This is a proof-of-concept of using code-generation to create a truly type-safe GraphQL client in Go. It is certainly not ready for production use!
To understand the issue, consider the example from the documentation of shurcooL/graphql:
// to send a query `{ me { name } }`:
var query struct {
Me struct {
Name graphql.String
}
}
// error handling omitted for brevity
client.Query(context.Background(), &query, nil)
fmt.Println(query.Me.Name)
// Output: Luke Skywalker
While your code is in principle type-safe, there's nothing to check that the schema looks like you expect it to. In fact, perhaps here we're querying the GitHub API, in which the field is called viewer
, not me
, so this query will fail. Even more common than misusing the name of the field is mis-capitalizing it: the GraphQL convention is myUrl
whereas the Go convention is MyURL
, and it's easy to forget or mistype the struct tag. (Even if you get it right, it adds up to a lot of handwritten boilerplate!) Other clients, such as machinebox/graphql, have even fewer guardrails to help you make the right query and use the result correctly. This isn't a big deal in a small application, but for serious production-grade tools it's not ideal.
These problems should be entirely avoidable: GraphQL and Go are both typed languages; and GraphQL servers expose their schema in a standard, machine-readable format. We should be able to simply write a query { viewer { name } }
, have that automatically validated against the schema and turned into a Go struct which we can use in our code. In fact, there's already good prior art to do this sort of thing: 99designs/gqlgen is a popular server library that does exactly this: it generates type-safe GraphQL resolvers from a schema.
This is a proof-of-concept of a GraphQL client that does the same sort of thing: you specify the query, and it generates type-safe helpers that make your query.
# queries.graphql
query getViewer {
viewer {
MyName: name
}
}
// generated.go (auto-generated):
type getViewerResponse struct { ... }
func getViewer(ctx context.Context, client *graphql.Client) (*getViewerResponse, error) { ... }
// your code (error handling omitted for brevity)
graphqlClient := graphql.NewClient("https://example.com/graphql", http.DefaultClient)
viewerResp, _ := getViewer(context.Background(), graphqlClient)
fmt.Println("you are", viewerResp.Viewer.MyName)
//go:generate go run github.com/Khan/genql
For a complete working example, see example/
.
go test ./...
tests code generation. (This is run by GitHub Actions.)
Most of the tests are snapshot-based; they use the schema, queries, and snapshots in generate/testdata
. The schema is in schema.graphql
; the queries are in TestName.graphql
. The test by default asserts that the output of the generator matches the snapshot TestName.graphql.go
. To update the snapshots, run with UPDATE_SNAPSHOTS=1
; it will fail the tests and print the diffs, but regenerate the snapshots. Make sure to check that the output is sensible!
make example
rebuilds the example, and tests that everything wires up to a real API correctly.
TODO(benkraft): Figure out how to get GitHub Actions to run the example -- it needs a token.
See DESIGN.md for documentation of major design decisions in this library.
(*) denotes things we need to use this in prod at Khan (+) denotes things we further need before recommending anyone else use this in prod
Generated code:
- add flag(s) to make a field use a pointer (for optionality or perf)
- redo support for interfaces, unions, fragments (see DESIGN)
- (optional) collapsing -- should be able to have
mutation { myMutation { error { code } } }
just return(code string, err error)
Config options:
- (+) proper config/arguments setup (e.g. with viper)
- (+) fix up context/client wiring (see DESIGN)
- get schema via HTTP (perhaps even via GraphQL introspection)
- send hash rather than full query
- whether names should be exported
- default handling for optional fields (pointers, HasFoo, etc.)
- response/function-name format (e.g. force exported/unexported, change "Response" suffix, etc.)
- generate mocks?
Other:
- (*) error-checking/validation/etc. everywhere
- (*) a name that's more clearly distinct from other libraries out there and conveys what this does
- (+) more tests
- (+) documentation
- custom scalar types (or custom mappings for standard scalars, if you want a special ID type say)
- allow mapping a custom type to a particular val (if you want to use a named type for some string, say)