Better notation for building servers

[serras]

Right now, building a server requires matching the exact order in which methods and services are defined in the correspoding type. When this type comes from importing a .proto, .avdl, or in the future GraphQL schema, keeping this order is quite difficult. Furthermore, @kutyel has noticed that the error message you get is quite un-informative.

So we need to find a better notation. Right now we are thinking of a blaze-builder-like notation using do, like:

service @"Book" $ do
  method @"id" ...
  method @"author" ...

@arianvp you might be interested in this

[arianvp]

How will you make sure that you get a type error when you 'miss' a certain rpc call?

Could perhaps fixed by using OverloadedSyntax and use IxMonad, but then you enforce implicit ordering again...

[serras]

Sure, the goal is to get a compile-time error if you miss something. I was also thinking along the lines of RebindableSyntax, but I'm not so happy with the idea, because that makes inference in the rest of the file terrible :(

[serras]

Here's my first attempt: https://github.com/higherkindness/mu-haskell/blob/better-server-api/core/rpc/src/Mu/Rpc/Examples.hs#L75

@kutyel @arianvp what do you think?

Unfortunately, I could not get a good error message for the case in which you are missing a method. The reason is that for the inference to work I need to have a functional dependency on ToHandlers. But that dependency precludes me from having this instance.

[serras]

@arianvp after some discussion with @kutyel, we came with a better API using tuples instead of weird heterogeneous lists. So you can now declare servers in this way:

singleService ( method @"SayHello" sayHello
              , method @"SayManyHellos" sayManyHellos
              , method @"SayHi" sayHi )

and GraphQL resolvers as follows:

resolver
  ( object @"Author" ( field @"name"   authorName
                     , field @"books"  authorBooks )
  , object @"Book"   ( field @"author" (pure . snd)
                     , field @"title"  (pure . fst) ) )

What do you think? I find it quite nice :)