An application store for dojo.
- Application state store designed to work with a reactive component architecture
- Out of the box support for asynchronous commands
- All state operations are recorded per process and undoable via middleware
- Supports the optimistic pattern with the ability to roll back on a failure
- Fully serializable operations and state
Dojo stores is a predictable, consistent state container for Javascript applications with inspiration from Redux and Flux architectures. However, the @dojo/framework/stores
package aims to provide more built-in support for common patterns such as asynchronous behaviors, undo support and more!
Managing state can become difficult to coordinate when an application becomes complicated with multiple views, widgets, components, and models. With each of these attempting to update attributes of state at varying points within the application lifecycle things can get confusing. When state changes are hard to understand and/or non-deterministic it becomes increasingly difficult to identify and reproduce bugs or add new features.
The @dojo/framework/stores
package provides a centralized store, designed to be the single source of truth for an application. It operates using uni-directional data flow. This means all application data follows the same lifecycle, ensuring the application logic is predictable and easy to understand.
Note: Do you need a centralized store? Lifting state up to parent widgets and using local state are likely to be sufficient in less complex applications.
To work with @dojo/framework/stores
there are three core but simple concepts - Operations, Commands, and Processes.
Operation
- Granular instructions to manipulate state based on JSON Patch
Command
- Simple functions that ultimately return operations needed to perform the required state change
Process
- A function that executes a group of commands that usually represent a complete application behavior
Operations are the raw instructions the store uses to make modifications to the state. The operations are based on the JSON Patch and JSON Pointer specifications that have been customized specifically for Dojo stores, primarily to prevent access to the state's root.
Dojo stores support four of the six JSON Patch operations: "add", "remove", "replace", and "test". The "copy" and "move" operations are not currently supported. Each operation is a simple object which contains instructions with the OperationType
, path
and optionally the value
(depending on operation).
import { OperationType } from '@dojo/framework/stores/State/patch';
import { Pointer } from '@dojo/framework/stores/state/Pointer';
const operations = [
{
op: OperationType.ADD,
path: new Pointer('/foo'),
value: 'foo'
},
{
op: OperationType.REPLACE,
path: new Pointer('/bar'),
value: 'bar'
},
{
op: OperationType.REMOVE,
path: new Pointer('/qux')
}
];
Dojo stores provides a helper package that can generate PatchOperation
objects from @dojo/framework/stores/state/operations
:
add
- Returns aPatchOperation
of typeOperationType.ADD
for thepath
andvalue
remove
- Returns aPatchOperation
of typeOperationType.REMOVE
for thepath
replace
- Returns aPatchOperation
of typeOperationType.REPLACE
for thepath
andvalue
test
- Returns aPatchOperation
of typeOperationType.TEST
for thepath
andvalue
These functions accept a Path
type. This is returned by the path
and at
methods on a store. More often this will be created using the path
and at
functions provided as part of the arguments to a command, which are described in more detail below. Rather than accepting a full string path, the path
and at
functions accept a series of arguments and provide type checking to verify that the path created actually exists on the state object.
Commands are simply functions which are called internally by the store when executing a Process
and return an array of PatchOperations
that tells the store
what state changes needs to be performed.
Each command is passed a CommandRequest
which provides path
and at
functions to generate Path
s in a typesafe way, a get
function for access to the store's state, and a payload
object for the argument that the process executor was called with.
