@ditojs/admin 2.100.0 → 3.0.0

package/src/utils/FormModel.js

@@ -0,0 +1,556 @@
+import { effectScope, computed, watch, nextTick, shallowRef } from 'vue'
+import {
+  isArray,
+  isPlainObject,
+  equals,
+  clone,
+  parseDataPath,
+  getValueAtDataPath
+} from '@ditojs/utils'
+import DitoContext from '../DitoContext.js'
+import { DataSchemaResolver } from './DataSchemaResolver.js'
+import { isNested, isSourceWithResource } from './schema/structure.js'
+import { isEmptySchema } from './schema/lookup.js'
+import {
+  shouldRenderSchema,
+  initializeData,
+  computeValue,
+  hasValueFromDataSchema,
+  processSchemaData,
+  processData
+} from './schema/data.js'
+
+// FormModel holds the state of the data edited by a form, view or dialog that
+// is derived from schema and data, independently of what is rendered:
+//
+// - Missing values are set to their defaults when the data is set up, except
+//   the values of components with `compute()`, which fall back to their
+//   defaults when `compute()` doesn't return a value, unlike in new data, see
+//   `initializeData()`.
+// - The items of lists with `orderKey` are numbered when the data is set up,
+//   which doesn't make the data dirty.
+// - Computed values, the results of `schema.compute()` and of the data schemas
+//   of the `computed` types (`schema.data`, `schema.dataPath`), are written
+//   into the data by watchers that the model owns, one scope per component,
+//   for all components whose `if` doesn't evaluate to `false`.
+// - Options, `schema.options`, are resolved per component when they are first
+//   read, by `compute()` through `context.options` or by the component that
+//   displays them, see `getOptions()`. Both get the same option objects, and
+//   computes that read options run again when they are loaded.
+// - Submitting waits for the loads of data schemas and options that are still
+//   pending, see `waitForPendingLoads()`.
+// - The data is dirty when its processed data differs from a snapshot taken
+//   when the data was set up, saved or applied, see `isDirty`. The values that
+//   the model derives until it settled, e.g. from loaded options, don't count,
+//   see `takeProcessedDataSnapshot()`. Only forms track this, by passing
+//   `getSourceSchema()`, see the constructor.
+//
+// Sources with their own resource are skipped, as their items are edited
+// through their own forms.
+//
+// The model handles components through their entries, the objects that
+// `processSchemaData()` passes to `before()` and `after()`, with the schema,
+// data, name, data path and component path of each component. Components that
+// read their options pass entries of the same shape, see `getOptions()`.
+//
+// `component` is the component that owns the data, e.g. `DitoForm`. Its
+// `dataPath`, `componentPath`, `rootData` and `mainSchemaComponent` are used
+// when present. The model needs to be stopped before the component unmounts.
+
+export class FormModel {
+  // The entries of the components with computed values and the scopes of the
+  // watchers that write these values into the data, by component path:
+  computedValueRecords = new Map()
+  // The entries of the components whose options were read and the resolvers
+  // of their options, by component path, see `getOptionsResolver()`:
+  optionsRecords = new Map()
+  // The promises of the loads of data schemas and options that are pending,
+  // see `waitForPendingLoads()`:
+  pendingLoads = new Set()
+  // The data that the snapshot was taken of, and its processed data to compare
+  // with in `isDirty`, see `takeProcessedDataSnapshot()`:
+  processedDataSnapshot = shallowRef(null)
+  // The data paths of the values that the model writes while it settles after
+  // the snapshot is taken, which `isDirty` takes from the current data, `null`
+  // once it settled, see `takeProcessedDataSnapshot()`:
+  derivedValueDataPaths = null
+
+  // `getSchema()` and `getData()` return the schema and the data that the
+  // model handles. `getSourceSchema()` returns the schema of the source that
+  // the data is an item of, as needed by `processData()`. Only forms pass it,
+  // to track whether their data is dirty, see `isDirty`.
+  constructor({ component, getSchema, getData, getSourceSchema = null }) {
+    this.component = component
+    this.getSchema = getSchema
+    this.getData = getData
+    this.getSourceSchema = getSourceSchema
+    this.rootScope = effectScope(true)
+    this.rootScope.run(() => {
+      watch(
+        [getSchema, getData],
+        ([schema, data]) => this.initializeData(schema, data),
+        { immediate: true }
+      )
+      // The entries are read through a computed property, so that the watchers
+      // of the entries can check synchronously whether their entry is still
+      // current, see `createComputedValueScope()`.
