State machines have been around forever, and still there's no better way to graphically represent application logic
By offloading your application state management into a platform-agnostic library like XState, you can render it out into a chart, use it to generate automated tests, avoid framework lock-in, and get all the benefits that a finite state machine can offer.
Here we'll step through some progressively more featureful examples of a state machine for a form. None of these are more correct than the others — instead, each represents a specific set of business logic and it's up to you to decide what level of state you want to support for your application.
Submit only with no validation
The most basic form has no client side validation.
The user hits submit. If the submission succeeds, we let the user know their submission has been received. If the submission fails, we allow them to retry.
The logic here is all focused around the submission of the form, rather than the time spent filling out the form. Since there's no required fields, and no conditions that stop us from submitting, all states up until the submitting state are identical from this machine's point of view.
With a machine like this, any validation would be done server-side, and an invalid form would cause the submit to fail and transition to a failure state.
// View in the XState Visualizer
// https://xstate.js.org/viz/?gist=6e7e6b7258c805c85ea273b03172605d
const formMachine = Machine({
id: 'form',
initial: 'editing',
context: {
retries: 0
},
states: {
editing: {
on: {
SUBMIT: {
target: 'submitting'
}
}
},
submitting: {
on: {
RESOLVE: 'success',
REJECT: 'failure'
}
},
success: {
type: 'final'
},
failure: {
on: {
RETRY: {
target: 'submitting',
actions: 'incrementRetries'
}
}
}
}
}, {
actions: {
incrementRetries: assign((context, event) => context.retries + 1)
}
})
Submit only if input is valid
Even in a world where server-side validation is free and easy, we're still going to want some basic client side validation. It doesn't make sense to allow users to submit empty forms, so let's add one input.
Consider a form with a single input for a message, and we want to ensure that the message is at least ten characters long before submitting.
There are a few new requirements our machine needs to implement in order to support that feature:
- a context value for the message
- an INPUT event that caches our message text in the machine context
- a guard on the SUBMIT event that stops the transition if our message is empty or too short
// View in the XState Visualizer
// https://xstate.js.org/viz/?gist=365c0512bcdd1d187a8fda64f4211daa
const formMachine = Machine({
id: 'form',
initial: 'editing',
context: {
retries: 0,
+ message: ''
},
states: {
editing: {
on: {
SUBMIT: [
+ {
+ target: 'editing',
+ cond: 'isMessageTooShort'
+ },
{
target: 'submitting'
}
],
INPUT: {
target: 'editing',
actions: 'cache'
}
}
},
…
}
}, {
actions: {
cache: assign((context, event) => event),
incrementRetries: assign((context, event) => context.retries + 1)
},
+ guards: {
+ isMessageTooShort: (context) => context.message.length < 10
+ }
})
Inputs with an error state
Forms that fail without telling you why are categorically inelegant. At the very least, we should know which input is invalid.
To do this, we'll use a parallel nested state machine.
Since a parallel nested machine is in every state at once, we give it one state for each input whose validation we care about.
Each of those states contains yet another machine (not parallel this time) with the state of the input: either valid or error
If this sounds complicated to you, that's because it is. State charts give an accurate representation of the complexity of a system. This is just exposing how complex forms actually are.
The good news is, once you're familiar with the pattern it's a copy paste job to extend it to new inputs.
// View in the XState Visualizer
// https://xstate.js.org/viz/?gist=28dfc6298fc233eb10a5ff8795570ce3
…
editing: {
on: {
SUBMIT: [
{
+ target: 'editing.message.error',
cond: 'isMessageTooShort'
},
{
target: 'submitting'
}
],
INPUT: {
target: 'editing',
actions: 'cache'
}
},
+ type: 'parallel',
+ states: {
+ message: {
+ initial: 'valid',
+ states: {
+ valid: {},
+ error: {}
+ }
+ }
+ }
},
…
Inputs with specific error states
A well designed form will tell you not only which field is invalid, but exactly what's wrong with it too. The trick to that is another state machine inside in the error state of an input
For demonstration purposes, we'll add a new guard isMessageEmpty
to show how we can target specific error states
// View in the XState Visualizer
// https://xstate.js.org/viz/?gist=28dfc6298fc233eb10a5ff8795570ce3
…
editing: {
on: {
SUBMIT: [
+ {
+ target: 'editing.message.error.empty',
+ cond: 'isMessageEmpty'
+ },
{
target: 'editing.message.error.tooShort',
cond: 'isMessageTooShort'
},
{
target: 'submitting'
}
],
INPUT: {
target: 'editing',
actions: 'cache'
}
},
type: 'parallel',
states: {
message: {
initial: 'valid',
states: {
valid: {},
error: {
+ initial: 'empty',
+ states: {
+ empty: {},
+ tooShort: {}
+ }
}
}
}
}
},
…
Forms with multiple inputs
Lets try a real-world scenario. Imagine a contact form on your website, where clients can pitch jobs.
