Client for managing application state with GraphQL operations.
Blips exposes a simple interface for managing your application state with GraphQL.
The state is contained inside a single store object. It can only be changed through mutations, you can read from it through queries and can also listen for changes with subscriptions.
I developed Blips because I wanted to use GraphQL with every project, regardless of what the API server looks like. So if my application is consuming a simple API (whatever kind), I can still have a store that manages the application state and write queries/mutations that would resolve with making requests to that API.
The following examples assume some familiarity with GraphQL. If you haven't used it before, or you have no idea about what it is, you should visit graphql.org and read about GraphQL in detail.
npm install blips graphql
new BlipsClient({ typeDefs [, resolvers] } [, initialState] [, config] )
Creating a client requires type definitions, optional resolvers, optional initial state and an optional configuration object.
// types.js
export default `
type Todo {
id: String!
label: String!
completed: Boolean!
}
type Query {
allTodos: [Todo]!
todo(id: String!): Todo!
}
type Mutation {
createTodo(id: String!, label: String!, completed: Boolean)
}
`
The types.js
file contains 3 definitions:
- a basic
Todo
object type containing three fields:id
,label
,completed
. - the root
Query
type - the root
Mutation
type
The !
marks the field as required. See
graphql.org to read more about
the type system.
// resolvers.js
export default {
Query: {
allTodos: (obj, args, { store }) => store.state.todos || [],
},
Mutation: {
createTodo: (
obj,
{ id, label, completed = false },
{ store: { state: { todos } } }
) => {
const newTodo = { id, label, completed }
todos = [...todos, newTodo]
return newTodo
},
},
}
The resolvers.js
file provides definitions for allTodos
query and the
createTodo
mutation. Each resolver function accepts three arguments:
obj
: The previous object.args
: The arguments provided to the field in the GraphQL query.context
: Provides access to important information like the currently logged in user, or the store itself.
See graphql.org to read more about resolvers.
// index.js
import { BlipsClient } from 'blips'
import resolvers from './resolvers'
import typeDefs from './types'
const initialState = {
todos: {
'3c4a086e-2151-4b54-acb2-13044ea553c1': {
id: '3c4a086e-2151-4b54-acb2-13044ea553c1',
label: 'Buy milk',
completed: false,
},
},
}
const schemaDef = { typeDefs, resolvers }
// new BlipsClient({ typeDefs [, resolvers] } [, initialState] [, config] )
const client = new BlipsClient(schemaDef, initialState)
You create a new client by providing your schema definitions ({ typeDefs, resolvers }
) and an optional initial state. In addition to the schemaDefs
and
initialState
, you can also privide a configuration
object as the last and
optional argument. The configuration
parameter is used for tapping into the
initial setup to define default variables or context properties and to configure
the fetch
method.
The client object has the following API:
state
: getter for your entire state.schema
: getter for your generated schema. state-clerk.query
: method for executing queries.mutate
: method for executing mutations.subscribe
: method for registering subscriptions.fetch
: method for sending queries to a real GraphQL API.
You define operations in the form of GraphQL queries, mutations or subscriptions and use the client methods for executing them.
// operations.js
export const todoQuery = `
query todoQuery {
todo {
label
completed
}
}
`
export const createTodoMutation = `
mutation createTodoMutation($label: String!, $completed: Boolean) {
createTodo(label: $label, completed: $completed) {
id
label
completed
}
}
`
Read more about queries and mutations
import { todoQuery, createTodoMutation } from './operations'
client
.query(todoQuery, {
variables: { id: '3c4a086e-2151-4b54-acb2-13044ea553c1' },
})
.then(res => {
console.log(res)
// {
// data: {
// todo: {
// label: 'Buy milk',
// completed: false
// }
// }
// }
})
client
.mutate(createTodoMutation, {
variables: {
id: '4ecca858-67f8-491e-94cc-48b262061819',
label: 'Learn Blips',
},
})
.then(res => {
console.log(res)
// {
// data: {
// createTodo: {
// id: '4ecca858-67f8-491e-94cc-48b262061819',
// label: 'Learn Blips',
// completed: false
// }
// }
// }
})
If you've used Redux before you can think of execution methods as dispatchers. All executors may receive three arguments:
operation
: can be in the form of source or documentASToptions
: object containing:variables
: the operation variablescontext
: object containing any additional data that you want to pass to the resolvers, like the currently logged in user, tokens etc. (This will extend the default context)
All executors return promises that eventually resolve with an object containing
a data
prop which holds the requested data. If, for some reason, the operation
was not successful, the data
prop will hold an error
object instead of the
requested field.