+      this.computedValueEntries = computed(() => this.getComputedValueEntries())
+      watch(
+        () => this.computedValueEntries.value,
+        entries => this.updateComputedValueRecords(entries),
+        { immediate: true }
+      )
+      if (getSourceSchema) {
+        // Data that is set up, e.g. loaded, saved or applied, isn't dirty:
+        watch(
+          getData,
+          () => this.takeProcessedDataSnapshot().catch(console.error),
+          { immediate: true }
+        )
+      }
+    })
+  }
+
+  get dataPath() {
+    return this.component.dataPath ?? ''
+  }
+
+  get componentPath() {
+    return this.component.componentPath ?? ''
+  }
+
+  get rootData() {
+    return this.component.rootData ?? this.getData()
+  }
+
+  // Returns whether the processed data differs from the snapshot taken when
+  // the data was set up, saved or applied, see `takeProcessedDataSnapshot()`.
+  // Excluded values don't count, unless their components have `process()`,
+  // see `processData()`, and neither do the values that the model derives
+  // until it settled.
+  get isDirty() {
+    const processedDataSnapshot = this.processedDataSnapshot.value
+    const data = this.getData()
+    // A snapshot only applies to the data that it was taken of. Data that was
+    // replaced since isn't dirty until its own snapshot is taken, and neither
+    // is missing data.
+    if (!data || processedDataSnapshot?.data !== data) {
+      return false
+    }
+    const processedData = this.getProcessedDataForDirtyCheck()
+    const processedDataToCompare = this.derivedValueDataPaths
+      ? takeOverValuesAtDataPaths(
+          processedDataSnapshot.processedData,
+          processedData,
+          this.derivedValueDataPaths
+        )
+      : processedDataSnapshot.processedData
+    return !equals(processedData, processedDataToCompare)
+  }
+
+  stop() {
+    this.rootScope.stop()
+    this.computedValueRecords.clear()
+    this.optionsRecords.clear()
+    this.pendingLoads.clear()
+  }
+
+  // Waits for the pending loads of data schemas and options, including the
+  // ones that they cause, e.g. options that depend on computed values that
+  // depend on loaded options, and for the watchers to write the resulting
+  // computed values into the data.
+  async waitForPendingLoads() {
+    while (this.pendingLoads.size > 0) {
+      await Promise.all(this.pendingLoads)
+      await nextTick()
+    }
+  }
+
+  // Takes the snapshot of the processed data that `isDirty` compares with,
+  // when the data is set up, saved or applied. Until the model settled, i.e.
+  // the pending loads finished and the computed values that depend on them are
+  // written into the data, see `waitForPendingLoads()`, the values that the
+  // model writes are recorded in `derivedValueDataPaths` and don't count, as
+  // they're derived from the data, e.g. from loaded options. Once settled, they
+  // are taken over into the snapshot, so that changes made in the meantime,
+  // e.g. by the user, count right away. Values that the model derives from
+  // these changes while it settles are taken over too, so reverting the
+  // changes afterwards leaves the data dirty, which only asks for confirmation
+  // unnecessarily.
+  async takeProcessedDataSnapshot() {
+    const data = this.getData()
+    const derivedValueDataPaths = data ? new Set() : null
+    this.derivedValueDataPaths = derivedValueDataPaths
+    this.processedDataSnapshot.value = data
+      ? { data, processedData: clone(this.getProcessedDataForDirtyCheck()) }
+      : null
+    if (!data) {
+      return
+    }
+    // Let the watchers and the rendering start their loads first:
+    await nextTick()
+    await this.waitForPendingLoads()
+    // Snapshots that were taken since settle by themselves.
+    const isSnapshotReplaced = (
+      this.derivedValueDataPaths !==
+      derivedValueDataPaths
+    )
+    // Check whether the model stopped first, as its component may be gone.
+    const isModelStopped = !this.rootScope.active
+    if (isSnapshotReplaced || isModelStopped) {
+      return
+    }
+    this.derivedValueDataPaths = null
+    const processedDataSnapshot = this.processedDataSnapshot.value
+    if (processedDataSnapshot?.data === this.getData()) {
+      this.processedDataSnapshot.value = {
+        data,
+        processedData: takeOverValuesAtDataPaths(
+          processedDataSnapshot.processedData,
+          this.getProcessedDataForDirtyCheck(),
+          derivedValueDataPaths
+        )
+      }
+    }
+  }
+
+  // Returns the processed data that `isDirty` compares with
+  // `processedDataSnapshot`, and that the snapshot is taken of. The data is
+  // processed like for the clipboard, not for the server, which gives
+  // references to new items random prefixes each time, see
+  // `SchemaGraph.getReferencePrefix()`.
