To add something here, this can be considered as a dynamic extendable global class space for type systems. Every Annotator refers to the host and therefore is able to see all attached types added by fragments. The order of attachement defines what type is discovered if you have to Classes with the same name (like the old ClassPath approach). Buddy ClassLoading as mentioned before is a proprietary mechanism of Eclipse. It does pretty much the same like Dynamic-Import, but has a better scoping. With our current approach, we don't need this mechanism at all. We can refer to what we need by OSGi wiring and the use of OSGi services, which is more OSGi like and reduces the static links betweens components.
On 1/9/07, Yurdaer Doganata <[EMAIL PROTECTED]> wrote:
I just wanted to add to the previous note that Type System Fragment Bundles conceptually become part of the host bundle when they are attached. That is they use the same class loader. As UIMA OSGi component bundles are activated and made available for UIMA deployment, before CAS is produced, a Type System Bundle is created dynamically and used by all components. -Yurdaer ----- Original Message ----- From: "Yurdaer Doganata" <[EMAIL PROTECTED]> To: <[email protected]> Sent: Tuesday, January 09, 2007 12:14 PM Subject: Re: OSGi enablement and JCas > Distributed type definitions can be supported by UIMA OSGi extension by > using the concept of "fragments" and "type system host bundle". In this > approach, all components depend on (require) the "type system host bundle" > which can dynamically extend its class space as the type system fragments > required by various components are attached to it as needed. This gives > the flexibility of reconfiguring the process flow by adding new components > without actually shutting down the JVM. Of course, the CAS object has to > be refreshed every time a new type is introduced. > > ----- Original Message ----- > From: "Michael Baessler" <[EMAIL PROTECTED]> > To: <[email protected]> > Sent: Tuesday, January 09, 2007 11:50 AM > Subject: Re: OSGi enablement and JCas > > >> Adam Lally wrote: >>> On 1/9/07, Eddie Epstein <[EMAIL PROTECTED]> wrote: >>>> On 1/9/07, Thilo Goetz <[EMAIL PROTECTED]> wrote: >>>> > a) backward compatibility and b) some people actually seem to be >>>> > using >>>> > it and don't want to be without it. >>>> >>>> Then perhaps OSGi components will not be compatible with distributed >>>> definitions, >>>> but other component packaging will still be supported that do. >>>> >>> >>> That makes sense to me. >>> >>> -Adam >>> >>> >> I don't like the idea that the UIMA OSGi extension does not support >> distributed type definitions but the "old" framework does. I think OSGi >> is the future and if >> we do not support distributed type definitions with OSGi we should >> deprecate that completely. But I don't think this is really possible. >> >> -- Michael >> >
