On 11/23/11 4:54 PM, Aliaksandr Autayeu wrote:
>
> Mixing independent things into one jar (in this case also one module)
> is really what we are doing with opennlp tools.
>
> As pointed out this is the case for cli and at least also for tokenizer,
> sentence detector,
> sequence based components (pos tagger, chunker and name finder), parser
> and the coref component.
The pipeline components are closer to each other than CLI tools to all of
them.
Looks like we agree about that everything should be divided by java
packages,
different package names for semantically divided pieces.
Adding an additional separation via maven modules will enforce that the
classes cannot be used by another component, which does not explicitly
depend on it inside the pom. And it gives us the option to provide just
the classes
which are necessary to perform a given task (multiple jars) vs.
everything (single jar)
to a user.
I think the additional complexity (and things which can go wrong) isn't
worth the small advantage of less classes on the users class path.
Anyway, if everyone thinks I am mistaken, I will not stand in the way.
Jörn