Managing your state in an asynchronous way might seem scary but unless you're actually fetching data from a web service, hitting an API or performing database queries, your data will be available with the next tick. This behaviour also comes with a few benefits:
- it forces you to introduce checks and error handlers which will eventually result in better code.
- you can write abstractions or helpers for managing your data and they will apply everywhere.
- it's much easier if you decide to switch from caching some of your state locally to exclusively fetching it from an external service.
Plus, using async/await
will still make your code look synchronous and badass.
In addition to polling queries or scheduling them to execute at different moments throughout your application life-cycle to keep your data up-to-date, you can also subscribe to changes through GraphQL subscriptions:
const allTodosSubscription = `
subscription allTodosSubscription {
allTodos {
id
label
completed
}
}
`
const asyncIterator = await client.subscribe(allTodosSubscription)
/* Either use for-await-of */
try {
for await (const tick of asyncIterator) {
console.log(tick);
// { data: { allTodos: [ ... ] } }
}
} catch (e) {
// { data: { error: { ... } } }
}
/* or transform the iterator into an observable and subscribe to it */
const sub = asyncIterator.toObservable().subscribe(
next => {
// { data: { allTodos: [ ... ] } }
},
error => {
// { data: { error: { ... } } }
}
)
// remember to dispose of any unused subscriptions
// sub.unsubscribe()
Blips uses Apollo's
graphql-subscriptions
PubSub implementation, where any query or mutation would publish
data over
a specific topic while your subscription resolvers subscribe to one or more
topics.
In order to use subscriptions with Blips, you need to have access to the
client's PubSub
instance. We can achieve that by passing a resolvers
function instead of an object when creating the client instance. This function
accepts as first argument an object containing the PubSub
instance and the
withFilter
method.
// resolvers.js
export default ({ pubsub, withFilter }) => ({
Query: { ... },
Subscription: {
allTodos: {
resolve: (obj, args, { store }) => store.state.todos || [],
subscribe: () => pubsub.asyncIterator([ 'TODO_UPDATED', 'TODO_CREATED', 'TODO_DELETED', ]),
}
}
Mutation: {
createTodo: (obj, { id, label, completed = false }, { store: { state: { todos } } }) => {
const newTodo = { id, label, completed }
todos = [ ...todos, newTodo ]
pubsub.publish('TODO_CREATED')
return newTodo
}
}
})
In the above example, the createTodoMutation
publishes over TODO_CREATED
as
soon as the todos get updated with the new entry. It doesn't need to publish any
data because the allTodos
subscription will return all the todos and it's
resolve
method will do just that.
When publishing data to subscribers, we need to make sure that each subscribers get only the data it needs.
To do so, we can use withFilter helper from this package, which wraps AsyncIterator with a filter function, and let you control each publication for each user.
// types.js
export default {
...
type Subscription {
todo(id: String!): Todo!
}
...
}
// resolvers.js
export default ({pubsub, withFilter}) => ({
Query: { ... },
Subscription: {
todo: {
subscribe: withFilter(
() => pubsub.asyncIterator('TODO_UPDATED'),
(payload, variables) => payload && payload.todo.id === variables.id
)
}
},
Mutation: {
updateTodo: (obj, { id, label, completed }, { store }) => {
const modified = store.patch(id, {
label,
completed,
})('todos')
pubsub.publish('TODO_UPDATED', { todo: modified })
return modified
}
}
})
The updateTodoMutation
will publish the updated todo over the TODO_UPDATED
topic. If we have subscriptions registered for each todo, only those who's ID
variable match to the updated todo's ID will emit.
Head over to graphql-subscriptions for a more detailed documentation on how to use this PubSub implementation.
Your access is not limited to the default context, nor limited to working with only the local state. Resolvers can completely ignore any context and just make async requests to external API servers.
