Migrating from v1

v2 is a clean break in the JavaScript API and a deliberate freeze of the rendered DOM. Understanding that split makes the upgrade far smaller than it first looks.

Your CSS and your DOM selectors keep working

Read this before estimating the work

Class names, element structure, #dl-form-field-wrapper-for-<name> ids, tab class derivation, .dl-tab-btn sibling ordering, dl-select[name=…] selectors — all unchanged, including misspelled classes like .dl-muliselect-selected-remove. Stylesheets need no changes. End-to-end suites that select on rendered output need no changes. See the DOM contract.

What changes is how you write the form.

Package and imports

- import DeclarativForm from 'declarativ_forms';
+ import { DeclarativeForm } from 'declarative-forms';

The package is renamed (declarative-forms), the export is now named rather than default, and the class name gains its missing e.

Stylesheet:

- @import "declarativ_forms/assets/default.css";
+ @import "declarative-forms/classic.css";

declarative-forms/assets/default.css is still a valid export path mapped to the same file, so that import can stay if you prefer a one-word diff.

styles.css is not the v1 look

The v1 stylesheet is classic.css. declarative-forms/styles.css is the package's own default — a modern light/dark theme that renders the identical DOM but looks different. Import it deliberately, not as part of the rename.

Constructor

Positional arguments become a single options object:

- new DeclarativForm(attrs, onChange, onCancel, confirmCaption, parentForm)
+ new DeclarativeForm({ fields, buttons, onConfirm, onCancel, confirmLabel })

// v1
new DeclarativForm({ fields }, save, () => {}, 'Back to References');

// v2
new DeclarativeForm({
  fields,
  onConfirm: save,
  onCancel: () => {},
  confirmLabel: 'Back to References',
});

parentForm is gone — nesting is handled internally, and child forms read outward via ctx.parentData / ctx.stackData.

Field descriptors gain kind

v1 inferred the field type from which optional key happened to be present. v2 uses an explicit discriminant, which is what makes the descriptors typable.

v1	v2
(no marker)	kind: 'text' (or omit — it is the default)
largetext: true	kind: 'textarea'
allowedValues	kind: 'select', key renamed to options
check: 'caption'	kind: 'checkbox', label: 'caption'
message	kind: 'message'
inputType: 'file'	kind: 'file'
acceptFileType	accept
calculate	kind: 'computed', key renamed to compute
detailedOptions	kind: 'cards', key renamed to cards
render	kind: 'custom'
arrayOf	kind: 'array', key renamed to of
inputType: 'password'	type: 'password'

- { name: 'note', largetext: true, allowNewlines: true }
+ { name: 'note', kind: 'textarea', allowNewlines: true }

- { name: 'lang', allowedValues: [['en', 'English']] }
+ { name: 'lang', kind: 'select', options: [{ value: 'en', label: 'English' }] }

- { name: 'agree', check: 'I agree' }
+ { name: 'agree', kind: 'checkbox', label: 'I agree' }

Select options also move from tuples to objects:

- allowedValues: [['bk', 'Book', 'Type: Book']]
+ options: [{ value: 'bk', label: 'Book', displayWhenSelected: 'Type: Book' }]

Plain string arrays still work unchanged.

Renamed keys and methods

v1	v2
mapFiledsOnEdit (typo)	mapFieldsOnEdit
onValuesCalculationFailedMessage	onOptionsError
newButtonLabel	unchanged
descriptor.domElement	form.field(name)
dom.onChange(force)	ctx.requestUpdate(force)
field.editArrayOfEntryMode	ctx.isEditingArrayEntry
closeModalIfOpen(cb?)	close(cb?)
updateCalculatedFields()	updateComputedFields()
updateForm()	update()
subscribeOnInput(cb)	unchanged, but now returns an unsubscribe function
mulitple-allowed attribute (typo)	multiple

Callbacks take one context object

The single largest ergonomic change. v1 gave each callback a different positional argument list; v2 gives them all the same object.

