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

[Han-Wen Nienhuys]

David Feuer schreef:
> On 5/1/06, Han-Wen Nienhuys <address@hidden> wrote:
>
> > I don't see who would benefit from a refactoring of the backend, so IMO
> > it's a waste of time.  By contrast, having a continuous distance measure
> > between different outputs would speed development up tremendously,
> > because we will be able to spot and repair bugs as soon as they crop up.
> >
> > You're free to submit patches to the backend, though.
>
> Refactoring:  anyone who wants to write a new backend or edit an
> existing one, and anyone who wants compact output files.

Given that we haven't had anyone writing or modifying the backend in the 
past 5 years, I doubt that there is much need for this.

> Distance measure:  I don't yet understand what kinds of changes you
> want to be considered significant and what kinds you want to be
> considered insignificant.

it should take geometry and appearance into account, ie. where the 
symbols end up on the page, and how large they are. What I would really 
want is to have a similarity measure for the image produced by two PDF 
files, but that's a rather hard problem, and it will be easier if we 
take advantage of the fact that we can access the backend.

I think the main problem is to make the measure invariant under changes 
that do not influence the end result.  Eg. due to internal restructuring 
grobs might be output in a different order between versions or runs. 
That should not affect the distance measure.

Basically my proposal is outputting (per grob) a tuple of

  1. discrete information - unlikely to change between versions,

  2. continuous information - floating point numbers that are likely to 
change subtly between versions

sort the tuples by the discrete component, and take a standard Rn metric 
for the sorted floating point numbers.

--

Han-Wen Nienhuys - address@hidden - http://www.xs4all.nl/~hanwen

LilyPond Software Design
 -- Code for Music Notation
http://www.lilypond-design.com