On Wed, May 22, 2013 at 9:37 AM, Esteve Fernandez <esteve@apache.org> wrote:

> Hi,
>
> there doesn't seem to be a SIG for ROS comm protocols and APIs, so I'm
> writing here anyway :-)
>

There is a ros-sig-ng-ros mailing list. There was some good discussion, but
it has not been very active, lately:

 https://groups.google.com/forum/?fromgroups#!forum/ros-sig-ng-ros

I am cross-posting there to archive your message.

Following Dirk's talk about the future of ROS 2.0, I thought of
> experimenting with alternate protocols for serialization and RPC. I
> hacked together rosthrift [1], a package for using Apache Thrift [2]
> as the transport for serialization and RPC, it generates code out of
> Thrift definition files and integrates nicely with catkin.
>
> I just added support for Python, but Apache Thrift supports many more
> languages. Apache Thrift is being used by many open source projects
> (HBase, Cassandra, Scribe, etc.) and unlike Protocol Buffers, it comes
> with a full RPC framework for C++, Java, Python, Ruby, Perl and many
> more. Disclaimer: I'm one of the developers of Apache Thrift, so I'm
> clearly biased :-)
>
> I also pushed a bunch of examples [3] that use rosthrift and txrospy
> [4] that mimic the AddTwoInts example from the tutorial, but use
> Apache Thrift instead of TCPROS, thus replacing much of the low level
> networking code.
>
> Is there something in the roadmap about the network stack for ROS 2.0?
> I'd love to contribute! Maybe we could create a SIG and move any
> discussion there.
>
> Cheers.
>
> 1 - https://github.com/esteve/rosthrift
> 2 - https://thrift.apache.org/
> 3 - https://github.com/esteve/ros_beginner_tutorials
> 4 - https://github.com/esteve/txrospy
> _______________________________________________
> ros-users mailing list
> ros-users@code.ros.org
> https://code.ros.org/mailman/listinfo/ros-users
>

I don't know of any definite decisions.

There has been some discussion of using zeromq as a possible low-level
transport framework. It may solve some, but not all, problems with the
current TCP implementation. I've looked at it and been impressed, but
lacking hands-on experience, would not want to make any strong
recommendation.
-- 
 joq