- isActive: (formData, stackData, field, parentData) => …
+ isActive: ({ data, stackData, field, parentData }) => …

- tab: (field) => …
+ tab: ({ field, data }) => …

- defaultValue: (formData) => …
+ defaultValue: ({ data }) => …

- placeholder: (formData) => …
+ placeholder: ({ data }) => …

- allowedValues: (formData) => …
+ options: ({ data }) => …

- onFormChange: (formData, form, triggerElement) => …
+ onFormChange: ({ data, form, trigger }) => …

The context is { data, form, field, parentData, stackData, isEditingArrayEntry }, plus trigger on onFormChange and element / requestUpdate / setValue on a custom field's render.

Imperative field access

- fields.find((f) => f.name === 'lang').domElement.setValue('de');
+ form.field('lang')?.setValue('de');

- extRefField.domElement.setLoadingStatus();
+ form.field('extRef')?.setLoading(true);

- document.querySelector('dl-select[name=existing-ref]').setValue(id);
+ form.field('existing-ref')?.setValue(id);

v2 never writes back onto your descriptor objects, so they can be shared, frozen or generated. form.field(name) returns a typed handle with getValue, setValue, focus, setLoading, element and wrapper.

Behaviour changes

Five differences that are not just renames.

1. The single-button quirk is gone

v1 silently overwrote onChangeCallback when buttons had exactly one key. Code commonly passed () => {} as the callback and put the real handler in the sole button, relying on the overwrite.

In v2 onConfirm and buttons are independent: declare buttons and onConfirm is not used.

- new DeclarativForm({ fields, buttons: { Save: { action: save } } }, () => {}, cancel);
+ new DeclarativeForm({ fields, buttons: { Save: { action: save } }, onCancel: cancel });

2. message fields leave getValues()

v1 emitted '' for them. v2 omits them entirely. Code that iterates over all value keys may need adjusting.

3. Buttons are <button>, not <div>

They keep the .btn class, and the stylesheet normalises the UA button styles, so rendering is unchanged — but they are now focusable and keyboard-operable.

4. openInModal() still returns an element

Unchanged, and deliberately so. It returns the raw div.dl-modal. If your code calls .close() on that return value, it was a silent no-op in v1 and still is — the method does not exist. Replace it with form.close().

5. Falsy values survive

v1 read values as getAttribute('data-value') || getAttribute('value') || _value || value, so '', 0 and false fell through to the next source. v2 returns them faithfully — notably, an unchecked checkbox is now the boolean false rather than an accidental string.

Other fixes you inherit

No action needed, but worth knowing the behaviour changed for the better:

• Stale async options are discarded. A slow options load from an earlier keystroke can no longer overwrite newer results.
• Two forms no longer fight. Tabs and tooltips were resolved with document.querySelector in v1; they are now scoped per form.
• Text is escaped by default. Labels, messages and option content go through textContent unless wrapped in html() — if you passed markup in any of these, wrap it.
• No import side effects. The keydown listener and stylesheet are installed lazily, so importing the library is SSR-safe.
• The confirm button no longer deletes itself when showing its loading state (v1 called Element.remove() where classList.remove() was meant).

Suggested order

Because dependencies are pinned per repository, migrate one at a time:

1. presspack — only user of suggested and acceptFileType. It receives the constructor as a function argument across a package boundary, which can now be typed. Its option labels contain raw HTML (<span part="tag" class="dl-option-tag">) and must be wrapped in html().
2. structured-text-editor — mostly reference_controller.js. Sole user of mapFiledsOnEdit, onValuesCalculationFailedMessage and editArrayOfEntryMode. Convert domElement access to form.field(). Its ~193 dl-select[name=…] end-to-end selectors need no change.
3. monsterwriter2 — largest surface: the React wrapper, appendInElement, subscribeOnInput, updateComputedFields, setActiveTab(undefined) as a resync, and closeModalIfOpen() called with no argument behind a @ts-ignore (v2's optional-callback close() removes the need). Its CSS overrides and dark-mode --dl-* declarations need no change. Then bump the monsterwriter-electron submodule.

Staying on v1

v1 remains available at its existing commits. If you pin by git SHA, nothing changes until you choose to move.