Query caching broken with some custom scalars and JIT
lsanwick opened this issue · comments
I'm building a schema that has a number of custom scalars like DateTime
that are automatically validated and parsed into a Luxon DateTime
object when provided as arguments.
This works great, except it seems to interact poorly with the default request cache, where subsequent requests will just provide the string representation of the argument.
Here's a toy implementation that shows the issue:
scalar DateTime
type Query {
testDateTime(time: DateTime!): String!
}
const parseDateTime = (astOrString) => DateTime.fromISO(dateTime, { zone: "utc" })
const resolvers = {
DateTime: new GraphQLScalarType<LuxonDateTime, string>({
name: "DateTime",
description:
"An ISO8601 formatted date time string (e.g. 2023-02-01T08:30:00Z), parsed into a Luxon DateTime",
serialize(value) {
if (value instanceof LuxonDateTime) {
return value.toISO();
}
throw new Error("DateTime must be a Luxon DateTime");
},
parseValue(dateTime) {
return DateTime.fromISO(dateTime, { zone: "utc" })
},
parseLiteral(ast) {
if (ast.kind === Kind.STRING) {
return DateTime.fromISO(ast.value, { zone: "utc" })
}
},
Query: {
dateTimeMessage(_root, { time }) {
return `Good day! It's ${time.toISODate()}`
}
}
})
Example query
query {
testDateTime(date: "2023-05-16T21:30:00.000Z")
}
First response (correct)
{
"data": {
"testDateTime": "Good day! It is 2023-05-16"
}
}
Second response (incorrect)
{
"data": null,
"errors": [
{
"message": "time.toISODate is not a function",
"locations": [
{
"line": 2,
"column": 3
}
],
"path": [
"testDateTime"
]
}
]
}
Disabling the caching with cache: false
on mercurius when registering the plugin fixes the issue.
Thanks for reporting!
Can you provide steps to reproduce? We often need a reproducible example, e.g. some code that allows someone else to recreate your problem by just copying and pasting it. If it involves more than a couple of different file, create a new repository on GitHub and add a link to that.
Okay, I've created a repoduction repo here: https://github.com/lsanwick/custom-scalar-caching-example
It appears that the issue appears with this configuration:
await fastify.register(mercurius, {
schema,
resolvers,
jit: true,
graphiql: true,
});
If you disable JIT, or disable caching while keeping JIT enabled, I don't see the issue.
jit: true,
cache: false,
@mcollina Is that sufficient to work off of, or do you need anything else?
I currently do not have much time to look into it. This looks like an issue inside graphql-jit.
I never experience this exact issue so maybe there is something that intersects between that and luxon.