form-state

form-state is a headless form state management library, built on top of mobx.

It acts as a buffer between the canonical data (i.e. the server-side data, or your app's GraphQL/Redux/etc. global store) and the user's WIP data that is being actively mutated in form fields (which is "too chatty"/WIP to push back into global stores).

It also keeps track of low-level form UX details like:

• Which form fields are dirty
• Which form fields are valid/invalid
• Which form fields are touched (i.e. don't show validation errors for untouched fields)
• Enabling/disabling buttons/form UX based on the overall form-wide state
• Submitting the form should touch (validate) all fields
• Auto-saving the form when appropriate (i.e. not on keystroke, but after blur/leaving the field)
• Queueing auto-saves if one is already in-flight
  • Auto-saves in a table with per-row forms will serialize to avoid cross-child write conflicts on the backend
• Not over-writing the user's WIP/actively-focused field when auto-saved data refreshes
• Building a wire payload that has only changed fields
  • form.changedValue will return the entity id + only changed fields to faciliate doing partial update APIs
  • Supports collections of children, i.e. a author: { books: [...} } will include only changed books if necessary
  • Child collections can be either exhaustive (if any child changes, submit them all) or incremental (only include changed children), to match the backend endpoint's semantics

Main Features

There are two main reasons why form-state exists:

1. Provide a FieldState interface for component libraries
2. Match the "three shapes" mental model of our forms

FieldState Interface & "One-Line" Forms

The core abstraction that form-state provides is a FieldState interface that looks like:

// Very simplified example
interface FieldState {
  // Can be used to read & write the value bound into the form field
  value: V;
  errors: string[];
  valid: boolean;
  touched: boolean;
}

Which combines all the logical aspects of "a single form field" into a single object/prop.

This facilitates the "gold standard" of form DX, which is "one line per form field", i.e:

function AuthorEditorComponent() {
  const author = useFormState(() => /* ... */ );
  return (
    <FormLines>
      <BoundTextField field={author.firstName} />
      <BoundTextField field={author.lastName} />
      <BoundSelectField field={author.city} options={...cities...} />
    </FormLines>
  )
}

Besides great developer ergonomics (low boilerplate, very DRY code), this approach also provides a very consistent UI/UX for users, because all forms get the highly-polish behavior of BoundTextField for free.

(See the BoundTextField in Beam for an actual implementation of this approach.)

The Three Shapes Mental Model

In general when working with any forms (i.e. not just form-state), there are three types/shapes of data involved:

1. The input data/shape from the server (i.e. a GraphQL/REST query)
2. The form data/shape that is being reactively bound to form fields (i.e. used as <TextField value={form.firstName} onChange={(v) => form.firstName = v} />)
3. The mutation data/shape that will submit the change to the server (i.e. the GraphQL mutation/REST POST)

form-state generally refers to each of these shapes as:

• The input type
  • (Hrm, in retrospect "input" is an unfortunate term b/c that is what GraphQL uses for its mutation types, i.e. input SaveAuthorInput...we should consider changing this).
• The form type
• The ...third type...

And then provides an API/DSL for managing the mapping between each of these in a standard/conventional manner.

Admittedly (and hopefully, b/c it makes the code simpler), the differences between each of these types can often be small, i.e.:

• The input type might have { author: { book: { id: "b:1" } } but the mutation wants { author: { bookId: "b:1" } }
• ...have other examples...

These are usually simple/mechanistic changes, but nonetheless just some boilerplate that form-state provides conventions for.

Basic Usage

See the sample.

Todo

• Add conditional readonly logic, like { type: "field", readOnlyIf: i => i.isInternal.value }

• Add omitFromValue so we can have two fields, book.author.id / book.author.name, where book.author.name is used for showing the author name, but book.author.id is the only field that is submitted to the server on mutation (maybe pair this with Ref based mutations)

• Undo/redo would in theory be neat and easy to do on top of the existing infra

Features

Incremental changedValue/dirty Management

In general, we have two types of forms:

• autoSave forms
• submit forms

autoSave forms

For auto save forms, the expectation is that you should:

• Pass useFormState an init.input and init.map that updates the form-state from the initial GraphQL request & any cache updates

• Pass useFormState an autoSave lambda that calls your GraphQL mutation, using changedValue.

