Just saving here, will clean up later:
-
Inline empty arrays: a zero-length array stored inline (data: [], as Python asdf writes for np.zeros(0) with inline storage) fails in the constructor at stack(data) (ArgumentError: stack on an empty collection is not allowed) before any stride logic runs. The zero-size fix here covers block-backed nodes only.
-
0-d ndarrays: shape: [] (permitted by the standard) yields default strides of length 1, failing the length(shape) == length(strides) check; NumPy gives () strides for 0-d arrays. Needs constructor, materialization, and writer support together.
Just saving here, will clean up later:
Inline empty arrays: a zero-length array stored inline (
data: [], as Python asdf writes fornp.zeros(0)with inline storage) fails in the constructor atstack(data)(ArgumentError: stack on an empty collection is not allowed) before any stride logic runs. The zero-size fix here covers block-backed nodes only.0-d ndarrays:
shape: [](permitted by the standard) yields default strides of length 1, failing thelength(shape) == length(strides)check; NumPy gives () strides for 0-d arrays. Needs constructor, materialization, and writer support together.