The get
function returns back state for a given Path
, for example get(path('my', 'deep', 'state'))
or get(at(path('my', 'array', 'item'), 9))
.
function addTodoCommand({ at, path, get, payload }: CommandRequest) {
const todosPath = path('todos');
const length = get(todosPath).length;
const operations = [add(at(todosPath, length), payload)];
return operations;
}
function calculateCountsCommand({ get, path }: CommandRequest) {
const todos = get(path('todos'));
const completedTodos = todos.filter((todo: any) => todo.completed);
const operations = [
replace(path('activeCount'), todos.length - completedTodos.length),
replace(path('completedCount'), completedTodos.length)
];
return operations;
}
A Command
, or the CommandRequest
argument to it, can be provided with generics that indicate the type of the state of the store they are intended to target and the payload that will be passed. This will provide type checking for all calls to path
and at
and usages of payload
, ensuring that the operations will be targeting real properties of the store, and providing type inference for the return type of get
.
interface MyState {
id: string;
}
interface Payload {
id: string;
}
const command = (request: CommandRequest<MyState, Payload>) => {
return [
add(path('id'), request.payload.id);
];
};
In order to avoid typing each Command
explicitly, a CommandFactory
can be created that will pass its generic types onto all commands it creates. Creating commands using a factory is essentially the same as creating them without one. It is simply a convenience to avoid repeating the same typings for each command.
interface Todo {
completed: boolean;
label: string;
}
interface TodoState {
todos: Todo[];
activeCount: number;
completedCount: number;
}
const createCommand = createCommandFactory<TodoState>();
const addTodoCommand = createCommand(({ at, get, path, payload }) => {
const todos = get(path('todos'));
// const todos = get(path('todoes')); Fails to compile because `todoes` is not a property on the state
// const value = todos + 3; Fails to compile because todos is typed as inferred to be an array of `Todo` objects
const operations = [
// Using the utilities provided by the `operations` module ensures that the paths provided are valid,
// and that the values being added or replaced are of the appropriate type
add(at(path('todos'), todos.length), payload)
];
return operations;
});
const calculateCountsCommand = createCommand(({ get, path }) => {
const todos = get(path('todos'));
const completedTodos = todos.filter((todo: any) => todo.completed);
const operations = [
replace(path('activeCount'), todos.length - completedTodos.length),
replace(path('completedCount'), completedTodos.length)
];
return operations;
});
Important: Access to state root is not permitted and will throw an error, for example, get(path('/'))
. This applies to Operations
also, it is not possible to create an operation that will update the state root. Best practices with @dojo/framework/stores mean touching the smallest part of the store as is necessary.
Commands support asynchronous behavior out of the box simply by returning a Promise<PatchOperation[]>
.
async function postTodoCommand({ get, path, payload: { id } }: CommandRequest): Promise<PatchOperation[]> {
const response = await fetch('/todos');
if (!response.ok) {
throw new Error('Unable to post todo');
}
const json = await response.json();
const todos = get(path('todos'));
const index = findIndex(todos, byId(id));
// success
return [
replace(at(path('todos'), index), {
...todos[index],
loading: false,
id: data.uuid
})
];
}
In modern browsers, commands can also directly modify a state
object that is passed in as part of the CommandRequest.
Any modifications to this object will be translated into the appropriate patch operations and executed against the store.
Commands that modify the state object can be synchronous or asynchronous, but if they are asynchronous they must return
a promise that will resolve when modifications are complete. Note that attempting to access state
is not supported in IE
and will immediately throw an error.
function calculateCountsCommand = createCommand(({ state }) => {
const todos = state.todos;
const completedTodos = todos.filter((todo: any) => todo.completed);
state.activeCount = todos.length - completedTodos.length;
state.completedCount = completedTodos.length;
});
async function postTodoCommand({ state }: CommandRequest): Promise<PatchOperation[]> {
const response = await fetch('/todos');
if (!response.ok) {
throw new Error('Unable to post todo');
}
const json = await response.json();
const todos = state.todos
const index = findIndex(todos, byId(id));
// success
state.todos[index] = {
...todos[index],
loading: false,
id: json.uuid
};
}
A Process
is the construct used to execute commands against a store
instance in order to make changes to the application state. Processes
are created using the createProcess
factory function that accepts an array of commands and an optional callback that can be used to manage errors thrown from a command. The optional callback receives an error
object and a result
object. The error
object contains the error
stack and the command that caused the error.
