None values in lattice_vectors

[CasperWA]

It seems from the definition of lattice_vectors that while the values may be None, they should be it for a vector as a whole.

This should be checked in the models by a pydantic validator.

[ml-evs]

Relevant part of the spec:

This property MUST be an array of dimensions 3 times 3 regardless of the elements of dimension_types. The vectors SHOULD by convention be chosen so the determinant of the lattice_vectors matrix is different from zero. The vectors in the non-periodic directions have no significance beyond fulfilling these requirements.
All three elements of the inner lists of floats MAY be null for non-periodic dimensions, i.e., those dimensions for which dimension_types is 0.

Which part indicates that they all have to be null? It seems to me that any non-null values should just be ignored.

[CasperWA]

This part:

• All three elements of the inner lists of floats MAY be null for non-periodic dimensions, i.e., those dimensions for which dimension_types is 0.

But I am still not sure this always falls out to be a complete lattice vector?

Concerning the determinant, we can validate that as well if you want? It would probably make sense, but since it is SHOULD, it also isn't REQUIRED, so it shouldn't raise if not true, but instead maybe return a warning?

[ml-evs]

It's not clear to me that the sentence is stipulating that all values must be null if one is, though it probably should.

I don't have strong feelings about the determinant. We could validate it in the case of dimension_types: [1, 1, 1] but don't think we want to go down the route of validating it in <3D...

[shyamd]

I agree with @ml-evs that it doesn't imply that it has to be null but rather null is only allowed when that direction is aperiodic. We already have a validator for this:

optimade-python-tools/optimade/models/structures.py

Line 643 in 3712c9c

def required_if_dimension_types_has_one(cls, v, values):