[Kde-bindings] KDE/kdebindings/ruby/qtruby

[David Palacio]

On Thursday 11 December 2008 05:45:58 Richard Dale wrote:
> On Thursday 11 December 2008 00:56:10 David Palacio wrote:
> > On Wednesday 10 December 2008 07:46:23 Richard Dale wrote:
> > > SVN commit 895236 by rdale:
> > >
> > > 	* Special case the Qt::Image.bits() method. Use the const variant so
> > > that Qt doesn't make a deep copy. Is that what Davor Ocelic and David
> > > Palacio would like it to do?
> > > 	* For other uses of 'uchar *' return types add a marshaller, hoping
> > > that will be null terminated.
> >
> > 1) I checked which functions returned an uchar* in the Qt includes. All
> > of them are expected to be used to directly transform the data.
>
> I think the problem with these is that to use them we would have set the
> internal pointer in an existing ruby string to the value. But then both Qt
> and Ruby would think they owned the 'char*' and try to free it later.
>
> I think my change to allow 'uchar*' return types by default expecting them
> to be null terminated is probably pointless, and it might be better to
> revert that commit so that we get a 'can't handle type' error for a method
> and we'll know it needs special casing.
>
> > 2) If the returned String when changed also affects the QImage, it makes
> > sense. That is the intention for the bits method. (Checks on IRB... it
> > does not).
> >
> > 3) I was blindly translating a C++ application to Ruby and stumbled on
> > bits(). Now I did it in another way, maybe not as efficient though.
>
> I think if you can do it another way that is better as it seems difficult
> to do it without risking double memory frees as above. Certainly we need to
> special case this kind of thing, and the general purpose marshalling code
> in qtruby won't handle this kind of tricky stuff.
>
> > If 1) is true and 2) does not change (which I think is unlikely to do),
> > this might have little sense.
>
> This is the code to Qt::Image.bits():
>
> static VALUE
> qimage_bits(VALUE self)
> {
>   smokeruby_object *o = value_obj_info(self);
>   QImage * image = static_cast<QImage *>(o->ptr);
>   const uchar * bytes = image->bits();
>   return rb_str_new((const char *) bytes, image->numBytes());
> }
>
> I think the rb_str_new() call will make a copy of 'bytes' and so it doesn't
> make any difference whether we use the const or non-const version of
> bits(). But it does avoid copying bytes twice, once on the QImage::bits()
> call and again in rb_str_new().
>
> -- Richard
> _______________________________________________
> Kde-bindings mailing list
> Kde-bindings at kde.org
> https://mail.kde.org/mailman/listinfo/kde-bindings
I'd rather not use such a trick. Even if it is for read-only use, you can use 
other methods for reading pixels.

I think what is needed now is to put some information on the wiki about uchar* 
methods.