+  getProcessedDataForDirtyCheck() {
+    const schema = this.getSchema()
+    const data = this.getData()
+    return data && !isEmptySchema(schema)
+      ? processData(schema, this.getSourceSchema(), data, this.dataPath, {
+          // Like `getComputedValue()`:
+          component: this.component.mainSchemaComponent ?? this.component,
+          rootData: this.rootData,
+          schemaOnly: true,
+          target: 'clipboard',
+          // Reading the dirty state doesn't call `compute()` and `process()`,
+          // see `processData()`:
+          shouldCallComputeAndProcess: false
+        })
+      : null
+  }
+
+  // Makes clean changes to the data, which don't make it dirty, while other
+  // changes still do, e.g. to apply what an action on the server already
+  // saved: Once the model settled, the values that changed in the processed
+  // data, including the values derived from the changes, are taken over into
+  // the snapshot. Arrays whose length changes are taken over as a whole, see
+  // `takeOverChangedValues()`. `makeChanges()` is called synchronously.
+  async applyCleanChanges(makeChanges) {
+    const data = this.getData()
+    const hasSnapshotOfData = this.processedDataSnapshot.value?.data === data
+    // Clone the processed data, as it shares nested values with the data,
+    // which `makeChanges()` may change in place.
+    const processedDataBeforeChanges = hasSnapshotOfData
+      ? clone(this.getProcessedDataForDirtyCheck())
+      : null
+    makeChanges()
+    if (hasSnapshotOfData) {
+      // Let the watchers write the values derived from the changes first:
+      await nextTick()
+      await this.waitForPendingLoads()
+      // Take the changes over into the current snapshot of the data, which may
+      // have been replaced in the meantime, e.g. once the model settled.
+      const processedDataSnapshot = this.processedDataSnapshot.value
+      if (processedDataSnapshot?.data === data && this.rootScope.active) {
+        this.processedDataSnapshot.value = {
+          data,
+          processedData: takeOverChangedValues(
+            processedDataSnapshot.processedData,
+            processedDataBeforeChanges,
+            this.getProcessedDataForDirtyCheck()
+          )
+        }
+      }
+    }
+  }
+
+  // Tracks the promise of a pending load until it settles, so that
+  // `waitForPendingLoads()` waits for it.
+  trackPendingLoad(promise) {
+    this.pendingLoads.add(promise)
+    promise.finally(() => this.pendingLoads.delete(promise))
+  }
+
+  // Sets missing values to their defaults and numbers the items of lists by
+  // their order key when the data is set up, see `initializeData()`.
+  initializeData(schema, data) {
+    if (data && !isEmptySchema(schema)) {
+      initializeData(schema, data, this.component, {
+        dataPath: this.dataPath,
+        rootData: this.rootData,
+        shouldProcess: entry => !isSourceWithResource(entry.schema),
+        // The model writes the values of components with `compute()`,
+        // including their defaults:
+        shouldSetDefaultsOfComponentsWithCompute: false
+      })
+    }
+  }
+
+  // Returns the entries of `processSchemaData()` for all components with
+  // computed values whose `if` doesn't evaluate to `false`, by component path.
+  // Called by a computed property, so that it runs again when the data
+  // structure changes, e.g. when list items are added or removed, or when `if`
+  // conditions change.
+  getComputedValueEntries() {
+    const schema = this.getSchema()
+    const data = this.getData()
+    const entries = new Map()
+    if (data && !isEmptySchema(schema)) {
+      processSchemaData(schema, data, {
+        dataPath: this.dataPath,
+        componentPath: this.componentPath,
+        shouldProcess: entry => (
+          !isSourceWithResource(entry.schema) &&
+          shouldRenderSchema(entry.schema, this.createEntryContext(entry))
+        ),
+        before: entry => {
+          if (hasComputedValueSource(entry.schema)) {
+            entries.set(entry.componentPath, entry)
+          }
+        },
+        options: { component: this.component, rootData: this.rootData }
+      })
+    }
+    return entries
+  }
+
+  // Keeps the records of the components that are still present with the same
+  // schema and data, creates records with new scopes for new ones, and stops
+  // the scopes of the others.
