This package is kept for historical reasons but is no longer maintained. You can instead use the Kingly state machine library which implements the sought-for architecture and patterns. The reasons for the deprecation are:
- The design of
xstate
makes it really hard to build a functional layer on top of it.xstate
seems to favor non-functional usage as it links itself to the SCXML standard. SCXML goals and design (in particular the choice of XML) reflect the interests of the telecommunications industry at the time of creation (speech processing, multi-modal interactions mostly). SCXML is oriented to process control, and processes can not be manipulated like functions. - The cost associated to the
xstate
library (15Kb in some cases) outweighs the benefits. On top of that, adding an interpreting layer that deals with the API surface and complexity ofxstate
compounds the problem. Conversely, the Kingly state machine library compiles an average state machine to below 1 KB JavaScript with zero dependencies. In practice, the extra functionalities proposed by the SCXML-orientedxstate
(actors, activities, etc.) can be replicated without the coupling and extending the API surface. Cf. Kingly's documentation for examples. - The interpreter layer is fragile as changes in
xstate
means maintenance tasks on the side of this library. A better design avoids unnecessary dependencies.
The xstate statecharts library has a few interpreters
already available with miscelleanous design goals. In order to integrate xstate
with React
through our react-state-driven
component, we needed to adapt the xstate interface to the interface for our Machine
component.
This is so because our component achieves a decoupling of the state machine on one side, and the
React component on the other side, but also separates out the event handling, state
representation and effect execution concerns out of the library. The usual technique of
programming to an interface instead of to an implementation is used to that purpose, which in
turn means that the external concerns have to abide by the interface set by the Machine
component.
It turns out that we could not reuse the default interpreter and other existing
interpreters for xstate
as they do not synchronously return the list of computed actions in
response to an input (listeners are instead used). Additionally some interpreters may also
produce effects which in our design is forbidden. As a result we created an interpreter which :
- matches the required interface to integrate xstate in React through our
Machine
component - computes and returns a list of actions in response to an input
- does not produce any effects
The benefits are the following :
- we were able to use
json patch
andimmer
for the state representation concern.Immutable.js
could also be used via simple interface adaptation. If that is your use case, you may even interface a reducer which updates state in place. How you update state is not a concern of theMachine
component - similarly (
xstate
machine library + this interpreter) can be replaced by the nativereact-state-driven
library, or any other machine interpreter satisfying the required interface - event handling can be done with the event library of your choice, by writing an adapter for the
accepted event handling interface (we did that for
rxjs
andmost
so far) - effect execution being separated out of the machine, it is easy to mock and stub effects for testing purposes. This will allow to enjoy the benefits of automated testing without breaking glasses.
- and React aside, we were able to integrate the interpreter with
cyclejs
with no major effort! Integration withAngular2
is in progress but seems to be going the same painless way. After all, the interpreter is just a function!
import { applyPatch } from "json-patch-es6"
// The machine may produce several outputs when transitioning, they have to be merged
const mergeOutputs = function (accOutputs, outputs) {
return (accOutputs || []).concat(outputs)
};
// The machine produces actions to update its extended state, the reducer executes those actions
const jsonPatchReducer = (extendedState, extendedStateUpdateOperations) => {
return applyPatch(extendedState, extendedStateUpdateOperations, false, false).newDocument;
};
const actionFactoryMaps = {
stringActions: {
'cancelAdmin': (extendedState, event) => {
return {
updates: [{ op: 'add', path: '/isAdmin', value: false }],
outputs: ['admin rights overriden']
}
}
},
};
// xstate machine
