Michel Hidalgo

Also, @ros2/team to kickstart the discussion.

> isn't this exactly the definition of re-entrant? Whoops, yeah. Updated the description. > you can also have problems if they share global state, but we probably want to avoid...

> What I meant was talking about data structures that contain their own locking or similar mechanisms to ensure they are thread safe, rather than putting the onus on the...

FYI @clalancette @ivanpauno

I'm not sure I follow. The purpose of `rcutils_error_handling_scope_t` is not to store anything, but to aid changing error handling macros (e.g. `RCUTILS_SET_ERROR_MSG`) behavior whenever errors occur while handling another...

> The error state is trivially copied, so I'd just recommend storing the error state locally before starting cleanup that might set the error state again. That's precisely the kind...

> The first issue can be solved by changing the contract of rcutils_get_env to take an allocator, allocate space, copy the contents of the environment variable into that space, and...

I can see value in a simpler `ros2 process kill` tool, which is clearly meant to support debugging.

I think we're getting closer @roger-strain. First of all, thank you for pushing. I spent some time thinking on what you wrote, in comments and in the document itself. We...

> One question I have on this is, do we expect there to be any `launch_ros.description.Executables` other than `Node`s? I think one reason I had steered toward the `ExecutableNode` approach...