On 12/03/10 03:25, Josh Faust wrote:

I'll invest the time when I can see that someone has an interest and a use-case for this form of shared memory transport.

I guess my question then is, if you don't have a use-case, and the common use will likely cause crashes, why release the code and tell others to use it?

I don't think common use will likely crash it: e.g. images and point cloud are not always changing size. And the nodes should not crash if the object size stays constant.

I have a theoretical use case for the sharedmem plugin which is "exchanging very large objects between processes that cannot be converted to nodelets, and with reduced delay or reduced network clogging". I am not sure this corresponds to a problem people really have, so I'm waiting for feedback on that. I would say this part of the code is in beta, sufficiently functional to raise interest and to be tried in controlled situations, but I agree it still need to be worked on.

Actually, all this generic message transport is a nice theoretical concept, but I'm not sure how much of it is solving a problem people really have. It was fun to develop though.

Let's wait and see.

-- 
Dr. Cedric Pradalier
http://www.asl.ethz.ch/people/cedricp