Semantics of AMENT_PREFIX_PATH variable?

[traversaro]

In #1358 and #1359, for supporting ROS2 the feature of reading from the AMENT_PREFIX_PATH variable was added, treating it as it contained the same directories contained in ROS_PACKAGE_PATH .

However, debugging another problem, I investigated this and it seems that while ROS_PACKAGE_PATH contains the directories in the form <install_prefix>/share, AMENT_PREFIX_PATH instead contains directories in the form <install_prefix> .

I am not sure what is the best thing to do in this case, but I wonder if the share subdirectory should be added to the values obtained from AMENT_PREFIX_PATH.

Related issue: robotology/robotology-superbuild#832 .

[jcarpent]

@traversaro Thanks for raising this issue.
As I'm not a ROS expert, I let @wxmerkt providing his thoughts.
@wxmerkt Could you provide your feedback on this issue?

[jcarpent]

I've just checked robotology/robotology-superbuild#832, and the solution sounds fine for me.
@traversaro Could you translate your change to Pinocchio?

[traversaro]

Sure, let's wait to see if @wxmerkt has any feedback as I will be interested in it even for iDynTree/iCub's models. I wonder if @olivier-stasse may have any comment as he is the one that originally created #1358 .

[wxmerkt]

Hi @traversaro and @jcarpent,
Apologies for the slow response to this; yesterday and today are filled with meetings. I haven't used ROS2 myself and hence not tested the AMENT_PREFIX_PATH - but what you wrote about changing to <install-prefix> sounds familiar. It may be worth testing in a Docker with a commonly used/binary-available robot description package to confirm any fixes

[olivier-stasse]

Dear all, the solution implemented in robotology/robotology-superbuild#832 sounds fine for me too.

[jcarpent]

@proyan Could you handle it?

[proyan]

Fixed with #1568.