[cairo] 1.6: a8-mask: Making non-aligned strides illegal

[Behdad Esfahbod]

On Sun, 2008-01-27 at 22:48 -0800, Carl Worth wrote:
> > > And also documenting cairo_image_surface_create_for_data that the
> > > stride value passed to it should always come from calling the function
> > > above.
> >
> > Humm, not sure how limiting that would be.  Do you mean it should be the
> > stride_for_width() of the actual image width,
> 
> "stride_for_width() of the actual image width" is what I intended.
> 
> >                                               or the stride for some
> > (any) width?   I mean, say I want to create a cairo surface for a
> > subsurface of a larger buffer.  The width passed to cairo may be way
> > smaller than the width used for stride computation.
> 
> So you might have noticed that I dropped the suggestion in the
> documentation for cairo_image_surface_create_for_data that playing
> games with stride is a good way to generate sub-surfaces. More on that
> below.

Right.  Subsurfacing is only one use case for using a greater width
value for stride computation than the actual width.

I still prefer at least allowing "any width" for stride.

> > Also while at this, how does this work for:
> >
> >   - 64-bit machines?  Is the requirement still just 32?  Guess yes.
> 
> Yes. The requirement comes from the fact that all of pixman is
> currently in terms of uint32_t. But, yes, the existence of 64-bit
> architectures is exactly why we want to be able to "upgrade" the
> alignment constraints in the future. Hence the newly required function
> call.

Right.  Also, in the patch I'd define a macro and use that instead of
hardcoding the 32-bit requirement in more than one place.

> >   - Machines that do not allow unaligned access.  Do we also require
> > that data be aligned properly?
> 
> Right. And this is why the sub-surface thing is a problem.
> 
> I tried to address this in the new documentation for
> cairo_image_surface_create_for_data by stealing language from
> malloc. What I'm trying to say there is that the data pointer needs to
> be properly aligned for the current platform. Take a look at the
> wording I used and see if you can improve it at all.

Sounds fine.  I marginally prefer requiring int alignment than any-type
alignment.  But who knows what we'll require in the future...

> -Carl
-- 
behdad
http://behdad.org/

"Those who would give up Essential Liberty to purchase a little
 Temporary Safety, deserve neither Liberty nor Safety."
        -- Benjamin Franklin, 1759