const hierarchicalMachine = {
context: { isAdmin: true },
id: 'door',
initial: 'closed',
states: {
closed: {
initial: 'idle',
states: {
'idle': {},
'error': {
onEntry: function logMessage(extS, ev) {return { updates: [], outputs: ['Entered .closed.error!', ev] }}
}
},
// NOTE : test input sequence : ['OPEN', {'CLOSE', overrideAdmin:true}, 'OPEN']
on: {
OPEN: [
{ target: 'opened', cond: (extState, eventObj) => extState.isAdmin },
{ target: 'closed.error' }
]
}
},
opened: {
on: {
CLOSE: [
{ target: 'closed', cond: (extState, eventObj) => eventObj.overrideAdmin, actions: ['cancelAdmin'] },
{ target: 'closed', cond: (extState, eventObj) => !eventObj.overrideAdmin }
]
}
},
}
};
// Test paraphernalia
export const testCases = {
HierarchicalMachineAndJSONPatchAndFunctionActionsAndObjectEvents: {
description: '(hierarchical, json patch, mergeOutput, action functions and strings, event as object, >1 inputs)',
machine: hierarchicalMachine,
updateState: reducers.jsonpatchReducer,
actionFactoryMap: actionFactoryMaps.stringActions,
mergeOutputs,
inputSequence: ['OPEN', { type: 'CLOSE', overrideAdmin: true }, 'OPEN'],
outputSequence: [null, ['admin rights overriden'], ["Entered .closed.error!", "OPEN"]]
},
}
QUnit.test("(hierarchical, json patch, mergeOutput, action functions and strings, event as object, >1 inputs)", function exec_test(assert) {
const testCase = testCases.HierarchicalMachineAndJSONPatchAndFunctionActionsAndObjectEvents;
const machineConfig = testCase.machine;
const interpreterConfig = {
updateState: testCase.updateState,
mergeOutputs: testCase.mergeOutputs,
actionFactoryMap: testCase.actionFactoryMap,
};
const interpreter = xstateReactInterpreter(Machine, machineConfig, interpreterConfig);
const testScenario = testCase.inputSequence;
const actualTestResults = testScenario.map(interpreter.yield);
const expectedTestResults = testCase.outputSequence;
testScenario.forEach((input, index) => {
assert.deepEqual(
actualTestResults[index],
expectedTestResults[index],
testCase.description
);
});
assert.ok(testCase.machine.context === initialContextHierarchicalMachine, `json patch does not mutate state in place`);
});
What happens here :
- the machine starts in the configured initial state with the configured extended state
- we made the following choices for our interpreter :
- use json patch for immutable state update
- outputs of the machines are arrays
- those arrays will be merged by simple concatenation
- we send a
OPEN
input to the machine which triggers :- because the guard is satisfied, and there is no actions defined, the machine will move to the
opened
control state, and outputsnull
, which is the value chosen for indicating that there is no output.
- because the guard is satisfied, and there is no actions defined, the machine will move to the
- we send an object input
{ type: 'CLOSE', overrideAdmin: true }
to the machine :- because
overrideAdmin
property is set in the event object, the transition chosen triggers thecancelAdmin
action, and the entry in theclosed
control state. ThecancelAdmin
action consists of updating theisAdmin
property of the extended state of the machine tofalse
. The machine outputs are[admin rights overrriden]
.
- because
- we then send the input
'OPEN'
to the machine :- because the property
isAdmin
is no longer set on the extended state, the machine will transition to the'closed.error'
control state. On entering that state, the machine will outputs as configured['Entered .closed.error!', ev]
withev
being"OPEN"
- because the property
In short, we have shown :
mergeOutputs
andupdateState
configuration- how to map action strings to action factories through the mapping object
actionFactoryMap
- how to directly include action factory in the xstate machine
- action factories produce two pieces of information to the interpreter :
- how to update the machine's extended state
- what are the machine outputs
Contrary to other interpreters, the interpreter does not interpret effects. In our React integration design, that responsibility is delegated to the command handler. The interpreter simply advances the machines, thereby updating the machine state, and producing the machine's outputs. The state of the machine is hence completely encapsulated and cannot be accessed from the outside. Our interpreter is just a function producing outputs in function of the state of the underlying machine. In our React machine component design, those outputs are commands towards to the interfaced systems.
