new (NULL)

 - But I would strongly object to the idea that anywhere
   that we have a text/len pair we should support
   NULL/-1.

     NULL/-1 => represents NULL 
     NULL/0  => represents ""

Regards,
                                        Owen