Unregister typesupport rather than delete

[brawner]

There was a use-after-free that I discovered while adding fault injection tests to rcl_action over at ros2/rcl#730. Somehow rcl_action's typesupport for its client and subscription were being duplicated. When it was cleaned up during this fail block in create_subscription, it would try to delete it again over in rmw_client_fini. Instead, I believe it needs to unregister it, which also deletes it, so that it won't get deleted twice. rmw_client.cpp already does this with unregister rather than deleting the pointer directly.

Before-and-after unit tests will have to wait until tomorrow.

Signed-off-by: Stephen Brawner brawner@gmail.com

[brawner]

Builds and tests of the following packages rcutils, rcl_logging_spdlog, rosidl_typesupport_c, rosidl_typesupport_cpp, rcl, rcl_action, rcl_lifecycle, rosidl_runtime_c, rmw_dds_common, rmw

• Linux
• Linux-aarch64
• macOS
• Windows

[hidmic]

@brawner I think #419 fixed this (among other things), no?

[brawner]

I think you're correct.

[hidmic]

@brawner are you keeping the branch for anything in particular?

[brawner]

No, deletebot didn't want to delete it I guess.