On Dec 4, 2010, at 4:40 AM, Jörn Kottmann wrote: > On 12/3/10 11:00 PM, Jason Baldridge wrote: >> I think it is more of a conceptual thing. You download a package like maxent >> so you can build models for your own prediction task, and it would be nice >> to not have lots of extra documentation and such getting in the way of >> understanding how to get up and running with the software, etc. I just think >> it makes sense as a separate release unit that the OpenNLP toolkit depends >> on without swallowing it whole. >> > > I clearly disagree here. Having a separate maxent distribution maybe a little > more convenient > for maxent only users, but it has other disadvantages. > > To release the tools project we would have to release maxent always before > which > will add a two - three week delay for every tools project release which > depends on the maxent change. > Its also more work to release one more item. > Its also more overhead from the project structure, we need a separate > documentation project > for maxent, etc. Do you also want a separate website for maxent ?
We had this problem in Lucene/Solr and eventually merged the two so that Solr could always be on the latest version of Lucene. That being said, it doesn't necessarily imply a separate website, etc. > The testing is also tangled, there are no good tests for maxent, expect the > quite > extensive regression test we do for the tools. In the end both projects must > be > ready for release, and maxent can not be really released separately. > In practice I think it means that we have usually to release both at the same > time, > expect for smaller minor fixes.
