unnecessary to externalize jakarta.data.metamodel.impl package

[njr-11]

This issue is for after version 1.0.

We shouldn't need to externalize an implementation package, jakarta.data.metamodel.impl, and should easily be able to avoid it by offering a static method to obtain the instance for each attribute type.

Instead of doing this,

import jakarta.data.metamodel.impl.AttributeRecord;
import jakarta.data.metamodel.impl.SortableAttributeRecord;
import jakarta.data.metamodel.impl.TextAttributeRecord;
...
@StaticMetamodel(Person.class) {
    ...
    SortableAttribute<Person> ssn = new SortableAttributeRecord<Person>(SSN);
    TextAttribute<Person> firstName = new TextAttributeRecord<Person>(FIRSTNAME);
    TextAttribute<Person> lastName = new TextAttributeRecord<Person>(LASTNAME);
    Attribute<Person> address = new AttributeRecord<Person>(ADDRESS);
}

we could have something much simpler that doesn't directly access any impl package things,

...
@StaticMetamodel(Person.class) {
    ...
    SortableAttribute<Person> ssn = SortableAttribute.of(SSN);
    TextAttribute<Person> firstName = TextAttribute.of(FIRSTNAME);
    TextAttribute<Person> lastName = TextAttribute.of(LASTNAME);
    Attribute<Person> address = Attribute.of(ADDRESS);
}

The above does not require an impl package to be externalized and appear in the spec or javadoc at all. The jakarta.data.metamodel.impl package and the classes within it could then be deprecated, and in a future version moved from external visibility to hidden.

[KyleAure]

If in the future we intend not to externalize jakarta.data.metamodel.impl won't that break users?

data/api/src/main/java/module-info.java

Line 1065 in b1e7c84

exports jakarta.data.metamodel.impl;

If this gets changed to opens or is removed entirely then the record implementations won't be available at compiletime.
Or is the idea we just stop documenting the constructor pattern but still have them available?