The result object contains the following:
- The
payload
passed to the process undoOperations
to undo theprocess
- A function to execute an additional
process
.
The array of Commands
are executed in sequence by the store until the last Command is completed or a Command
throws an error. These processes often represent an application behavior. For example, adding a todo in a simple todo application which will be made up with multiple discreet commands.
A simple process
to add a todo and recalculate the todo count:
const addTodoProcess = createProcess('add-todo', [addTodoCommand, calculateCountCommand]);
An after middleware
can be provided which will be called when an error occurs or the process is successfully completed:
const addTodoProcessMiddleware = () => {
return {
after: (error, result) => {
if (error) {
// do something with the error
// possibly undo the operations
result.store.apply(result.undoOperations);
}
// possible additional state changes by running another process using result.executor(otherProcess)
};
};
};
const addTodoProcess = createProcess('add-todo', [addTodoCommand, calculateCountCommand], [ addTodoProcessMiddleware ]);
The Process
creates a deferred executor by passing the store
instance addTodoProcess(store)
which can be executed immediately by passing the payload
, addTodoProcess(store)(payload)
. Or more often passed to your widgets and used to initiate state changes on user interactions. The payload
argument for the executor
is required and is passed to each of the Process
's commands in a payload
argument.
const addTodoExecutor = addTodoProcess(store);
addTodoExecutor({
foo: 'arguments',
bar: 'get',
baz: 'passed',
qux: 'here'
});
Initial state can be defined on store creation by executing a Process
after the store has been instantiated.
// Command that creates the basic initial state
const initialStateCommand = createCommand({ path }) => {
return [
add(path('todos'), []),
add(path('currentTodo'), ''),
add(path('activeCount'), 0),
add(path('completedCount'), 0)
]);
}
const initialStateProcess = createProcess('initial', [ initialStateCommand ]);
// creates the store, initializes the state and runs the `getTodosProcess` shown below.
const store = createStore();
initialStateProcess(store)();
// if a process contains an async command, like fetching initial data from a remote service the return promise can be used
// to control the flow.
getTodosProcess(store)().then(() => {
// do things once the todos have been fetched.
});
The payload
argument for the process executor can be specified as the second generic type when using createProcess
const process = createProcess<any, { foo: string }>('foo', [command]);
const processExecutor = process(store);
// The executor will require an argument that satisfies `{ foo: string }`
processExecutor({ foo: 'bar' });
processExecutor({ foo: 1 }); // Compile error
The process executor's payload
type will also be inferred by the payload
type of the commands if not specified explicitly, however, the payload
type for all the commands must be assignable when this is not the case the payload generic type needs to be explicitly passed.
const createCommandOne = createCommandFactory<any, { foo: string }>();
const createCommandTwo = createCommandFactory<any, { bar: string }>();
const commandOne = createCommandOne(({ get, path, payload }) => []);
const commandTwo = createCommandTwo(({ get, path, payload }) => []);
const processOne = createProcess('one', [commandOne]);
const executorOne = processOne(store); // payload for executor inferred based on `commandOne`
executorOne({ foo: 'foo' });
executorOne({ bar: 'foo' }); // compile error
// compile error as payload types for commandOne and commandTwo are not assignable
const processTwo = createProcess('two', [commandOne, commandTwo]);
// Explicitly passing a generic that satisfies all the command payload types enables payload type widening
const processTwo = createProcess<any, { foo: string; bar: string }>('two', [commandOne, commandTwo]);
const executorTwo = processTwo(store);
executorTwo({ foo: 'foo' }); // compile error, as requires both `bar` and `foo`
executorTwo({ foo: 'foo', bar: 'bar' }); // Yay, valid
Alternatively, the payload can be typed at command creation
const createCommandOne = createCommandFactory<MyState>();
const commandOne = createCommandOne<{ foo: string }>(({ get, path, payload }) => []);
Although Dojo stores is a big atom state store, you never get access to the entire state object. To access the sections of state that are needed we use pointers to return the slice of state that is needed i.e. path('path', 'to', 'state')
. State is never directly updated by the user, with state changes only being processed by the operations returned by commands.
