Support filter on expand

[campoy]

It would be very useful to filter what fields are included in an expand operation.

For instance, we could expand only the predicates pointing to values of type Person from a Movie as such:

{
    q(func: type(Movie)) {
        expand(_all_) @filter(type(Person)) {
            first_name
        }
    }
}

This would expand the predicates of type uid or [uid] and show only the nodes associated to the type Person.

[gitlw]

@campoy It seems the syntax expand(type(Person)) is more concise and intuitive. What do you think?

[campoy]

expand(type(Person)) and expand(Person) @filter(type(Person)) are different things, and they're both valid.

Given a type Person as follows:

type Person {
  name: string
  age: int
  best_friend: uid   // points to a Person
  lives_with: [uid]  // holds an array of objects: some of type Person, some of type Pet
  pets: [uid]        // holds an array of objects of type Pet
}

expand(Person)

expand(Person) is replaced by the predicates defined in the Person type.

{
  q(func: eq(name, "Francesc")) {
    expand(Person)
  }
}

Is the same as writing

{
  q(func: eq(name, "Francesc")) {
    name
    age
    best_friend
    lives_with
    pets
  }
}

You can read more on expand(Person) on issue #3903.

expand(X) @filter(type(Person))

This will expand whatever is requested in the expand function, either _all_ or Person. But then only the values of type Person will be kept.

So given a value of Person

{
  "uid": "0x1234",
  "dgraph.type": "Person",
  "name": "Frodo",
  "age": 25,
  "best_friend": {
     "uid": "0x2345",
     "dgraph.type": "Person",
     "name": "Sam",
     "age": 24
  },
  "lives_with": [
      {
        "uid": "0x2345",
        "dgraph.type": "Person",
        "name": "Sam",
        "age": 24       
      },
      {
        "uid": "0x3456",
        "dgraph.type": "Pet",
        "name": "Milou"
      }
  ],
  "pets": [
      {
        "uid": "0x3456",
        "dgraph.type": "Pet",
        "name": "Milou"
      }
  ]
}

The following query:

{
  q(func: eq(name, "Frodo")) {
    expand(_all_) @filter(type(Person)) {
       expand(Person)
    }
  }
}

would:

• expand all the fields in the types associated to "Frodo", in this case just those of "Person"
• filter the predicates to only those of type "Person", therefore dropping name, age, pets, and some of the values in lives_with
• expand all of the predicates in Person for each one of the nodes in pointed by the predicates expanded before

The result would be the following:

{
  "best_friend": {
     "name": "Sam",
     "age": 24
  },
  "lives_with": [
      {
        "name": "Sam",
        "age": 24       
      }
  ]
}

I wonder if this is clear enough. I think I need to spend some time writing more specs for these changes.
For now let's keep them blocked.

[campoy]

Hey @martinmr , could you have a look at this issue?
I think you might be the best person to handle this right now.

[martinmr]

I have a PR for this already but there is a weird bug that I haven't had the chance to debug yet.

[Necmttn]

Hey, @martinmr is there any status update about this issue?

[campoy]

This issue is related to #4399, as expand(_all_) might expand value predicates and filters should apply on them too (or not, to be decided in that issue).