Added the ability to specify the JSON encoder

[switchtrue]

json.dumps() allows for a JSON encoder to be specified by the cls argument. This PR allows for a JSONEncoder to be passed through when constructing the GraphQLView object.

Supporting this PR allows for serialization of custom types which can be particular useful when defining custom graphene scalar types.

[coveralls]

Coverage increased (+0.02%) to 96.512% when pulling 8077fee on mleonard87:feature/specify-encoder into 2825345 on graphql-python:master.

[rwe]

I'm just a bystander here, but this change looks good to me. My comments aren't actual issues, just suggestions.

[rwe]

Any reason to continue using json.dumps here rather than the encoder directly?

return self.json_encoder(separators=(',' ':')).encode(d)

I believe they're functionally equivalent but the .encode() route reduces the dependency on the builtin json module, since there are several json libraries that could be reasonably used: http://artem.krylysov.com/blog/2015/09/29/benchmark-python-json-libraries/

[switchtrue]

I'm not an expert on json encoding/decoding in Python. What you've said sounds great but I'd rather leave this to someone in a better position to comment.

[rwe]

Should this also include a json_decoder member? Roughly equivalent change with the json.loads -> self.json_decoder(…).decode(s) a couple places below.

[switchtrue]

Completely agree, I've just updated the branch with json_decoder support.

[coveralls]

Coverage increased (+0.04%) to 96.532% when pulling a3f84aa on mleonard87:feature/specify-encoder into 2825345 on graphql-python:master.

[coveralls]

Coverage increased (+0.04%) to 96.532% when pulling a3f84aa on mleonard87:feature/specify-encoder into 2825345 on graphql-python:master.

[coveralls]

Coverage increased (+0.04%) to 96.532% when pulling a3f84aa on mleonard87:feature/specify-encoder into 2825345 on graphql-python:master.

[coveralls]

Coverage increased (+0.04%) to 96.532% when pulling a3f84aa on mleonard87:feature/specify-encoder into 2825345 on graphql-python:master.

[rwe]

@syrusakbary This PR looks good, is anything preventing it from merging?

[syrusakbary]

Hi @erydo @Mleonard87,

Thanks for taking time for making this PR. I think the serialization/deserialization is better to be assumed by the scalar type itself rather than the transport formatting language (json, msgpack, ...) so is universal and completely independent of it.

However I might be missing some of your context here. Would be great if you could provide an example of a Scalar type that could not be created without the "customize json serialization class" approach.

[rwe]

@syrusakbary I think you're mostly right, but a couple considerations:

1. There are some better-performing serialization libraries that, if installed, are nice to use: http://artem.krylysov.com/blog/2015/09/29/benchmark-python-json-libraries/
2. One of the use-cases that we'd run across was error serialization, for example; we have an internally-standard JSONEncoder that handles things like datetime and so on. Though I suppose that could be done in a format_error implementation, it's harder to backport a custom encoder to instead produce a dict that would encode identically through a vanilla encoder.

Error serialization doesn't go through the Scalar system, though I think pretty much everything else does.

[syrusakbary]

For error serialization, you can provide a custom error function format_error.
If you want to have another json serializer/deserializer (other than json, such as simplejson or ujson) I think that's will be a good idea to add into this repo.

But then the solution for that is not a custom decoder for json (as this PR is doing)

[flipbit03]

I have just stumbled upon this, I am having the same need.

We have a SQLAlchemy model that is using a custom class called 'MVStringBag', it's a bag of translateable strings with a default. It inherits from sa.STRING.

This custom type is used in several SQLAlchemy' Model Classes to provide 'automatic' multi-language strings in our API:

class OrderedPostPart(UnhideDeclBase):

    (...redacted...)

    # Groupname
    groupname = Column(
        MVStringBag.as_mutable(MVStringBagSAType), nullable=False
    )  # type: MVStringBag

    (...redacted...)

It allows you to send a key to it, specifying the wanted language and, if that's not available, it'll return the default string.

class MVStringBagSAType(types.TypeDecorator):
    impl = sqlalchemytypes.String

    (...redacted...) 

To generate the GraphQL Classes, I am using the Meta: feature you guys provide to automatically plug all of the Model classes' attributes to GraphQL Scalars and, since MVStringBag inherits from String, it'll be (correctly) a graphene.String (or a graphene.JSONString, in what i have seen in the wild) on the GraphQL side. which is the final type I expect to return to the API, so that translation is totally transparent and the api returns a simple and standard string.

But this class cannot be serialized because it's not one of the default types. So passing a custom JSONEncoder would be really useful.