There is no concept of reducers
, meaning that there is no confusion about where logic needs to reside between reducers
and actions
. Commands
are the only place that state logic resides and return operations
that dictate what state
changes are required and processed internally by the store
.
Additionally, this means that there is no need to coordinate actions
and reducers
using a string action key. Commands are simple function references that can be reused in multiple processes
.
Store data can be connected to widgets within your application using the StoreProvider
widget provided by @dojo/framework/stores
.
Container Property API:
renderer
: A render function that has the store injected in order to access state and pass processes to child widgets.stateKey
: The key of the state in the registry.paths
(optional): A function to connect theContainer
to sections of the state.
There are two mechanisms to connect the StoreProvider
to the Store
:
- The recommended approach is to register
paths
on container creation to ensure invalidation will only occur when state you are interested in changes. - A catch-all when no
paths
are defined for the container, it will invalidate when any data changes in the store.
import { WidgetBase } from '@dojo/framework/core/WidgetBase';
import { Store } from '@dojo/framework/stores/Stores';
import StoreProvider from '@dojo/framework/stores/StoreProvider';
interface State {
foo: string;
bar: {
baz: string;
};
}
class MyApp extends WidgetBase {
protected render() {
return w(StoreProvider, {
stateKey: 'state',
renderer: (store: Store<State>) => {
return v('div', [store.get(store.path('foo'))]);
}
});
}
}
A pre-typed container can be created by extending the standard StoreProvider
and passing the State
type as a generic.
interface State {
foo: string;
}
export class MyTypeStoreProvider extends StoreProvider<State> {}
However in order for TypeScript to infer this correctly when using w()
, the generic will need to be explicitly passed.
w<MyTypeStoreProvider>(MyTypeStoreProvider, {
stateKey: 'state',
renderer(store) {
return v('div', [store.get(store.path('foo'))]);
}
});
To be notified of changes in the store, use the onChange()
function, which takes a path
or an array of path
's and a callback for when that portion of state changes, for example:
store.onChange(store.path('foo', 'bar'), () => {
// do something when the state at foo/bar has been updated.
});
or
store.onChange([store.path('foo', 'bar'), store.path('baz')], () => {
// do something when the state at /foo/bar or /baz has been updated.
});
In order to be notified when any changes occur within the store's state, simply register to the stores .on()
method for a type of invalidate
passing the function to be called.
store.on('invalidate', () => {
// do something when the store's state has been updated.
});
An optional transformer
can be passed to a process that is used to transform the executor
's payload to the command
payload type. The return type of the transformer
must match the command
payload
type of the process
. The argument type of the executor
is inferred from transformers payload
type.
interface CommandPayload {
bar: number;
foo: number;
}
// `CommandPayload` types the command payload and the argument of the process executor
const createCommand = createCommandFactory<any, CommandPayload>();
// `payload` is typed to `CommandPayload`
const command = createCommand(({ get, path, payload }) => {});
const process = createProcess('example', [command]);
interface TransformerPayload {
foo: string;
}
// The transformer return type must match the original `CommandPayload`
const transformer = (payload: TransformerPayload): CommandPayload => {
return {
bar: 1,
foo: 2
};
};
// when no transformer passed to the process the executor `payload` is typed to `CommandPayload`
const processExecutor = process(store);
// Works
processExecutor({ bar: 1, foo: 2 });
// These shouldn't work as `payload` does not match `CommandPayload` interface
processExecutor({ bar: '', foo: 2 });
processExecutor({ foo: '' });
// when a `transformer` passed to the process the `transformer` return type must match `CommandPayload`
// and the executor `payload` type becomes `payload` type of the `transformer`
const processExecutor = process(store, transformer);
// Works as the `transformer` payload type is `{ foo: string }`
processExecutor({ foo: '' });
// Shouldn't work as `payload` does not match the `TransformerPayload` type
processExecutor({ bar: 1, foo: 2 });
processExecutor({ bar: 1, foo: '' });
Each Command
will be passed the result of the transformer as the payload
for example: { bar: 1, foo: 2 }
Optimistic updating can be used to build a responsive UI despite interactions that might take some time to respond, for example saving to a remote resource.
