Feature request

Feature description

Currently and by default, most DDS-based ROS middlewares lookup and choose a suitable network interface on the machine that's hosting them. It makes for easy configuration of multi-machine ROS systems, but it may also result in undesired cross-talk. This is specially true in CI setups, where multiple tests running in parallel may inadvertently interact with each other.

Isolation via ROS_DOMAIN_ID, though possible, poses a maintenance burden as IDs must be carefully allocated and assigned by other means. Additionally, the set of available domain IDs is a rather limited one (see RTI Connext documentation, this relevant eProsima Fast-RTPS ticket and the RTPS standard).

Both burden and limitation are acceptable when a ROS system is intended to span several machines in a network, but when that's not the case it's an unnecessary source of problems. Providing an easy way for users to force DDS-based middlewares to use loopback would preclude all of these issues.

Implementation considerations

From a UX perspective, negative ROS_DOMAIN_IDs could provide the means to externally enable this restriction (credits to @dirk-thomas). ROS_DOMAIN_IDs sign would act as a flag:

• if it's a positive number, a suitable network interface will be chosen (just like it is today);
• if it's negative, loopback will be used instead.

Meanwhile, the actual domain ID would become the magnitude (i.e. the absolute value) of the ROS_DOMAIN_ID number. For instance, ROS_DOMAIN_ID=-1 would read as "Use 1 for a domain ID over loopback" while ROS_DOMAIN_ID=1 would read as "Use 1 for a domain ID over a suitable network interface".