WIP: Allow to pass existing TimeSource via NodeOptions

[roncapat]

Addresses #2659 (task 1).

This allows to put an existing TimeSource in NodeOptions so that a Node in the same process space can avoid re-creating a TimeSource.

For example, the TimeSource may be extracted via rclcpp::Node::get_node_time_source_interface() from an existing node, and passed via NodeOptions to all the others that share the process space.

This is, for example, how I tested it locally in an application of mine.

Main benefit: a single subscriber can be shared when using use_sim_time: better scaling for large rclcpp-composed simulations.

[fujitatomoya]

@roncapat is this still in draft / WIP status?

[roncapat]

@fujitatomoya works nicely for me locally, but I'd like at least a review to understand what are your opinion on this / if I need to change something.

[fujitatomoya]

@roncapat thanks for the PR and feedback.

a several comments for now, i think we need to add more unit tests to verify this feature.

• if the TimeSource is shared with use_clock_thread_ disabled, this main Node must be spinning to update the clock for everybody else, otherwise other nodes sharing TimeSource cannot get the update for the simulation clock information. user application must understand this condition.
• i think if the TimeSource is shared, we cannot reset this in the subordinate Nodes during destruction. only main Node can do the reset once the TimeSource is no longer required to keep?
• The subordinate Nodes will not have <nodename>/use_sim_time parameter neither clock subscription QoS parameters, but only main Node. i think this is also important for user to know, i would suggest doc section in the code and ros2_documentation.
• unit test required to make sure TimeSource is shared within multiple Nodes instances w/ and w/o simulation time. (currently only copy assignment test is added.)
• LifecycleNode support is missing? or is this on purpose for any constraints?

[fujitatomoya]

this actually tries to insert the same element node_cloeck_ if TimeSource is not specified in NodeOption, but that is just fine since std::unordered_set checks the hash to avoid duplication.

[roncapat]

Would you prefer to leave it as it is, maybe with a comment, or make some kind of check to avoid this tentative re-insertion instead?

[roncapat]

Thank you for the useful comments. I will surely try to satisfy every request, starting with LifecycleNode support for this option.

Two quick replies:

• if the TimeSource is shared with use_clock_thread_ disabled, this main Node must be spinning to update the clock for everybody else, otherwise other nodes sharing TimeSource cannot get the update for the simulation clock information. user application must understand this condition.

Do you mean by documentation or via a console warning?

I think if the TimeSource is shared, we cannot reset this in the subordinate Nodes during destruction. only main Node can do the reset once the TimeSource is no longer required to keep?

I think not, the shared_ptr of each node gets just "unlinked" from the TimeSource object, i.e. reference counting descreases, but the object itself gets destroyed at the last reset() only.
std::shared_ptr::reset() on cppreference.com:

Releases the ownership of the managed object, if any. After the call, *this manages no object. 
Equivalent to shared_ptr().swap(*this);

If *this already owns an object and it is the last shared_ptr owning it,
the object is destroyed through the owned deleter.