In the case of adding a todo item for instance, with optimistic updating, we can immediately add the todo before we even make a request to the server and avoid having an unnatural waiting period or loading indicator. When the server responds we can then reconcile the outcome based on whether it is successful or not.
In the success scenario, we might need to update the added Todo item with an id that was provided in the response from the server and change the color of the Todo item to green to indicate it was successfully saved.
In the error scenario, it might be that we want to show a notification to say the request failed and turn the Todo item red, with a "retry" button. It's even possible to revert/undo the adding of the Todo item or anything else that happened in the process.
const handleAddTodoErrorProcess = createProcess('error', [ () => [ add(path('failed'), true) ]; ]);
const addTodoErrorMiddleware = () => {
return {
after: () => (error, result) {
if (error) {
result.store.apply(result.undoOperations);
result.executor(handleAddTodoErrorProcess);
}
}
};
};
const addTodoProcess = createProcess('add-todo', [
addTodoCommand,
calculateCountsCommand,
postTodoCommand,
calculateCountsCommand
],
[ addTodoCallback ]);
addTodoCommand
: Adds the new todo into the application statecalculateCountsCommand
: Recalculates the count of completed and active todo itemspostTodoCommand
: posts the todo item to a remote service and using the process after middlware we can make changes if there is a failure- on failure: the previous two commands are reverted and the
failed
state field is set totrue
- on success: Returns operations that update the todo item
id
field with the value received from the remote service
- on failure: the previous two commands are reverted and the
calculateCountsCommand
: Runs again after the success ofpostTodoCommand
To support "pessimistic" updates to the application state, i.e. wait until a remote service call has been completed before changing the application state simply put the async command before the application store update. This can be useful when performing a deletion of a resource, when it can be surprising if an item is removed from the UI "optimistically" only for it to reappear back if the remote service call fails.
function byId(id: string) {
return (item: any) => id === item.id;
}
async function deleteTodoCommand({ get, payload: { id } }: CommandRequest) {
const { todo, index } = find(get('/todos'), byId(id));
await fetch(`/todo/${todo.id}`, { method: 'DELETE' });
return [remove(path('todos', index))];
}
const deleteTodoProcess = createProcess('delete', [deleteTodoCommand, calculateCountsCommand]);
Note: The process requires the counts to be recalculated after successfully deleting a todo, the process above shows how easily commands can be shared and reused.
A Process
supports concurrent execution of multiple commands by specifying the commands in an array when creating the process:
const myProcess = createProcess('my-process', [commandOne, [concurrentCommandOne, concurrentCommandTwo], commandTwo]);
In this example, commandOne
is executed, then both concurrentCommandOne
and concurrentCommandTwo
are executed concurrently. Once all of the concurrent commands are completed the results are applied in order before continuing with the process and executing commandTwo
.
Note: Concurrent commands are all assumed to be asynchronous and resolved using Promise.all
if any returns a Promise
.
Processing operations and updating the store state is handled by an implementation of the MutableState
interface
defined in Store.ts
. This interface defines four methods necessary to properly apply operations to the state.
get<S>(path: Path<M, S>): S
Takes aPath
object and returns the value in the current state that that path points toat<S extends Path<M, Array<any>>>(path: S, index: number): Path<M, S['value'][0]>
Returns aPath
object that points to the providedindex
in the array at the providedpath
path: StatePaths<M>
A typesafe way to generate aPath
object for a given path in the stateapply(operations: PatchOperation<T>[]): PatchOperation<T>[]
Apply the provided operations to the current state
The default state implementation is reasonably optimized and in most circumstances will be sufficient. If a particular use case merits an alternative implementation it can be provided to the store constructor
const store = new Store({ state: myStateImpl });
An implementation of the MutableState
interface that leverages Immutable under the hood is provided as an example. This implementation may provide better performance if there are frequent, deep updates to the store's state, but performance should be tested and verified for your app before switching to this implementation.
