tf2 in groovy defines unnecessary CATKIN_DEPENDS (e.g. on rospy) about geometry2

[tfoote]

fixed in groovy 6c8ea39

[tkruse]

I also miss boost as DEPENDS, used in https://github.com/ros/geometry_experimental/blob/hydro-devel/tf2/include/tf2/buffer_core.h
On the other hand, I am not sure whether console_bridge should be exported as dependency, given it's not used in the public headers.

I could go on.

E.g.
tf2_bullet: misses geometry_msgs and tf2 as exports
tf2_geometry_msgs: tf2_ros is exported but not used
tf2_kdl: misses geometry_msgs (both as find_package as well as in catkin_depends), and tf2 as catkin_depends

Correct me if I am wrong, but a review process for the catkin_package macro should go like this:

1. Check the package author did not do something unorthodox and installs headers from somewhere else than /include//*.h (if he did, this should probably in itself be a style ticket, in conventional packages)
2. Look what is used in the /include//*.h files (and only those)
3. Make sure all packages mentioned there are declared in catkin_package as DEPENDS or CATKIN_DEPENDS
4. Anything in catkin_package not mentioned in any of the headers should go, unless it's something exceptional (genmsg, not sure what else would count)

The more packages get this worng, the more people will be confused and annoyed about catkin. I think except for packages catkinized by maybe Dirk or Vincent, most packages get the DEPENDS and CATKIN_DEPENDS wrong. Maybe that is a hint that those keywords are not helpful and self-explaining, but confusing?

[tkruse]

Note I am not just talking about the tf2 packages here, I am talking about hydro in general. Sampling over at a few catkin packages in hydro, I see many packages flawed that way, and that is a major problem in the general introduction of catkin.