Would it be better if there was a way to specify this as part of the mini-language itself?

It gets a little tricky. Syntactically what is the difference between MyEntity( someAttribute ) and someAttribute( someSubAttribute )?

how about something like MyEntity: someAttribute

The plan was to simply have this pass through GraphParser. So if you are asking me if I want to change the GraphParser grammar, no not really.

If we don't change GraphParser, there would need to be a pre-analysis of this text to determine if there is a =n entity "prefix". I'm not against it, just not sure what it gains us. But if others think its "nicer", I'm fine with it.

I should mention, I had planned to change this a bit to allow either name or class:

@interface NamedEntityGraph {
    ...
    String rootEntityName() default "";
    Class<?> rootEntityType() default void.class;

Mainly because I know most users will prefer to use the class reference. I will agree that given just #rootEntityName your suggestion is nice.

6 in one, half dozen in the other imo

how about something like MyEntity: someAttribute

So I did end up adding this to the parser syntax. But in practice, for the named graphs, this is only needed for package-level definition. For entity-level I just required that the annotation be placed on the root entity for the graph (in keeping with JPA variant) and so we always know the entity type there.

@Entity
@NamedEntityGraph( graph="title, isbn")
class Book {...}

@NamedEntityGraph( name="book-title-isbn", graph="Book: title, isbn")
package org.hibernate.test;

For entity-level I just required that the annotation be placed on the root entity for the graph (in keeping with JPA variant) and so we always know the entity type there.

Perfect!

This is indeed nice.