Middleware provides a hook to apply generic/global functionality across multiple or all processes used within an application. Middleware is a function that returns an object with optional before
and after
callback functions.
Multiple middlewares can be provided, and in this case any before
callbacks will be called in the order they are provided, the process will run, and then after
callbacks will be called in the order provided.
If provided, the after
callback will be passed the error and the result of a process to perform a specific action. If multiple middlewares are provided, their after
callbacks will be
executed in the order they were provided.
In the example below, the after
callbacks returned by callback, logger, and snapshot will be called in that order.
const myProcess = createProcess('my-process', [commandOne, commandTwo], [callback, logger, snapshot]);
If provided, the before
callback functions will be passed payload passed to the process and a reference to the store. They can be synchronous or asynchronous, designated by returning a promise,
but if they are asynchronous they will make any processes they are attached to asynchronous as well.
In the example below, the before
callbacks returned by authenticator and logger would be called in that order.
const myProcess = createProcess('my-process', [commandOne, commandTwo], [authenticator, logger]);
Specifying a middleware on an individual process explicitly works for targeted behavior but can become cumbersome when the middleware needs to be applied to multiple processes throughout the application.
The createProcessWith
higher order function can be used to specify middlewares that need to be applied across multiple processes
. The function accepts an array of middleware and returns a new createProcess
factory function that will automatically apply the middleware to any process that it creates.
const customCreateProcess = createProcessWith([logger]);
// `myProcess` will automatically be decorated with `logger` middleware.
const myProcess = customCreateProcess('my-process', [commandOne, commandTwo]);
The example below demonstrates the usage of both before
and after
callbacks
const logger: ProcessCallback = () => {
const beforeLogs: string[] = [];
const afterLogs: string[] = [];
return {
before(payload, store) {
const value = `${beforeLogs.length}th payload: ${JSON.stringify(payload)}`;
beforeLogs.push(value);
store.apply([
{
op: OperationType.ADD,
path: new Pointer(`/beforeLogs/${beforeLogs.length}`),
value
}
]);
console.log('logger before called');
},
after(error, result) {
const value = error ? 'Process failed' : 'Process Succeeded';
afterLogs.push(value);
result.apply([
{
op: OperationType.ADD,
path: new Pointer(`/afterLogs/${afterLogs.length}`),
value
}
]);
console.log('logger after called');
}
};
};
const afterOnly: ProcessCallback = () => ({
after(error, result) {
console.log('after only called');
}
});
const beforeOnly: ProcessCallback = () => ({
before(payload, store) {
console.log('before only called');
}
});
const callback: ProcessCallback = () => ({
before(payload, store) {
console.log('callback before called');
},
after(error, result) {
console.log('callback after called');
}
});
const myProcess = createProcess('my-process', [commandOne, commandTwo], [logger, afterOnly, beforeOnly, callback]);
// Subsequent usages of this process will log the following to console:
// logger before called
// before only called
// callback before called
// logger after called
// after only called
// callback after called
Middleware that provides a collector
that saves state to LocalStorage
and a load
function to hydrate a store from LocalStorage
.
export const myProcess = createProcess(
'my-process',
[command],
collector('my-process', (path) => {
return [path('state', 'to', 'save'), path('other', 'state', 'to', 'save')];
})
);
import { load } from '@dojo/framework/stores/middleware/localStorage';
import { Store } from '@dojo/framework/stores/Store';
const store = new Store();
load('my-process', store);