Sphinx autoapi also now only shows subpackages 1 level deep. This prevents us having a huge long list. Also don't show private or special members. Also show imported members so ifcopenshell.file and ifcopenshell.entity_instance works in docs too.
On older numpy versions, np.ndarray was less forgiving and wasn't allowing passing 1 argument instead of required 2.
And turned out numpy doesn't yet have typing for shapes (https://github.com/numpy/numpy/issues/16544), so all matrices and other shapes specified as `npt.NDArray[np.float64]`.
Fixed type discrepancies for `get_edges` and `get_faces` and also had to fix `import_ifc` as Blender apparently has problems with storing np.int32 in custom attributes (https://projects.blender.org/blender/blender/issues/121072), tested that Blender is okay with np.int32 in other cases we had (addressing BMesh.verts[i] where `i` is np.int32).
ifcpatch ConvertLengthUnit - change used units from plural to singular names, to make it consistent across the api
more test coverage - test converting to more units and back from them
the error is below, not sure how I haven't noticed it before 🤔
assert subject.format('imperial_length(1, 1)') == "1'"
> raise VisitError(tree.data, tree, e)
E lark.exceptions.VisitError: Error trying to process rule "imperial_length":
E
E cannot access local variable 'output_unit' where it is not associated with a value
Tried to reduce amount of IFC attributes retrieved by reusing already retrieved values.
E.g. previously get_nest() was retrieving same attribute 3 times on each call, now it does it only once and it's 2 times faster (2.4 ms -> 1.2 ms).
Checking start/end dates are done very, very often in date calculations (often looping through every day in a date range). So I'd prefer more opaque code but less overhead.
Andrej730