RosSubChannelElement should be created as a shared_ptr

[guihomework]

completely fixes #61

[meyerj]

The member variable tmp is already a shared pointer (actually a boost::intrusive_ptr) in rtt_rostopic_ros_msg_transporter.hpp:257, so the change in this PR should actually not make a difference.

Only intrusive_ptrs allow assignment from raw pointers like in this case. shared_ptrs with an external reference counter indeed would have to be constructed explicitly before the assignment to tmp (or using make_shared).

Did you observe a difference after having applied this patch? Actually the test added in #66 (only in branch toolchain-2.9) should verify that the subscriber was actually shutdown after the port has been disconnected. The toolchain-2.9 branch is fully backwards-compatible and you can also use it for toolchain 2.8 in ROS indigo.

[guihomework]

Dear @meyerj I will re do my own tests on 2.8.3 first and then on 2.9 if you say it should not change anything, because without my patch on 2.8.3 I could still see the rtt node being displayed as subscriber with rostopic info.
In order to test the code change I know I have to clean completely my build space since the typekits must be rebuilt to integrate the changes in this header file.
stay tuned

[guihomework]

@meyerj
After cleaning my build space AND devel space (using catkin tools) it this time worked.
I had cleaned only my build space only , applying or removing my patch and I could definitely see changes, now with the devel space cleaned as well your patch is sufficient. I suppose the fact that code is in headers affects building with catkin somehow.

sorry for the noise.

[meyerj]

I suppose the fact that code is in headers affects building with catkin somehow.

Yeah, that's kind of tricky and the main reason why orocos-toolchain/rtt#85 has been introduced in toolchain 2.9. Actually you can keep your build-space to save some time, but you should always clear the install-space (and devel-space in case you are using catkin) before rebuilding without this include directories patch because otherwise your compiler will prefer the installed headers from previous builds.

[guihomework]

Okay, it makes sense now.
I am saving so much time building my workspace without rtt-ros-integration (since the released version works for me).
thanks for the nice support.