directive to apply omitempty on nested fields

[suhabe]

I'm getting this error when using the @genqlient directive on the following mutation and schema:

Error:
client.graphql:339: for is only applicable to operations and arguments

Mutation:

# @genqlient(for: "AddAuthorInput.name", omitempty: true)
mutation AddAuthorWithPost($author: AddAuthorInput!) {
    addAuthor(input: [$author]) {
        author {
            id
        }
    }
}

Schema:

input AddAuthorInput {
	name: String
	posts: [PostRef!]
}

input PostRef {
	postID: ID
	postUuid: String
	title: String
	subPosts: [PostRef!]
}

Is the directive above written the correct way? I tried following the explanation at: https://github.com/Khan/genqlient/blob/main/docs/FAQ.md#-nullable-fields

Also, is possible to add a directive on a nested subfield? So for example, something like:
# @genqlient(for: "AddAuthorInput.posts.title", omitempty: true)

Is it possible to add a directive on a field of struct type that's nested in the top level type? So for example, something like:
# @genqlient(for: "PostRef.title", omitempty: true)

[suhabe]

Also, when the directive is applied to the entire query, I get this error:
client.graphql:339: omitempty may only be used on optional arguments

# @genqlient(omitempty: true)
mutation AddAuthorWithPost($author: AddAuthorInput!) {
    addAuthor(input: [$author]) {
        author {
            id
        }
    }
}

I was expecting the directive to apply omitempty everywhere in the query, including deeply nested fields of the input variable, regardless of whether they are optional or not.

[suhabe]

By the way, this functionality is important when using Dgraph for the use case described here: https://discuss.dgraph.io/t/best-way-to-avoid-duplicate-xid-found/10369

Dgraph expects a block of a mutation's input to include only the XID field and have no other fields set in order for the block to be used as a reference to another block.

[suhabe]

After investigating further, I found that the best way to use genqlient with Dgraph is to have omitempty on every generated field and always make the field type a pointer (e.g. *int instead of int). The reason is that Dgraph relies on distinguishing between a zero value and no value.

Is there a global configuration option to enable these two changes to the generated code?

[benjaminjkraft]

I think use_struct_references, added in #155, is what you're looking for! It's not in a release yet (see #163) but you can get it on latest main branch.

BTW, the directive is not written quite right, see #151.

[kevinmichaelchen]

Also, is possible to add a directive on a nested subfield? So for example, something like:
# @genqlient(for: "AddAuthorInput.posts.title", omitempty: true)

This appears to be unsupported as of v0.5.0.

Hasura generates structs with lots of fields (Eq, Gt, Gte, IsNull, Lt, etc) but sometimes none of those fields have omitEmpty tags. The problem is there's an extra layer of nesting in between.

Trying to access a nested field results in this error: for must be of the form "MyType.myField"