const xStateRxAdapter = {
subjectFactory: () => new Rx.Subject(),
// NOTE : must be bound, because, reasons
merge: Rx.Observable.merge.bind(Rx.Observable),
create: fn => Rx.Observable.create(fn)
};
const showXstateMachine = machine => {
const interpreterConfig = {
updateState: machine.updateState,
mergeOutputs: machine.mergeOutputs,
actionFactoryMap: machine.actionFactoryMap,
};
const fsm = xstateReactInterpreter(xstateMachineFactory, machine.config, interpreterConfig);
return React.createElement(Machine, {
eventHandler: xStateRxAdapter,
preprocessor: machine.preprocessor,
fsm: fsm,
commandHandlers: machine.commandHandlers,
componentWillUpdate: (machine.componentWillUpdate || noop)(machine.inject),
componentDidUpdate: (machine.componentDidUpdate || noop)(machine.inject)
}, null)
};
// Displays all machines (not very beautifully, but this is just for testing)
ReactDOM.render(
div([
showXstateMachine(xstateMachines.xstateImageGallery)
]),
document.getElementById('root')
);
export const NO_IMMER_UPDATES = nothing;
export const immerReducer = function (extendedState, updates) {
if (updates === NO_IMMER_UPDATES) return extendedState
const updateFn = updates;
return produce(extendedState, updateFn)
};
export const mergeOutputs = function (accOutputs, outputs) {
return (accOutputs || []).concat(outputs || [])
};
export const xstateMachines = {
xstateImageGallery: {
preprocessor: rawEventSource => rawEventSource
.startWith([INIT_EVENT])
.map(ev => {
const { rawEventName, rawEventData: e, ref } = destructureEvent(ev);
if (rawEventName === INIT_EVENT) {
return { type: INIT_EVENT, data : void 0}
}
// Form raw events
else if (rawEventName === 'onSubmit') {
e.persist();
e.preventDefault();
return { type: 'SEARCH', data: ref.current.value }
}
else if (rawEventName === 'onCancelClick') {
return { type: 'CANCEL_SEARCH', data: void 0 }
}
// Gallery
else if (rawEventName === 'onGalleryClick') {
const item = e;
return { type: 'SELECT_PHOTO', data: item }
}
// Photo detail
else if (rawEventName === 'onPhotoClick') {
return { type: 'EXIT_PHOTO', data: void 0 }
}
// System events
else if (rawEventName === 'SEARCH_SUCCESS') {
const items = e;
return { type: 'SEARCH_SUCCESS', data: items }
}
else if (rawEventName === 'SEARCH_FAILURE') {
return { type: 'SEARCH_FAILURE', data: void 0 }
}
return NO_INTENT
})
.filter(x => x !== NO_INTENT)
,
// DOC : we kept the same machine but :
// - added the render actions
// - render must go last, in order to get the updated extended state
// - added an init event to trigger an entry on the initial state
config: {
context: { query: '', items: [], photo: undefined, gallery: '' },
initial: 'init',
states: {
init: {
on: { [INIT_EVENT]: 'start' }
},
start: {
onEntry: [renderGalleryAppImmer('start')],
on: { SEARCH: 'loading' }
},
loading: {
onEntry: ['search', renderGalleryAppImmer('loading')],
on: {
SEARCH_SUCCESS: { target: 'gallery', actions: ['updateItems'] },
SEARCH_FAILURE: 'error',
CANCEL_SEARCH: 'gallery'
}
},
error: {
onEntry: [renderGalleryAppImmer('error')],
on: { SEARCH: 'loading' }
},
gallery: {
onEntry: [renderGalleryAppImmer('gallery')],
on: {
SEARCH: 'loading',
SELECT_PHOTO: 'photo'
}
},
photo: {
onEntry: ['setPhoto', renderGalleryAppImmer('photo')],
on: { EXIT_PHOTO: 'gallery' }
}
}
},
actionFactoryMap: {
'search': (extendedState, { data: query }, xstateAction) => {
const searchCommand = { command: COMMAND_SEARCH, params: query };
return {
outputs: [searchCommand],
updates: nothing
}
},
'updateItems': (extendedState, { data: items }, xstateAction) => {
return {
updates: extendedState => {extendedState.items = items},
outputs: NO_OUTPUT
}
},
'setPhoto': (extendedState, { data: item }, xstateAction) => {
return {
updates: extendedState => {extendedState.photo = item},
outputs: NO_OUTPUT
}
}
},
updateState: immerReducer,
mergeOutputs: mergeOutputs,
commandHandlers: {
[COMMAND_SEARCH]: (trigger, query) => {
runSearchQuery(query)
.then(data => {
trigger('SEARCH_SUCCESS')(data.items)
})
.catch(error => {
trigger('SEARCH_FAILURE')(void 0)
});
}
},
inject: new Flipping(),
componentWillUpdate: flipping => (machineComponent, prevProps, prevState, snapshot, settings) => {flipping.read();},
componentDidUpdate: flipping => (machineComponent, nextProps, nextState, settings) => {flipping.flip();}
}
npm xstate-interpreter
npm run test
The factory xstateReactInterpreter
returns an interpreter with a yield
function by which
inputs will be sent to the machine and outputs will be collected. It also returns an instance of
the executable state machine.
