detect and report local graph changes

[wjwwood]

Fixes #168

This required, apparently for the first time, that the node is given to the rmw_destroy_client and rmw_destroy_service calls, so that will change the rmw API.

I'm still debugging some issues with the tests.

[wjwwood]

So, it seems that Connext's request-reply uses some strange topic names. For example, the service server creates:

+S add_two_intsRequest <example_interfaces::srv::dds_::AddTwoInts_Request_>
+P add_two_intsReply <example_interfaces::srv::dds_::AddTwoInts_Response_>

But the client creates:

+S add_two_intsReply_ac1702d4000020e70000000180002003 <example_interfaces::srv::dds_::AddTwoInts_Response_>
+P add_two_intsRequest <example_interfaces::srv::dds_::AddTwoInts_Request_>

Note what looks like a GUID attached to the reply topic. It's not clear to me how the replier (service server) delivers the reply to the requester (service client).

I spent the better part of last night trying to figure out a better way to figure out when the service server is available for how Connext implements Request-Reply. I have some ideas, but as of right now this is not ready for merging.

[dirk-thomas]

Should this be changed back to auto?

Same below.

[wjwwood]

I think I did that so my autocomplete would work, but I can change them back: 90d4c86

[dirk-thomas]

This recent change broke rmw_connext_dynamic_cpp: http://ci.ros2.org/job/ci_linux/1870/testReport/

While not being actively tested at the moment maybe it can be easily fixed @wjwwood ?

[wjwwood]

I believe it is now failing because the wait_for_service is still not implemented, see:

http://ci.ros2.org/job/ci_linux/1870/testReport/(root)/projectroot/requester_replier_cpp__Primitives__rmw_connext_dynamic_cpp/

The tests could be disabled, but I don't know if it is worth it since we're not regularly trying to figure out which failures are known and which are not for that rmw implementation. The fact that this function is not implemented is already tracked in an issue: #203

I'm not sure about the other failures.

[dirk-thomas]

Since the merge numerous very simple tests which only do pub/sub are failing. I don't think the referenced ticket about wait_for_service is related to all of them. When I tried it with the new composition demos the executable crashed when calling one of the new functions around add_information. Therefore I am pretty sure that this is a regression of this PR.

[wjwwood]

Well I'm sorry it is broken, because it wasn't my intention to break it, in fact I spent some extra time trying to keep it up to speed. However, fixing this is not at the top of my priority list because it's not one of our supported rmw implementations right now.