• Have your save mutation response return the acked/updated entity/fragment

  autoSave: async () => {
    const input = formState.changedValue;
    await saveAuthor(input);
  };

With these in place, we will correctly handle interleaved edits/saves, i.e.:

1. User changes { firstName: bob }
2. We submit { id: 1, firstName: bob } to the backend
3. While waiting for the response, the user sets { lastName: smith }
4. The GraphQL mutation acks that { id: 1, firstName: bob } is committed
5. The init.map updates formState to realize firstName is no longer dirty, but lastName keeps its WIP change
6. formState will trigger a 2nd autoSave for just the lastName change

Submit forms

For submit forms, the expectation is that you should:

• Pass useFormState an init.input and init.map that updates the form-state from the initial GraphQL request & any cache updates
• In your onClick lambda, use formState.changedValue to call your GraphQL mutation
• Have your save mutation response return the acked/updated entity/fragment

If you do this, you should not have to call commitChanges manually, because code like:

  onClick: async () => {
    const input = formState.changedValue;
    await saveAuthor(input);
    // checks if formState.dirty is true before closing
    closeModal();
  };

Will "just work" because the control flow will be:

• User changes { firstName: bob } and clicks Submit
• onClick runs and we submit { id: 1, firstName: bob } to saveAuthor
• Before the await promise resolves, the GraphQL response of saveAuthor { ...AuthorFragment } will:
  • Update the apollo cache
  • Re-render the AuthorEditor component with the new data
  • Call init.map to update formState with the new data
  • Realize the firstName is no longer dirty
• When closeModal runs, no "You have unsaved changes?" will appear

Basically, in a correctly-setup form, you should never have to call commitChanges manually, and doing so risks losing edits that the user made while any saves (either auto save or submit save) were in-flight.

Fragments

Normally, form-state expects all fields in the form to be inputs to the GraphQL mutation/wire call. For example, the author.firstName field will always be submitted to the saveAuthor mutation (albeit with author.changedValue you can have firstName conditionally included).

However, sometimes there is "other data" that your UX needs to render the form, which is not strictly a form field, but would be handy for the data to "just be on the form" anyway, as you're passing it in around code.

A stereotypical example of this is GraphQL fragments, where an AuthorFragment might have a lot of misc read-only info that you want to display next to/within your form, but is not technically editable.

In form-state, you can model with as a Fragment, which is set up as:

// Your input type, likely generated from GraphQL mutation
type AuthorInput = { firstName?: string };

// Your wire data likely from your page's GraphQL query to get
// the author to edit + also "misc other data"
type AuthorFragment = { firstName: string; miscOtherData: {} };

// For your page's form state, add-in the "extra data"
type AuthorForm = AuthorInput & {
  // The `Fragment` type tells form-state this is not a regular form field
  data: Fragment<AuthorFragment>;
};

// Tell the form config the "fragment" is not a real field
const config: ObjectConfig<AuthorForm> = {
  firstName: { type: "value", rules: [require] },
  data: { type: "fragment" },
};

// Now in the component...
const data = useGraphQLQuery();
const form = useFormState({
  config,
  init: {
    input: data,
    map: (d) => ({
      firstName: data.author.firstName,
      data: fragment(data),
    }),
  },
});

Internal Implementation Notes

form-state keeps the "actual data" (basically a POJO of your form data) separate from the "mobx proxies that track reactivity" (the ObjectState interface with .get / .set / .errors other methods).

This works well b/c the "actual data" returned from ObjectState.value or FieldState.value is always a non-proxy POJO that can be dropped on the wire without causing serialization issues.

However, it does mean that form-state internally uses a few "that looks odd" tricks like _tick.value++ to ensure code like formState.value.firstName will be reactive, even though the .firstName is not actually a proxy access (but doing formState.firstName.value would be).

(To be clear, both formState.firstName.value and formState.value.firstName return the same value, and also have the same reactivity semantics, this is just noting that form-state's internals need to do a few extra tricks to get the latter to be reactive.)