Subject: Re: [boost] RFC: edit_distance / edit_alignment library
From: Erik Erlandson (eje_at_[hidden])
Date: 2013-07-24 13:19:33
----- Original Message -----
> Erik Erlandson wrote:
> >> > I am grappling with how best to represent the returned "edit
> >> > script".
> >>
> >> the most generic way to return the edit script
> >> would be to template the algorithm on an output-handling class:
> >>
> >> template <typename ITER1, typename ITER2, typename OUTPUT>
> >> void diff(ITER1 begin1, ITER1 end1, ITER2 begin2, ITER2 end2, OUTPUT&
> >> output)
> >
> > I had a design using this approach, which has obvious advantages.
> > The one thing I wasn't crazy about is that to do this, the implementation
> > layer is committed to always computing and/or storing start+length
> > information and edit-op cost information, so that the output handler
> > has its space of choices about which information it cares about.
>
> Not necessarily; you can allow a variety of methods in the output handler,
> with different names or signatures, and use "enable-if" SFINAE
> techniques to
> enable appropriate parts of the implementation algorithm.
True, it might be possible to apply Clever MPL Tricks, maybe involving boost::function_types, to allow smart template specializations from an output-class method. Considering how the Meyers and SSSP algorithms work, I'm now feeling less sure about how much it even matters. Good fodder for next round of tinkering.