The <512 byte solution to your GraphQL ambiguous null
woes.
Works with:
- Apollo Client
- URQL
- graffle
- window.fetch()
- any GraphQL client that returns the JSON
{ data, errors }
Not needed with Relay as it has native error handling support via the @throwOnFieldError and @catch directives.
You read null
from a field in GraphQL... but is that a data-null (explicit
non-existence) or an error-null (something went wrong)? This is an important
distinction: your boyfriend's profile page showing Partner: none
is very
different than it showing Error loading partner
!
If you're not using an error-handling GraphQL client, then for each null
you
see in a GraphQL response you must check through the errors
list to determine
if it relates to an error or not. To do so, you need to know the path of the
currently rendering data - instead of just passing the data to your component,
you need to pass the root list of errors, and the path of the data being
rendered.
Managing this yourself is a huge hassle, and most people don't bother - instead
either rejecting requests that include errors (and losing the "partial success"
benefit of GraphQL) or treating all null
as ambiguous: maybe it errored, maybe
it's null, we don't know. Or worse, they treat an error-null as if it is a
data-null, and cause much heartbreak!
Well, no more!
GraphQL-TOE transforms your GraphQL response ({ data: {...}, errors: [...] }
)
into a new object that looks exactly like data
, except it throws when you
access a position that is null
due to an error. As such, your application can
never read an error-null (because the error will be thrown if you try) - so if
you read a null
value you know it's definitely a data-null and can render it
as such. And for the errors... you can handle them as any other throw error:
with native JavaScript methods like try
/catch
, or those built on top of them
such as React's <ErrorBoundary />
!
Stop writing code that second-guesses your data; re-throw GraphQL errors!
yarn add graphql-toe
# OR: npm install --save graphql-toe
# OR: pnpm install --save graphql-toe
import { toe } from "graphql-toe";
const result = await fetch(/* ... */).then((res) => res.json());
const data = toe(result);
If result.data
is null
or not present, toe(result)
will throw immediately.
Otherwise, data
is a derivative of result.data
where errored fields are
replaced with throwing getters.
Under 512 bytes gzipped (v1.0.0-rc.1 was 471 bytes according to bundlephobia)
Works with any GraphQL client that returns { data, errors }
.
Errors are thrown as-is; you can pre-process them to wrap in Error
or
GraphQLError
if needed:
import { GraphQLError } from "graphql";
import { toe } from "graphql-toe";
const mappedResult = {
...result,
errors: result.errors?.map(
(e) =>
new GraphQLError(e.message, {
positions: e.positions,
path: e.path,
originalError: e,
extensions: e.extensions,
}),
),
};
const data = toe(mappedResult);
import { toe } from "graphql-toe";
// Result of query `{ users(first: 3) { id, name } }`
const graphqlResult = {
data: {
users: [
{ id: 1, name: "Alice" },
null, // < An error occurred
{ id: 3, name: "Caroline" },
],
},
errors: [
{
path: ["users", 1],
message: "Loading user 2 failed!",
},
],
};
// Return the transformed data that will Throw On Error:
const data = toe(graphqlResult);
console.log(data.users[0]); // Logs { id: 1, name: "Alice" }
console.log(data.users[1]); // Throws "Loading user 2 failed!"
Different frameworks and libraries have different approaches to feeding the GraphQL result into GraphQL-TOE:
import { useQuery } from "@apollo/client";
import { toe } from "graphql-toe";
import { useMemo } from "react";
function useQueryTOE(document, options) {
const rawResult = useQuery(document, { ...options, errorPolicy: "all" });
return useMemo(
() => toe({ data: rawResult.data, errors: rawResult.error?.graphQLErrors }),
[rawResult.data, rawResult.error],
);
}
Now simply replace all usages of useQuery()
with useQueryTOE()
.
Note: apply similar changes for mutations and subscriptions.
Use @urql/exchange-throw-on-error:
import { Client, fetchExchange } from "urql";
import { throwOnErrorExchange } from "@urql/exchange-throw-on-error";
const client = new Client({
url: "/graphql",
exchanges: [fetchExchange, throwOnErrorExchange()],
});
import { request } from "graffle";
const result = await request("https://api.spacex.land/graphql/", document);
const data = toe(result);
import { toe } from "graphql-toe";
const response = await fetch("/graphql", {
headers: {
Accept: "application/graphql-response+json, application/json",
"Content-Type": "application/json",
},
body: JSON.stringify({ query: "{ __schema { queryType { name } } }" }),
});
if (!response.ok) throw new Error("Uh-oh!");
const result = await response.json();
const data = toe(result);
Relay has native support for error handling via the @throwOnFieldError and @catch directives - use that instead!
The
@semanticNonNull
directive lets schema designers mark fields where null
is never a valid
value; a null
in such a position must mean an error occurred (and thus there
will be an entry in the errors
list matching the path).
With toe()
you can treat these @semanticNonNull
fields as non-nullable since
we know an error-null can never be accessed; and thus your JavaScript/TypeScript
frontend code will need fewer null checks!
In TypeScript, use semanticToStrict from graphql-sock to rewrite semantic-non-null to traditional non-null for type generation.
Together, this combination gives you:
- More accurate codegen types
- Improved DX with fewer null checks
- Safer, cleaner client code
Creates copies of data impacted by errors, using getters to throw when error positions are accessed.
Highly efficient: only response sections impacted by errors are copied; with no errors the underlying data is returned verbatim.
There's growing consensus amongst the GraphQL Working Group that the future of GraphQL has errors handled on the client side, with server-side error propagation disabled. This fixes a number of issues, among them the proliferation of null types (and the associated nullability checks peppering client code), and inability to safely write data to normalized caches if that data came from a request containing errors.
Over time, we hope all major GraphQL clients will integrate error handling deep into their architecture so that users don't need to think about it. In the mean time, this project can add support for this future behavior to almost any GraphQL client by re-introducing thrown errors into your data.
Handle errors the way your programming language or framework is designed to — no need for GraphQL-specific logic.
Read more on the motivation behind this here: https://benjie.dev/graphql/nullability/
import { toe } from "graphql-toe";
// Example data from GraphQL
const result = {
data: {
deep: {
withList: [
{ int: 1 },
{
/* `null` because an error occurred */
int: null,
},
{ int: 3 },
],
},
},
errors: [
{
message: "Two!",
// When you read from this path, an error will be thrown
path: ["deep", "withList", 1, "int"],
},
],
};
// TOE'd data:
const data = toe(result);
// Returns `3`:
data.deep.withList[2].int;
// Returns an object with the key `int`
data.deep.withList[1];
// Throws the error `Two!`
data.deep.withList[1].int;
- Add support for incremental delivery
Version 0.1.0 of this module was released from San Francisco the day after GraphQLConf 2024, following many fruitful discussions around nullability.
Version 1.0.0 of this module was released just before GraphQLConf 2025, as the result of what we call Conference-Driven Development.