Preserve the serialization order for mutations

[kyasu1]

Hi!

I run into a problem when sending multiple mutations in a single request to Hasura server backed with PostgreSQL. The mutation query requires to execute update and insert operations sequentially in a transaction of PostgreSQL, however elm-graphql is serializing their GraphQL operations into alphabetical order which is the problem.

More specifically, with this code we expect the updateUser runs first then the insertUser.

SelectionSet.map2 Tuple.pair updateUser insertUser

However actually this is serialized to something like follows.

query: “mutation {
  insert_user12345: insert_user(…)
  update_user12345: update_user(…)
}”

With this query, Hasura issues PostgreSQL query in order of insert then update which will fail in my schema. I think this behavior is counter intuitive and can be a pitfall for users. Also I have searched a bit for another projects and I found this and GraphQL specs

http://spec.graphql.org/October2021/#sec-Normal-and-Serial-Execution

which says the execution order must be in serial for mutations.

The current implementation uses standard Dict package for serializing fields

elm-graphql/src/Graphql/Document/Field.elm

Lines 127 to 141 in 0680565

	mergedFields : List RawField -> List RawField
	mergedFields children =
	let
	mergeThing : MergedFields
	mergeThing =
	mergeFields children
	in
	(mergeThing.leaves |> Dict.values |> List.map leafToField)
	++ (mergeThing.composites |> Dict.values |> List.map compositeToField)
	type alias MergedFields =
	{ leaves : Dict String ( { typeString : String, fieldName : String }, List Argument )
	, composites : Dict String ( { name : String, args : List Argument }, List RawField )
	}

which cause serialization of the fields in alphabetical order.

To prevent this re-ordering, in this PR, I have naively replaced it with j-mass/elm-ordered-containers and it seems to give us the expected result.

Thanks

[dillonkearns]

Thank you for the PR @kyasu1! It is indeed very important that the order be preserved for mutations. I wasn't thinking about that when I worked on #556. Thank you for the report and the PR!

I went ahead and made a few changes to get this green and also to handle some additional cases (so that ordering is preserved between leaf and composite fields correctly).

[dillonkearns]

This is now live in Elm package version 5.0.8. Thanks again!