[Biojava-dev] Feature interface change

[Thomas Down]

On Fri, Aug 23, 2002 at 08:58:21AM +1200, Schreiber, Mark wrote:
> 
> It would make biojava less sequence centric which imho would be a good
> thing. You could represent almost any kind of biological data this way,
> mapping it to sequence if needed but also to experimental results,
> literature searches etc.
> 
> How much breakage would occur to the biojava api? It would be good to
> have this sooner rather than latter.

I think doing this properly will mean some pretty serious
breakage.  There are actually a number of fairly big issues
that have been talked about in various context lately.  Off
the top of my head:

   - FeatureCards
   - Consistent use of identifiers.
   - Use of ontologies (I think this one relates quite stronly to
     the FeatureCards).
   - Revamping the ChangeEvent system to handle event forwarding
     better (I've already been trying out a few ideas in this
     direction in the latest version of the BioSQL code).
   - Object query language (more powerful/flexible than the
     FeatureFilter `language').
   - Probably some others.  SeqIO changes?

Pretty much all of these imply at least some API breakage.  I'm
tempted to route around this by forking into `evolution'
(1.x, try to remain completely API compatible with current stuff)
and revolution (2.0, lots of cool stuff, lots of API breakage).
Maybe even start with a completely fresh repository for the
2.0 stuff.

When I suggest this, I'm thinking particularly about Tomcat's
fork into 3.x and 4.x branches (where 4.x was a complete re-write).
It caused some pain, but 4.x ended up a much nicer package. 
It certainly seems a better idea to me than trying to implement
major changes to the current codebase (and still breaking everyone's
code in practice...).

Anyway, I think this is all likely to be fairly long term stuff...
Any other thoughts?

     Thomas.