Add support for rosidl::Buffer in rosidl Python path for rclpy

[nvcyc]

Description

This pull request adds rosidl::Buffer support for uint8[] fields in the Python code generation path.

This pull request consists of the following key changes:

• Python message setters (_msg.py.em): For uint8[] fields, the generated setters now allow assigning from rosidl_buffer.Buffer or plain array.array. Existing code that sets msg.data = array.array('B', ...) continues to work unchanged.
• Python/C conversion (_msg_support.c.em): Messages with uint8[] fields use the sequence's is_rosidl_buffer flag to determine the conversion path:
  • Python to C: For non-CPU rosidl_buffer.Buffer objects, calls _get_buffer_ptr() and sets the C++ Buffer pointer plus is_rosidl_buffer = true so the RMW path uses the existing buffer without copying. For array.array objects, uses the normal byte-copy path.
  • C to Python: When the RMW deserializes into a vendor-backed buffer (is_rosidl_buffer == true), calls _take_buffer_from_ptr() to wrap the pointer in a Python Buffer (taking ownership) and set it on the message. CPU-based data is converted to array.array as before.

Is this user-facing behavior change?

This pull request does not change existing rclpy behavior. Existing code using array.array for uint8[] fields continues to work. The rosidl_buffer.Buffer type is only encountered when a subscriber receives data from a non-CPU buffer backend (and has opted in via acceptable_buffer_backends).

Did you use Generative AI?

Yes. Claude (claude-4.6-opus) via Cursor was used to assist with creating an initial prototype version of the changes contained in this PR.

Additional Information

This PR is part of the broader ROS 2 native buffer feature introduced in this post.

[hidmic]

First pass. Ownership is bit fuzzy.

[hidmic]

@nvcyc why not simply return true? Is this missing some cleanup?

[nvcyc]

Returning here would skip iterating through the rest of the members in the same message, so we need goto @(member.name)__done; to make sure that it goes to the next member within the loop starting at:

rosidl_python/rosidl_generator_py/resource/_msg_support.c.em

Line 208 in 7901121

@[for member in message.structure.members]@

[hidmic]

Ah, I missed EmPy's end for statement. Thanks for the clarification

[hidmic]

@nvcyc meta: I know this code predates your implementation but wow. I can see looping over pop instead of clear having a massive impact in runtime performance.

[nvcyc]

Agreed. We can replace this part with either

• PyObject * ret = PyObject_CallMethodNoArgs(field, "clear"); but this only works with Python 3.13 or higher (reference)
• PySequence_DelSlice(field, 0, length) that should work in any Python version