Store node pointer in publisher and subscription

[nnmm]

This is the successor of #987.

By storing the node pointer in the publisher and subscription, it is possible to check for the correct node pointer being passed to rcl_publisher_fini() and rcl_subscription_fini(). This PR also adds a new return code for this purpose.

[fujitatomoya]

I will go ahead to close this, see #987 (comment) and #987 (comment)

[fujitatomoya]

lgtm, all comments posted on #987

[fujitatomoya]

CI:

• Linux
• Linux-aarch64
• Windows

[fujitatomoya]

@iuhilnehc-ynos can you take another review on this?

i believe that node argument to fini function is by contract to make it clear, we are not removing this. but extra check makes sense to me.

thanks,

[iuhilnehc-ynos]

@fujitatomoya

Storing the node pointer in publisher and subscription just to only check if the node pointer is correct or not in rcl_publisher_fini() and rcl_subscription_fini() is not reasonable to me.

I believe that not storing the node pointer in the entities(rcl_publisher_impl_s, rcl_subscription_impl_s, rcl_service_impl_s, rcl_client_impl_s) is something about the decoupling design.

[fujitatomoya]

I believe that not storing the node pointer in the entities(rcl_publisher_impl_s, rcl_subscription_impl_s, rcl_service_impl_s, rcl_client_impl_s) is something about the decoupling design.

I see that one of the design, but node and publisher / subscription are not exactly decoupled.
adding extra check to see if that is the correct node would make sense?

or are you suggesting something else?