Handle RMW_DEFAULT_DOMAIN_ID.

[hidmic]

Precisely what the title says. Connected to ros2/rcl#689.

[ivanpauno]

why is this needed?
there's similar logic in rcl to avoid passing the default domain id IIRC

[ivanpauno]

See ros2/rcl#689 (comment)

[wjwwood]

nitpick: line wrap looks unnecessary.

[hidmic]

why is this needed? there's similar logic in rcl to avoid passing the default domain id IIRC

No there isn't. Or there shouldn't be. The bug that ros2/rcl#689 fixed was forcing a 0 when ROS_DOMAIN_ID=. The point of having an invalid domain ID as default is for rmw implementations to decide what to do with it. We can discuss whether we want to have such sentinel or not.

[hidmic]

CI up to rcl and test_rmw_implementation against all RMW implementations:

• Linux
• Linux-aarch64
• macOS
• Windows

[ivanpauno]

why is this needed? there's similar logic in rcl to avoid passing the default domain id IIRC

No there isn't. Or there shouldn't be. The bug that ros2/rcl#689 fixed was forcing a 0 when ROS_DOMAIN_ID=. The point of having an invalid domain ID as default is for rmw implementations to decide what to do with it. We can discuss whether we want to have such sentinel or not.

It has historically been used to determine in rcl if ROS_DOMAIN_ID variable has to be check or not (see https://github.com/ros2/rcl/blob/72965a0e3fa9ad0af6d0fe301a1d02edc746a1fb/rcl/src/rcl/init.c#L146-L147).
It has never been used for letting the rmw implementation decide what to do with the default domain id (same in foxy, eloquent, dashing, etc).

Not addressing ros2/rcl#689 (comment) will break rmw implementations not maintained by us.

[hidmic]

It has historically been used to determine in rcl if ROS_DOMAIN_ID variable has to be check or not

True. In rcl, where it happens to be the same number. This patch is simply making sure that rmw_init() with default options as provided by the implementation does not fail.

Not addressing ros2/rcl#689 (comment) will break rmw implementations not maintained by us.

If an implementation does not handle RMW_DEFAULT_DOMAIN_ID, it is broken already.

---

IMHO the issue is that both domain ID defaults are conflated. rcl has a default to decide whether it has to check the environment or not, and rmw has a default with a sentinel value. If no ID is passed to rcl and the environment does no provide one either, the rmw default should be honored.

We can discuss the rmw default though. It could be 0, sure, if we think no rmw can benefit from knowing if that ID is the default and not a user-defined one that happens to have the same value.

[ivanpauno]

If an implementation does not handle RMW_DEFAULT_DOMAIN_ID, it is broken already.

Currently, rcl never passes RMW_DEFAULT_DOMAIN_ID, so it's not already broken.

[hidmic]

Currently, rcl never passes RMW_DEFAULT_DOMAIN_ID, so it's not already broken.

It is if the rmw can't handle it. And that's regardless of whether it's most common downstream package guards it from that situation or not.