const resolvers = {
Mutation: {
createTodo: async () =>
await fetch('api/v1/todos', {
method: 'post',
body: JSON.stringify(args)
}).then(res => res.json())
}
}
const createTodoMutation = `
mutation createTodoMutation($label: String!, completed: Boolean) {
createTodo(label: $label, completed) {
id
label
completed
}
}
`
const newTodo = await client.mutate(createTodoMutation, { variables: { label: 'Buy milk' } })
import { BlipsClient } from 'blips'
import resolvers from './resolvers'
import typeDefs from './types'
const initialState = {}
const client = new BlipsClient({ typeDefs, resolvers }, initialState, { fetch: { uri: 'http://localhost:3000/graphql' } })
const allTodosQuery = `
query allTodosQuery($first: Int) {
allTodos(first: $first) {
id
label
completed
}
}
`
// client.fetch(query [, options] [, operationName] )
const todos = await client.fetch(allTodosQuery, { variables: { first: 10 } })
You can extend the context provided to the resolvers in two ways:
1. When creating the client instance, by passing a context
object through
the options
argument. This will extend the default context will be available
to all your resolvers:
const client = new BlipsClient({ typeDefs, resolvers }, initialState, {
context: { foo: 'bar' },
})
2. Through the options
argument of every executor, which will make it
available only for that execution context:
client.query(allBookmarksQuery, { context: { user: loggedUser } })
example:
// types
const typeDefs = `
type Bookmarks {
id: Int!,
post: Post!,
user: User!
}
type Query {
allBookmarks: [Bookmark]!
}
`
// resolvers
const resolvers = {
Query: {
allBookmarks: (object, args, { store, user }) => {
// using identifier { user_id: user.id } to return only the current user's bookmarks
return store.get('bookmarks', { user_id: user.id })
// Returns:
// [
// { id: 1, post_id: 11, user_id: 23 },
// { id: 3, post_id: 155, user_id: 23 }
// ]
},
},
}
// initial state
const initialState = {
bookmarks: [
{ id: 1, post_id: 11, user_id: 23 },
{ id: 2, post_id: 356, user_id: 77 },
{ id: 3, post_id: 155, user_id: 23 },
],
}
const client = new BlipsClient({ typeDefs, resolvers }, initialState)
const allBookmarksQuery = `
query allBookmarksQuery {
allBookmarks {
id
post
user
}
}
`
const myBookmarks = client.query(allBookmarksQuery, {
context: { user: loggedUser },
})
This basics guide contains examples of using Blips with vanilla JS. The
preferred way to use with React is with
react-blips. Read the documentation
there and/or check out the *-react-blips
examples to get an
idea of how it works.
Since this is GraphQL
, some of your operations may result in multiple queries
for the same resource. Imagine the following scenario:
const state = {
comments: {
'3c4a086e-2151-4b54-acb2-13044ea553c1': {
id: '3c4a086e-2151-4b54-acb2-13044ea553c1',
message: 'Nice job!',
created: 1510737448000,
user_id: 'b1e7ed4d-7baa-4209-8ef3-0ccea2b420b0'
},
'f0ff08d2-9ea4-48b3-a77f-6b1517b5c827': {
id: 'f0ff08d2-9ea4-48b3-a77f-6b1517b5c827',
message: 'Great read!',
created: 1510737617221,
user_id: 'b1e7ed4d-7baa-4209-8ef3-0ccea2b420b0'
},
},
users: {
'b1e7ed4d-7baa-4209-8ef3-0ccea2b420b0'; {
id: 'b1e7ed4d-7baa-4209-8ef3-0ccea2b420b0',
firstName: 'John',
lastName: 'Doe',
comments: ['3c4a086e-2151-4b54-acb2-13044ea553c1', 'f0ff08d2-9ea4-48b3-a77f-6b1517b5c827']
// ...
}
}
}
const types = `
...
type Comment {
id: String!
message: String!
user: User!
created: Int!
}
...
`
const resolvers = {
// ...
Query: {
userCommentsQuery: (obj, { uid }, { store }) => store.get('comments', { user_id: uid }),
},
Comment: {
user: ({ user_id }, args, { store }) => store.get('users', user_id),
},
// ...
}
const userCommentsQuery = `
query userCommentsQuery($uid: String!) {
userComments(uid: $uid) {
idea
message
created
user {
id
firstName
photo
}
}
}
`
We have a comments
collection where each of the comments contain a user_id
field that is a reference to it's poster. When userCommentsQuery
is executed,
the user
field resolver for the comment will be called twice, even though it
will return the same user. That is bad design for a graphql server (querying
the database multiple times for the same resource), but it doesn't apply when
managing the client state. If you were to not add a resolver for the Comment
's
user
field, you would get the list of comments and then map over it to expand
the user data. That would still result in querying the store for the same
resource multiple times.
Now, because Blips manages all state asynchronously, we can use dataloader to batch and cache our resolvers, and that is a nice win and another benefit of managing your state asynchronously!
With graphql-tag
you can write your queries and type definitions in
.graphql|gql
files rather than .js
that export template literals. All you
need to do is add the loader to your webpack
config.
You can further split your queries and/or type definitions in multiple modules and use a package that will merge them before creating the client instance. See merge-graphql-schemas or gql-merge