Add tf2 Buffer accessor in FrameManager

[IanTheEngineer]

As this modifies headers (though only adding API), this may need to target melodic-devel.

This PR would mitigate the Issue related to using tf2 in the FrameManager #1215 by creating a tf2_ros TransformListener and Buffer for use by those who wish to use the RViz API and rely solely on the tf2 libraries. In particular, MoveIt! is in the middle of a migration to tf2 ( moveit/moveit#830 ) and this PR would unblock that migration.

[IanTheEngineer]

@dhood @wjwwood any chance you could give this PR a lookover?

The benefits include allowing users of RViz' API to transition away from the deprecated tf library, toward tf2, by utilizing a shared Buffer object.

As far as I can tell, with this implementation, the only downside is that there lacks an option to re-use a tf2_ros::Buffer passed in as an argument to the constructor, but I can add that if it is desired.

[dirk-thomas]

@wjwwood Can you please take a look at this since it is necessary for the migration of MoveIt to tf2. Thanks.

[dhood]

you are right that this will break ABI and will need to target melodic.

the downside that I see is that we now have duplicated the listener effort for TF, which is a shame since the tf1 listener is actually using a tf2 buffer underneath. I know you had worked to expose it in #1215 (comment); did you consider that approach here already?

[IanTheEngineer]

@dhood I agree that reusing the tf listener would be ideal, but the underlying tf2_ros::Buffer object is not designed to be accessible outside of inheretance.
I ran into slicing issues in attempting to retrieve the pre-alocated & protected tf_buffer object and maintain its access inside a shared pointer, without it being deleted from underneath the shared_ptr and causing a coredump. In the end, I figure the best solution is probably the simplest (implemented in this PR) even if it is not the most efficient. However, I am open to recommendations on a better implementation (or pointers as to how my C++ foo is failing).

[dhood]

@wjwwood could you pitch in on this please if you get a chance? (and the split std::shared_ptr and boost::shared_ptr interface: maybe stick to boost for consistency in the API?)

[wjwwood]

Yeah, I was hoping to do it today. We'll see...

[wjwwood]

I'd really rather not have duplicated tf buffers. So I'm going to pursue why the subclass didn't work and/or adding a new method in tf that would give access to the underlying tf2::Buffer without subclassing.

#1215 (comment)

[wjwwood]

Closing since I believe this was worked around in ros/geometry#163. Please comment if that's not the case.

[IanTheEngineer]

Thanks @wjwwood! I meant to close it myself. I believe this PR is unneeded due to the accessor method added to ros/geometry#163 Thanks for the reviews :)