Archives referencing skipped packages' local_setup.bat about ci

[dirk-thomas]

The requested change colcon/colcon-core#13 made this visible. I think letting the user know about it makes sense. Maybe we should ignore those packages on Windows (e.g. by placing an IGNORE marker file in their directories)?

[dhood]

Yeah I agree that for source builds this visibility makes sense, it's just misleading in the specific case of the packaging job.

The packaging job never used to install anything in the share directory for packages that got skipped by CMake.

e.g.:

$ cd ~/Downloads/ros2-osx-beta1
$ ls share | grep pendulum
pendulum_msgs

$ cd ~/Downloads/ros2-osx-colcon-bash-fix/
$ ls share | grep pendulum
pendulum_control
pendulum_msgs

Do you think it is possible to restore that behaviour (technically)? If so I think that would be preferred over adding ignore markers (the skip in cmake is nicely self-contained and self-describing and so should be sufficient)

[dirk-thomas]

Do you think it is possible to restore that behaviour (technically)?

What exact behavior do you want to "restore" / are you looking for? Revert the referenced patch and don't output a message (until COLCON_TRACE is set)?

[dhood]

The behaviour I'm referring to is more about the cause of the problem, i.e. that the packaging job is installing stuff for skipped cmake packages (where it didn't with ament_tools)/that skipped cmake packages get listed in the generated prefix.bash file. If we address that, we wouldn't need to change the behaviour of outputting the message or not, because those files wouldn't be looked for to start with. But I am not sure what "underneath" has caused this change in behaviour.

[dhood]

Or, put another way:

Why is the prefix.bash on OS X / the prefix.bat on Windows trying to source files that never have existed? That is my understanding of what is causing that output message. Contrary to colcon/colcon-bash#3, it is not because a directory has been deleted; the scripts it's looking for never got installed.

[dirk-thomas]

So when a package specific setup file doesn't exist you want the prefix script to not complain about it. For this case that makes sense. But that will also apply to other cases: e.g. you build and install a package B but not its dependency A. Then the resulting prefix script should not print a message that A isn't present.

We can certainly change it that way - we just have to be certain that this is the way we want it to be in all the cases which have the same "state".

[dirk-thomas]

Currently the build tool also checks that these files actually exist (https://github.com/colcon/colcon-core/blob/f093f0bcffe2f0d1ea66d434e40fcad95638cee0/colcon_core/shell/sh.py#L84-L96) before generating the prefix script. So I am not yet sure how this gets into the state and how we could distinguish the case described.