We can imagine several inputs would be required
- Name - required
- Email - required, formatted like an email
- Website - not required
- Budget - required, must be a positive number at least $1000
- Deadline - required, must be future date at least three days away
- Message - required, must be at least 10 characters long
These are general business requirements, so there's lots of room for flexibility here.
We can copy and paste our message
state for each of these form fields, and add the additional error states we want to track.
states: {
+ name: {
+ initial: 'valid',
+ states: {
+ valid: {},
+ error: {
+ initial: 'empty',
+ states: {
+ empty: {}
+ }
+ }
+ }
+ },
+
+ email: {
+ initial: 'valid',
+ states: {
+ valid: {},
+ error: {
+ initial: 'empty',
+ states: {
+ empty: {},
+ badFormat: {}
+ }
+ }
+ }
+ },
+
+ budget: {
+ initial: 'valid',
+ states: {
+ valid: {},
+ error: {
+ initial: 'empty',
+ states: {
+ empty: {},
+ badFormat: {},
+ tooLow: {}
+ }
+ }
+ }
+ },
+
+ deadline: {
+ initial: 'valid',
+ states: {
+ valid: {},
+ error: {
+ initial: 'empty',
+ states: {
+ empty: {},
+ tooEarly: {}
+ }
+ }
+ }
+ },
message: {
initial: 'valid',
states: {
valid: {},
error: {
initial: 'empty',
states: {
empty: {},
tooShort: {}
}
}
}
}
}
It would not be wrong to forego the nested error type state when the only type is empty
, and instead have valid
and empty
as two sibling states. I've found keeping a consistent error pattern is easier to grok in my own head, but your mileage may vary.
We get away easy with the website input, because any input we don't need to validate doesn't need to be part of the state machine.
Next, lets implement the guards for these inputs
guards: {
isNameEmpty: context => context.name.length === 0,
isEmailEmpty: context => context.email.length === 0,
isEmailBadFormat: context => !context.email.includes('@'),
isBudgetEmpty: context => context.budget.length === 0,
isBudgetBadFormat: context => !context.budget.match(MONETARY_VALUE_REGEX),
isBudgetTooLow: context => parseFloat(context.budget) < 1000,
isDeadlineEmpty: context => !context.deadline,
isDeadlineTooEarly: context => context.deadline <= THREE_DAYS_FROM_NOW,
isMessageEmpty: context => context.message.length === 0,
isMessageTooShort: context => context.message.length < 10
}
You can decide for yourself how fine-grained the validation should be. If you want different language on the website for a budget that's a little under or way under, add a new state and a new guard to accommodate that.
There are real names in the world that are only a single character long, and it's not unrealistic to have special characters or even a number in a valid name, so don't go overboard on trying to catch bad formats there. That's one extra state to track and at least one segment of the population you'd be excluding.
Emails are impossible to accurately validate client-side, and especially not with regex. Maintaining that it has at least one @ is about the safest check you can do.
Once you're happy with your guards, we can put them together and add the conditional transitions on the SUBMIT event
SUBMIT: [
+ {
+ target: 'editing.name.error.empty',
+ cond: 'isNameEmpty'
+ },
+ {
+ target: 'editing.email.error.empty',
+ cond: 'isEmailEmpty'
+ },
+ {
+ target: 'editing.email.error.badFormat',
+ cond: 'isEmailBadFormat'
+ },
+ {
+ target: 'editing.budget.error.empty',
+ cond: 'isBudgetEmpty'
+ },
+ {
+ target: 'editing.budget.error.badFormat',
+ cond: 'isBudgetBadFormat'
+ },
+ {
+ target: 'editing.budget.error.tooLow',
+ cond: 'isBudgetTooLow'
+ },
+ {
+ target: 'editing.deadline.error.empty',
+ cond: 'isDeadlineEmpty'
+ },
+ {
+ target: 'editing.deadline.error.tooEarly',
+ cond: 'isDeadlineTooEarly'
+ },
{
target: 'editing.message.error.empty',
cond: 'isMessageEmpty'
},
{
target: 'editing.message.error.tooShort',
cond: 'isMessageTooShort'
},
{
target: 'submitting'
}
],
And that's it! We've now mapped out a robust contact form with field level validation, that can't end up in impossible states and allows for retry on failure.
Check out the completed chart on XState Visualizer
Top comments (0)