Re: Backend and non-backend (was Re: Stencil bounding box)

[Johannes Schindelin]

Hi,

On Wed, 3 May 2006, David Feuer wrote:

> On 5/2/06, Han-Wen Nienhuys <address@hidden> wrote:
> 
> > There are various ways of comparing numbers. The point is that you need
> > to know which pairs of numbers/rectangles/etc. to compare. For that, you
> >   need to get the elements in a canonical order, and that *must* be done
> > with discrete quantities. Some possibilities
> 
> I don't understand why it has to be done in a canonical order, rather
> than looking for the closest match for each element.

Because the closest match can be the wrong match. This is about regression 
testing.

> > for the actual distance function, there are a lot of possibilities.
> > You'll probably need to experiment what works best.
> 
> If we're looking to measure small changes, area of union minus area of
> intersection, divided by the area of the union, would probably be
> good.

If a really nasty bug creeps in, which makes the new bounding box very 
tiny, but keeps the location, this distance will be about 100%.

Ciao,
Dscho