Missing a way to statically reference a sort by an embedded or associated property

[dstepanov]

I have an embedded entity specs and I want to sort by it's property:

Sort.asc( "specs.weight")

This works but I cannot reference it using the metamodel because navigate always returns an expression:

NumericExpression<Train, Double> navigate = _Train.specs.navigate(_TrainSpecs.weight);

Maybe there should be a way to create a path:

_Train.specs.path(_TrainSpecs.weight).asc();

[gavinking]

Ahhhh, this is probably on oversight in the new APIs.

asc() and desc() or declared by SortableAttribute, and not by, say, ComparableExpression.

We should look into this one.

[gavinking]

Ahhhh, this is probably on oversight in the new APIs.

Well, actually, not so fast ... in fact in JPQL we generally don't allow sorting by anything other than a value which already occurs in the select list. (There's actually a good reason for this prohibition, namely that it would interact a bit badly with distinct when translated to SQL.)

So the current design is deliberately restrictive. I agree it's too restrictive, but I'm not sure what a design with the correct level or restrictiveness would look like.

[njr-11]

A related problem is that in addition to needing a metamodel constant to supply to Sort.asc("specs.weight"), you would also need the constant for,

@By("specs.weight") LessThan<Double> maxWeight

and

@OrderBy("specs.weight")

and

@Select("specs.weight")
Optional<Double> weightOf(String trainId);

The Data 1.0 spec/metamodel included a solution to this, which is to have the _ character in the field name delimiting the attribute names of hierarchical structures or relationships, such as embedded classes
. This allowed you to write,

Sort.asc(_Train.SPECS_WEIGHT)

or

_Train.specs_weight.asc()

Data 1.0 Javadoc includes an example Entity with Embeddable and metamodel of the above form,
https://jakarta.ee/specifications/data/1.0/apidocs/jakarta.data/jakarta/data/metamodel/staticmetamodel

Depending on how well the NavigableAttribute pattern turns out (It still needs a lot of thinking about / work), I was wondering whether we would de-emphasize the existing pattern from Data 1.0.

[njr-11]

#1283 adds a TCK test for this (using the Data 1.0 approach for now) that can be switched over to NavigableAttribute after we get it to cover this.