Sirtaj Singh Kang (ssk@physics.unimelb.edu.au) spake thusly:
> 
> On Sun, 23 May 1999, Jo Dillon wrote:
> [snip]
> >   Whereas Qt uses RGB in one table and A in another, if I recall correctly.
> 
> I don't know what happens inside a QPixmap, but QImage also stores the
> alpha channel for each pixel in the MSB of the 32-bit data.

  Hum. Maybe it was Imlib that had a separate alpha table. I recall
there was some sort of difficulty along these lines a few months ago.
QPixmap's are X Pixmaps, which simply have no clue about alpha channels.

> That would be absolutely disastrous considering how X servers grow in
> memory footprint over time. I looked at some of the Imlib image loading
> routines after writing the PNG reader. It looked almost exactly the same
> as my code (except that imlib saves the gamma info etc in a separate
> struct, whereas I got libpng to use it while decoding) so I don't think
> that's where the speed improvement comes from. I guess it has to be
> somewhere in the conversion to an X Pixmap. 
> 
> Anyone feel like checking this? 

  I think that's it - I did peek at this code ages ago and there is
some highly optimised stuff for different depths of X image and so forth.
I think it also supports 16bpp XImage's directly, whereas QImages don't.

-- 
	Jo