On Fri, 27 May 2011 02:15:37 -0700, Thilo Götz <[email protected]> wrote:
It is also worth pointing out that if you control the code, you
can write/package your annotators so they just run out of a jar
file. The resource loading mechanism of UIMA can load resources
from the classpath, i.e., from a jar file. I always package my
annotators that way. Then even if you package them in a pear
(which you don't have to), the pear install doesn't have to do
any of the error prone file rewriting magic.
It has been put forward that if you have resources like dictionaries
as separate files on disk, they are easier to maintain. That's
an argument I only buy for a very limited set of circumstances
(e.g., you have a web app and you let users edit the dictionary
on the fly or something). Generally, if we're talking about a
real deployment, you don't want to do that anyway. It's just as
easy to replace a jar.
Thanks, Thilo. That's a great idea. I'll try it.
Does anyone here know the status of using OSGi to load annotators?
I seem to remember something on alphaworks.ibm.com a few years ago, but
haven't heard anything since.
Dynamic Module Enabler for UIMA:
http://www.alphaworks.ibm.com/tech/dmeuima/download
Is there any effort or interest in bringing this DME thing into the UIMA
sandbox?
My company is to moving to using OSGi extensively. We are adopting the
Virgo platform from Eclipse.
http://www.eclipse.org/projects/project_summary.php?projectid=rt.virgo
We might be willing to contribute to a project to "OSGi'ify" UIMA. I'll
have to check with my boss.
I'm told that OSGi has features that can be used to deliver resource files
along with the Java class files. Simplifying the installation process in
comparison to PEAR files would be my main goal with OSGI'ifying annotators
and analysis engines. To make it as easy as possible to consume
annotators and deploy them on-demand into a multi-tenant UIMA cluster. I
would want to distribute the artifacts using maven.
Is anyone else interested in such a thing using OSGi?
Thanks,
Greg
