Add GraphQL support

[chrisabrams]

I don't see it anywhere - is this an adaptor in another repo? GraphQL is definitely one piece of the project I am excited to use.

[mglukhovsky]

@chrisabrams: we're working through details on how to add a GraphQL adapter offline -- so far we've started with a rich API to query the backend, but GraphQL is on the list.

This issue is a good opportunity to discuss proposals and define what a GraphQL adapter would look like.

[taion]

Worth keeping an eye on facebook/relay#541, since Relay doesn't actually have subscription support yet.

[danielmewes]

It's worth noting that GraphQL is not at the core of Horizon. It can (and very likely will) be an additional protocol that can use Horizon functionality and infrastructure. rethinkdb/rethinkdb#3711 (comment) is a good summary of why Horizon is different from GraphQL

[coffeemug]

I renamed the issue to "Add GraphQL support" (sorry, e-mail users!)

GraphQL will not make it into v1, but I'm hoping we can add a GraphQL adapter on top of the protocol quickly after launch. If anyone has proposals for what specifically you're looking for wrt GraphQL, please post it here. I think the general goal should be to have a complete backend out of the box, so users could build React/Relay apps without initially writing any backend code. Once the app gets sophisticated, GraphQL should work similarly to the basic protocol -- it should be possible to import Horizon into Node as a library, and add custom handlers to do advanced operations.

I don't have details on how GraphQL will work yet; once the first version of Horizon ships, I expect we'll get to this pretty quickly.

[alexgorbatchev]

complete backend out of the box, so users could build React/Relay apps without initially writing any backend code

That! I think it would be a good idea to have GraphQL support during the initial public announcement, I believe it would make a bigger splash that way.

[coffeemug]

Adding GraphQL for the initial launch would be incredible, but unfortunately we don't have the resources to make it happen. If someone wants to contribute, it would be a really fun project to work on (wink wink!)

[taion]

To reiterate, I don't think it's a question of resourcing. It's just about impossible right now because the semantics defined for subscriptions in GraphQL are restricted to just having a support for a subscription message type.

I think you're fundamentally blocked on facebook/relay#541 here, because until that lands, there are no real semantics for e.g. subscription updates. You have a way to establish subscriptions, but nothing for subscription updates, and no semantics (or client-side support) for consuming such updates.

[coffeemug]

I don't think that's quite right. We could add GraphQL support without subscriptions (i.e. support fetch semantics but not watch semantics). It wouldn't give users the realtime features of horizon, but that's ok -- once GraphQL facebook/relay#541 we can add watch semantics pretty easily.

[taion]

Ah, I see. I take that back, then. Thanks.

[tslocke]

If anyone has proposals for what specifically you're looking for wrt GraphQL, please post it here

I realise this is stating the obvious, but maybe it doesn't hurt to sanity check that people are on the same page...

Composable UI components!

I want to build my UI as a collection of components, each annotated with the data needs of that one component (in GraphQL) and have the framework figure out the composite query that gets sent to the server.

I want my dev process to look like:

1. Write components
2. Tell Horizon which one is the root component
3. hzops deploy : )

OK there may be a few steps missing in there, but I think keeping this ideal in mind will give a good direction for the design of the GraphQL adapter.

The weird thing in all this is that Relay currently has a crazy high level of boilerplate. As we saw with Meteor, I think Horizon needs a message like "look how little code is required to do amazing things!", so excessive boilerplate is a problem.

So far, Horizon wants to stay out of the view layer, but the real promise of GraphQL is fundamentally about moving data access into UI components. Very interested to hear thoughts on this.

Maybe the hard part is finding the right boundary for where Horizon ends and Relay begins.