Dataloader's documentation

[cglacet]

The dataloader documentation is quite unclear in the current state, especially this piece of code:

class User(graphene.ObjectType):
    name = graphene.String()
    best_friend = graphene.Field(lambda: User)
    friends = graphene.List(lambda: User)

    def resolve_best_friend(root, info):
        return user_loader.load(root.best_friend_id)

    def resolve_friends(root, info):
        return user_loader.load_many(root.friend_ids)

It's not clear how this could work, because as it's presented here, neither root.best_friend_id nor root.friend_ids would be defined. Wouldn't it make sense to describe a real use case? The current documentation is more or less useless because when you first read it you can't tell what is an approximation and what is not (example of another non-working bit of code User.objects.filter(id__in=keys)). I'm not even sure yet about all this because I know graphene rewrites classes but I don't know all the details since I only used it (and never contributed). So maybe there is to trickery hidden in these statements, but I really feel there are several errors in the provided examples.

Maybe the solution to this is only to add empty objects/resolver (pass) but with well defined types. So everything is clearly defined (but unimplemented).

I've tried to make sense about graphene + aioloader by writing three simple helpers function that create some batch functions:

from collections import defaultdict
from typing import Callable, Dict, List, Optional, Tuple, TypeVar

class OneToOne:
    T = TypeVar("T")
    K = TypeVar("K", str, int)
    Query = Callable[[List[K]], List[T]]
    Key = Callable[[T], K]


OneToMany = OneToOne


class ManyToMany:
    T = TypeVar("T")
    Query = Callable[[List[str]], List[Tuple[T, str]]]


def one_to_one(query: OneToOne.Query, key: OneToOne.Key):

    async def batch_function(keys: List[str]) -> List[Optional[OneToOne.T]]:
        user_objects = {key(obj): obj for obj in query(keys)}
        return [user_objects.get(k) for k in keys]

    return batch_function


def one_to_many(query: OneToMany.Query, key: OneToMany.Key):

    async def batch_function(keys: List[str]) -> List[List[OneToMany.T]]:
        objects = query(keys)
        user_objects = defaultdict(list)
        for obj in objects:
            user_objects[key(obj)].append(obj)
        return [user_objects.get(k) or [] for k in keys]

    return batch_function


def many_to_many(query: ManyToMany.Query):

    async def batch_function(keys: List[str]) -> List[List[ManyToMany.T]]:
        obj_key_pairs = query(keys)
        user_objects: Dict[str, List[ManyToMany.T]] = defaultdict(list)
        for obj, k in obj_key_pairs:
            user_objects[k].append(obj)
        return [user_objects.get(k) or [] for k in keys]

    return batch_function

Describing this kind of functions might even help people understanding what is expected from them. I'm not sure yet these types (or even helper functions) are 100% correct, let me know if you find better solutions.

[stan-sack]

@cglacet is there any chance you can give an example of a query you're using for a many_to_many loader here? I cant see how a single ORM query would return a list of object key pairs? I'd say that apart from that your code seems like the best solution i've seen