best practices for defining indexes

[Philip Ogren <ph...@ogren.info>]

I don't have an intuition of what would be best practices for managing 
indexes I define might be.  It seems awkward to define them in the 
analysis engine descriptor file unless I refer to the index only in the 
corresponding implementation which is not the case for me.  It makes 
sense to import them instead as needed.  But this still seems difficult 
to maintain because how do I know where an index might be used?  The 
same problem exists for types - so what I do is I create an uber type 
system descriptor file the imports each type that I have defined (I 
create one descriptor file per type in the package directory where the 
generated jcas class goes.)  Should I define all my indexes in one file 
and then import this file in each analysis engine descriptor file?  This 
is what I am inclined to do - but would be glad to hear other opinions 
on this.

Also, is there a naming convention that  I should observe for naming 
indexes?  My thought is to concatenate the package name and class name 
of the corresponding JCas class of the type being indexed and the 
feature that is indexed.  That way I can find determine the index I need 
at runtime.  Thoughts?

Thanks,
Philip