Transformations require depends on

[jokasimr]

I noticed that some targets of depends_on are not NXtransformations.
I don't think that is correct according to the nexus standard.

In any case, it causes problems in our transformation chain resolution because we expect that a target of a depends_on has a depends_on attribute and not a depends_on dataset, and non-NXtransformation components generally don't have a depends_on attribute, even if they might have a depends_on dataset.

Specifically, the detector component in the Estia nexus files depends on the /entry/instrument/multiblade_detector/transformations/detector_rotation entry which is a NXpositioner and does not have a depends_on attribute.

It does however have other attributes (transformation_type and vector) that indicates it was really intended to have been a NXtransformation but was mislabeled.

To catch this issue and similar I'm adding a requirement that any target of a depends_on has to be a NXtransformation.

[jokasimr]

I think I might have misunderstood some things 🤔

There's probably still an issue here somewhere, but my interpretation is wrong.

NXtransformation does not exist, only NXtransformations (plural).

[jokasimr]

Okay, I think the correct fix here is to say that the target of any depends_on (field or attribute) must itself have a depends_on attribute, otherwise the chain is broken.
Is this correct?

[SimonHeybrock]

Okay, I think the correct fix here is to say that the target of any depends_on (field or attribute) must itself have a depends_on attribute, otherwise the chain is broken. Is this correct?

Yes, chains should always end in depends_on='.'.

[SimonHeybrock]

As you said yourself, this is not correct. The transformations themselves are typically just fields, not groups, so they do not have an NX_class.

[jokasimr]

Superseded by #71