- the machine is initialized per its configuration and specifications
- the interpreter returns a
yield
function to call the machine with an input - the machine's actions are in fine functions (termed action factories);
- whose input parameters are the machine's extended state and event
- which return :
- a description of the updates to perform on its extended state as a result of the transition
- the outputs for the state machine as a result of receiving the input
- on transitioning, the machine produces
updates
andoutputs
. The interpreter :- perform actual updates on the machine's extended state, according to the
updateState
configured reducer - outputs from the machine's triggered action factories are merged with the configured
mergeOutputs
and returned
- perform actual updates on the machine's extended state, according to the
JSDoc types available is /src/types
:
/**
* @typedef {function(ExtendedState, ExtendedStateUpdate): ExtendedState} ExtendedStateReducer
*/
/**
* @typedef {*} Output
*/
/**
* @typedef {Container<Output>} Outputs
* `Container` is a foldable functor, for instance `Array`
*/
/**
* @typedef {function(Outputs, Outputs): Outputs} OutputReducer
*/
/**
* @typedef {String} xStateActionType
* The type of xstate action. In xstate v4.0, this is the property `type` of the xstate action
*/
/**
* @typedef {*} xStateEvent
* cf. xState types. Usually either a string or an object with a `type` property which is a string
*/
/**
* @typedef {*} xstateAction
* cf. xState types. Usually an object with at least a `type` and `exec` property
* The exec property when set (i.e. truthy) holds an action factory function.
* The `type` property when set holds an identifier used to map to an action factory
*/
/**
* @typedef {function(ExtendedState, xStateEvent, xstateAction):x} xStateActionFactory
* The type of xstate action. In xstate v4.0, this is the property `type` of the xstate action
*/
/**
* @typedef {Object} interpreterConfig
* @property {ExtendedStateReducer} updateState
* @property {OutputReducer} mergeOutputs
* @property {Object.<xStateActionType, xStateActionFactory>} actionFactoryMap
*/
updateState
andmergeOutput
should be pure, monoidal operations- i.e. with an empty value, and associativity properties
- all functions involved in the machine and interpreter configuration should be pure functions
- if you use a function as xstate action, that function must be a named function!!
- type contracts
- integrating
xstate-interpreter
withreact-state-driven
means that the xstate machine will receive an init event. This means a dummy initial state and an init transition should be configured towards the real initial state of the machine. Alternaively, the machine can be configured to simply ignore unaccepted events. In any case, the xstate machine cannot reuse the reserved initial event.
- activities and delays are not currently interpreted
xstate
has automatically configured actions (logs, assign, invoke, etc). If you use them you will have to define a matching action factory. Our interpreter comes without any predefined action factory.- you can specify xstate actions as strings or functions or objects. I recommend to pick up your poison instead of juggling with 3 different types. (Named) Functions are the best option in my eyes, provided they do not prevent the machine visualizer from doing its job.
- the second parameter of the xstate machine factory i.e.
actions
is absorbed into the configuration of the interpreter to avoid confusion or duplication - if the machine does not have any actions configured for an occurring transition, it outputs
a constant indicating that there is no output (in this version the constant is
null
). The machine being a function, always outputs something as a result of being called.