Backport some colcon features

[mathias-luedtke]

Partial backport of #361

• align the code/behavior between both version
• shellcheck :)
• fix ABI check feature not functional on ROS Indigo? #258, fix ABI check feature failing without obvious output but "ERROR: no symbols info in the ABI dump" #322
• backport of CMAKE_ARGS (highly perferred over CATKIN_CONFIG)

In addition it should simplify the review.

CoS (Condition of Satisfaction, suggested by @130s)

• Code review.
  • @miguelprada
  • @130s
• (Locally) verify the change.
  • @130s ABI error ABI check feature not functional on ROS Indigo? #258 -> Ongoing in WIP: Do NOT Merge: [CI] Testing ABI check potential fix  130s/openni2_camera#2

[mathias-luedtke]

Due to the many renames it is best to review each commit individually.

[miguelprada]

I've gone over the proposed changes and, even if I don't fully grasp every little change, nothing stands out for me that could potentially break anything.

Tested in universal_robot and ur_modern_driver, and also in some of our most convoluted internal pipelines (both before and after switching from CATKIN_CONFIG to CMAKE_ARGS), and every job passed.

For my part, it's ok to merge this.

[mathias-luedtke]

@130s: Can you test /review it as well?
These fixes will end up in master/legacy.

[130s]

(CoS suggested is moved to the OP)

[mathias-luedtke]

Indigo is now EOL.

[mathias-luedtke]

@130s: Can we merge this? Indigo jobs started to fail as a result of ros/rosdistro#21284.

[130s]

I confirmed ABI feature issue is fixed with this PR 130s/openni2_camera#2 (comment)

Thanks for the great changes and backporting them!

[130s]

Tagged 0.8.0 mainly in a desire to clearly label a change where a ROS distro is now EOLed but still supported by ici.

[miguelprada]

FYI, I just found a small regression introduced in this PR related to an arguably non-standard use-case (i.e. see #232).

Previously the export command here was masking the (error) return value of catkin_topological_order, and the execution continued even with an empty workspace. After bdd783f, this error is caught and the job fails.

I don't think this should be acted upon, since that use case is not very common (and I'm no longer doing that myself). Furthermore, I'm not sure that trick would make sense anymore when #361 is merged. I wanted to leave this comment here just for future reference, in case I stumble into this ever again.

[mathias-luedtke]

Yes, this slipped through.
We could fix this on demand.
With #361 this problem does not exist anymore, you can just specify TARGET_WORKSPACE.
In addition colcon works with empty source spaces as well.