node_coords_in_element() in SquareNedelecFirstKindShapeFunctions and SquareRaviartThomasShapeFunctions do not return coords

[pei-tian]

Bug Description

SquareNedelecFirstKindShapeFunctions.make_node_coords_in_element() and SquareRaviartThomasShapeFunctions.make_node_coords_in_element() each return an inner node_coords_in_element() function. That inner function constructs a Coords value named coords, but does not return it. It therefore returns None instead of the expected value.

Source:

warp/warp/_src/fem/space/shape/square_shape_function.py

Line 817 in 987f70a

def node_coords_in_element(

warp/warp/_src/fem/space/shape/square_shape_function.py

Line 935 in 987f70a

def node_coords_in_element(

Repro Script:

import warp as wp
from warp.fem.space.shape import (
    SquareNedelecFirstKindShapeFunctions,
    SquareRaviartThomasShapeFunctions,
)

wp.init()


def run(shape_type, name: str, device: str = "cpu"):
    shape = shape_type(degree=1)
    node_coords_in_element = shape.make_node_coords_in_element()

    @wp.kernel
    def get_first_node_coords(out: wp.array(dtype=wp.vec3)):
        out[0] = node_coords_in_element(0)

    out = wp.empty(1, dtype=wp.vec3, device=device)

    try:
        wp.launch(get_first_node_coords, dim=1, inputs=[out], device=device)
        wp.synchronize_device(device)
    except RuntimeError as error:
        print(f"{name}: BUG REPRODUCED")
        print(error)
        return

    print(f"{name}: OK")
    print(out.numpy().tolist())


run(SquareNedelecFirstKindShapeFunctions, "Square Nedelec N1")
run(SquareRaviartThomasShapeFunctions, "Square Raviart-Thomas RT")

Expected output:
For a valid node index such as 0, both functions should return its coordinate instead of None.​​​​​​

Square Nedelec N1: OK
[[0.5, 0.0, 0.0]]
Square Raviart-Thomas RT: OK
[[0.5, 0.0, 0.0]]

Actual output:

Square Nedelec N1: BUG REPRODUCED
Error while parsing function "get_first_node_coords" at /tmp/repro_square_n1_rt_node_coords.py:17:
    out[0] = node_coords_in_element(0)
;array_store() value argument type (builtins.NoneType) must be of the same type as the array (vec3f)

Square Raviart-Thomas RT: BUG REPRODUCED
Error while parsing function "get_first_node_coords" at /tmp/repro_square_n1_rt_node_coords.py:17:
    out[0] = node_coords_in_element(0)
;array_store() value argument type (builtins.NoneType) must be of the same type as the array (vec3f)

Notes:
Both functions construct coords but do not return it from the node_coords_in_element().
Adding return coords makes the test complete successfully.

System Information

• Warp: 1.15.0rc3
• JAX: 0.10.2
• Python: 3.12
• OS: L4T R38.2
• Device: NVIDIA Jetson Thor
• CUDA: 13.0.0

[shi-eric]

This is addressed by 46c4595.

The change returns the computed coordinates from the device functions created by
SquareNedelecFirstKindShapeFunctions.make_node_coords_in_element() and
SquareRaviartThomasShapeFunctions.make_node_coords_in_element().

Public API behavior:

• Python scope: No constructor or method signatures changed.
• Kernel scope: The generated device functions now return valid Coords values,
  so their results can be used inside Warp kernels instead of being inferred as
  None.

Test coverage:

• No tests were added or modified. This leaves the exact return-value regression
  without committed coverage.
• A direct behavioral probe of both shapes and all four valid node indices passed
  on CPU and CUDA and produced the expected edge-midpoint coordinates.

Spotted Improvements:

• Add focused regression coverage in warp/tests/fem/test_fem_shape.py for the
  returned coordinates of both square vector shape functions.