Bertrand Delacretaz wrote:
On 6/30/07, Grant Ingersoll <[EMAIL PROTECTED]> wrote:
...My main concern w/ extracting Nutch is all the dependencies on
Hadoop, etc. But it does seem like the shortest path for me....
I've mentioned Tika to a few colleagues lately, and one thing that
comes up often is that there are many document/format parsing
libraries around, which should ideally be usable as Tika plugins with
as little changes as possible.
But these libraries' dependencies are all around the place, and
probably conflicting in many cases.
It might be good to take that into account in the design of Tika, and
use solid classloading and isolation mechanisms. OSGI comes to mind,
assuming it doesn't bloat the whole thing.
Yes, in many cases a solid classloading mechanism is a must and OSGi
definitly implements this properly.
I think, we can leave this open (= do not need to require OSGi) if we
have an open way of registering the plugins. Registering in an OSGi
environment might then be slightly different compared to registering in
a non OSGi environmnent. Of course, using the latter one might result in
classloading problems :) But then it's up to the developer to decide in
which environment tika should run with all the pros and cons that come
with this decision.
Carsten
