fix: correct property introspection in meta.Type

[ehennestad]

Two defects in openminds.internal.meta.Type, both found while building the round-trip test synthesizer in the previous PR in this stack. Neither is currently reachable from the serializer, which is why they have gone unnoticed, but both are reachable by any other consumer of the meta layer — including the traversal core proposed in #69.

isPropertyValueScalar reported scalar properties as non-scalar

A property can be restricted to a scalar in two ways: a mustBeScalarOrEmpty validator, or a (1,1) size declaration. The method only looked for the validator on linked and embedded properties. For every other property it consulted the size declaration instead.

The generated type classes declare almost everything as (1,:) and express the scalar constraint through the validator, so any property holding a primitive value that is restricted to a scalar was reported as non-scalar. openminds.core.miscellaneous.Membership/startDate is one example: it is declared (1,:) datetime {mustBeScalarOrEmpty, mustBeValidDate}, assigning two values correctly fails, and isPropertyValueScalar returned false.

The serializer is unaffected because it only asks about linked and embedded properties, which take the branch that worked.

Both constraints are now checked for every property.

The size branch also raised 'Not implemented.' for any dimension that was neither fixed nor unrestricted, with a comment questioning whether that could happen. A dimension that is not fixed leaves the property unconstrained in size, so that case now reports non-scalar rather than erroring.

isPropertyMixedType errored on a property with no declared class

The method read Validation.Class.Name without checking that a class was declared. A property declared without one produces an empty Validation.Class, and the read fails with Insufficient number of outputs from right hand side of equal sign to satisfy assignment rather than answering the question. openminds.neuroimaging.device.MRIScannerUsage/fieldOfView is declared fieldOfView (1,:) with no class and is registered in LINKED_PROPERTIES with an empty allowed-type list, so it reproduces this today.

A property without a declared class cannot be a mixed type, so the check now returns false. getMixedTypeForProperty had the same unguarded read and now raises OPENMINDS_MATLAB:MetaType:NotAMixedType instead.

Note that the fieldOfView declaration itself looks like a gap in the generated classes rather than something to fix here. It is worth raising separately against the pipeline.

Tests

MetaTypeTest gains three cases. The scalar constraint is pinned twice: once against ommtest.helper.PropertyDeclarationFixture, a small class added for this purpose that carries one property of each declaration shape, and once against Membership/startDate so the behaviour is also pinned for a real generated type.

The no-declared-class case is tested only against the fixture. Testing it against MRIScannerUsage/fieldOfView would tie the test to a schema anomaly that may well be corrected upstream, at which point the test would start failing for the wrong reason.

The round-trip synthesizer added in the previous PR inspected validators itself to work around the scalar defect. That workaround is removed here, which also serves as a check that the fix covers the case it was written for.

🤖 Generated with Claude Code