This is experimental support.
The implementation is based on conduit and is boring. There is a new
resolver data constructor that should create a source event stream. The
executor receives the events, pipes them through the normal execution
and puts them into the response stream which is returned to the user.
- Tests are missing.
- The executor should check field value resolver on subscription types.
- The graphql function should probably return (Either
ResponseEventStream Response), but I'm not sure about this. It will
make the usage more complicated if no subscriptions are involved, but
with the current API implementing subscriptions is more
difficult than it should be.
Returning resolvers from other resolvers isn't supported anymore. Since
we have a type system now, we define the resolvers in the object type
fields and pass an object with the previous result to them.
Fixes#39.
String containing control sequences should be encoded as simple strings
even if they contain newlines, since the block strings can contain only
SourceCharacters.
It makes using variables with queries more approachable, but some work
still has to be done.
- The type `Subs` should be renamed and moved out of `Schema`, together with
`AST.Core.Value` probably.
- Some kind of conversion should be possible from a user-defined input
type T to the Value. So the final HashMap should have a type like
`HashMap name a`, where a is an instance of a potential typeclass
InputType.
Fixes#18.
- `Language.GraphQL.Encoder` moved to `Language.GraphQL.AST.Encoder`.
- `Language.GraphQL.Parser` moved to `Language.GraphQL.AST.Parser`.
- `Language.GraphQL.Lexer` moved to `Language.GraphQL.AST.Lexer`.
- All `Language.GraphQL.AST.Value` data constructor prefixes were removed. The
module should be imported qualified.
- All `Language.GraphQL.AST.Core.Value` data constructor prefixes were removed.
The module should be imported qualified.
- `Language.GraphQL.AST.Transform` is now isn't exposed publically anymore.
These functions are from Language.GraphQL.Schema.
There are actually only two generic types in GraphQL: Scalars and objects.
Enum is a scalar value. According to the specification enums may be
serailized to strings. And in the current implementation they used
untyped strings anyway, so there is no point to have differently named
functions with the same implementation as their scalar counterparts.
It is not a schema (at least not a complete one), but a resolver list,
and the resolvers should be provided by the user separately, because the
schema can originate from a GraphQL document. Schema name should be free
to provide a data type for the real schema later.
This replaces the most usages of MonadPlus, which is not appropriate for
the resolvers, since a resolver is unambiguously chosen by the name (no
need for 'mplus'), and the resolvers are often doing IO.
Now the errors in the resolvers can be handled and 3 tests throwing
errors pass now. Another test fail but it requires distinguisching
nullable and non-nullable values.
One AST is meant to be a target parser and tries to adhere as much as possible
to the spec. The other is a simplified version of that AST meant for execution.
Also newtypes have been replaced by type synonyms and NonEmpty lists are being
used where it makes sense.
Aside of making the definition of Schemas easier, it takes care of
issues like nested aliases which previously wasn't possible. The naming
of the DSL functions is still provisional.
The `Schema` has been overhauled to make `Output` monomorphic.
Traversing the `GraphQL` document is handled implicitly while defining
the `Schema`.
The 4th end-to-end test from `graphql-js` has been ported.
`Character` is now a synonym of the sum type of `Droid` and `Human`.
For now I don't see the need to implement GraphQL Schema interfaces with
type classes or lens. Plain Haskell ADTs should be good enough.
The first end-to-end test taken from `graphql-js` passes but this still
needs to be extended to support more general cases.
- `Data.GraphQL.Schema` has been heavily modified to support the
execution model. More drastic changes are expected in this module.
- When defining a `Schema` ordinary functions taking fields as input are
being used instead of maps. This makes the implementation of `execute`
easier, and, arguably, makes `Schema` definitions more *Haskellish*.
- Drop explicit `unordered-containers` dependency. `Aeson.Value`s and
field functions should be good enough for now.
