Move to federated model for rosdep (ros ticket #1495) about rosdep HOT 2 CLOSED

[mquigley] I have implemented a good portion of this now. However, the lookup order I'm implementing is a bit different:

As you state, things would be cleaner if all code was in properly released stacks. That may happen for WG code, but I don't really see that happening for lab contributions without a sea change in the way labs operate IMHO.

In light of this, my current approach is to try to not force all repos to have stacks (even degenerate ones), but instead to follow this lookup order:

rosdep satisfy foo

1. go to the "foo" package
2. look up all its package dependencies.
3. for each package dependency, go to its "owner", where "owner" is either the stack containing it, or a top-level entry in ROS_PACKAGE_PATH (e.g., a svn checkout of a repo).
4. if its owner has a /rosdep.yaml, load and merge it with the working set of rosdeps.
5. detect and abort on merge collisions, as you state.

The only change is that rosdep.yaml files living in top-level ROS_PACKAGE_PATH entries are treated as if they were stack-based rosdep.yaml files. This is purely so that people don't have to do stacks if they don't want to (e.g., sail-ros-pkg can have a rosdep.yaml in its /trunk)

OK by you?

from rosdep.

[kwc] Awesome, and OK by me. As you note, it's hard for external repos to stack everything and the fake stack.xml files for a repo seemed kludgey. I think you came up for a good "c" solution that circumvents the dependency issue well w/o requiring stack.xml files or additional rosdep.yaml syntax.

from rosdep.