+  updateComputedValueRecords(entries) {
+    const previousRecords = this.computedValueRecords
+    this.computedValueRecords = new Map()
+    for (const [componentPath, entry] of entries) {
+      let computedValueRecord = previousRecords.get(componentPath)
+      if (
+        computedValueRecord?.entry.schema === entry.schema &&
+        computedValueRecord.entry.data === entry.data
+      ) {
+        previousRecords.delete(componentPath)
+      } else {
+        computedValueRecord = {
+          entry,
+          scope: this.createComputedValueScope(entry)
+        }
+      }
+      this.computedValueRecords.set(componentPath, computedValueRecord)
+    }
+    for (const { scope } of previousRecords.values()) {
+      scope.stop()
+    }
+  }
+
+  // Creates the scope of the watchers that write the computed values of the
+  // entry into the data. Entries stop being current before their scope is
+  // stopped, as watchers that don't belong to components run in the order in
+  // which they're triggered, e.g. when list items are removed or `if`
+  // conditions change. The watchers skip these entries, see `isEntryCurrent()`.
+  createComputedValueScope(entry) {
+    const scope = this.rootScope.run(() => effectScope())
+    scope.run(() => {
+      const isEntryCurrent = computed(() => this.isEntryCurrent(entry))
+      if (entry.schema.compute) {
+        watch(
+          // Return a new object each time, so that the value is also written
+          // when only the value in the data changed, e.g. through user input.
+          () =>
+            isEntryCurrent.value
+              ? { value: this.getComputedValue(entry) }
+              : null,
+          computedResult => {
+            if (computedResult) {
+              this.writeComputedValue(entry, computedResult.value)
+            }
+          },
+          { immediate: true }
+        )
+      }
+      if (hasValueFromDataSchema(entry.schema)) {
+        const dataSchemaResolver = this.createDataSchemaResolver(
+          entry.schema,
+          entry
+        )
+        watch(
+          () =>
+            isEntryCurrent.value && !dataSchemaResolver.isLoading
+              ? { value: dataSchemaResolver.value }
+              : null,
+          resolved => {
+            if (resolved) {
+              this.writeComputedValue(entry, resolved.value)
+            }
+          },
+          { immediate: true }
+        )
+      }
+    })
+    return scope
+  }
+
+  // Returns whether the entry is still one of the current entries, with the
+  // same schema and data, see `getComputedValueEntries()`.
+  isEntryCurrent({ componentPath, schema, data }) {
+    const currentEntry = this.computedValueEntries.value.get(componentPath)
+    return currentEntry?.schema === schema && currentEntry.data === data
+  }
+
+  // Calls `schema.compute()` with the main schema component of the data, like
+  // `processData()`, and with the options resolved by the model, so that the
+  // result doesn't depend on whether the component is rendered.
+  getComputedValue(entry) {
+    const { schema, data, name, dataPath } = entry
+    return computeValue(schema, data, name, dataPath, {
+      component: this.component.mainSchemaComponent ?? this.component,
+      rootData: this.rootData,
+      getOptions: schema.options ? () => this.getOptions(entry) : null
+    })
+  }
+
+  // Returns the options of the component of the entry, resolved from
+  // `schema.options` when they're first read, `undefined` while loading.
+  getOptions(entry) {
+    return this.getOptionsResolver(entry).value
+  }
+
+  // Returns the resolver of the options of the component of the entry, shared
+  // by all callers with the same component path, schema and data, so that
+  // computes and the component that displays the options get the same option
+  // objects, and the options only load once.
+  getOptionsResolver(entry) {
+    const { schema, data, componentPath } = entry
+    let optionsRecord = this.optionsRecords.get(componentPath)
+    if (
+      !optionsRecord ||
+      optionsRecord.entry.schema.options !== schema.options ||
+      optionsRecord.entry.data !== data
+    ) {
+      optionsRecord = {
+        entry,
+        resolver: this.createDataSchemaResolver(schema.options, entry)
+      }
+      this.optionsRecords.set(componentPath, optionsRecord)
+    }
+    return optionsRecord.resolver
+  }
+
+  createDataSchemaResolver(dataSchema, entry) {
+    return new DataSchemaResolver(dataSchema, {
+      createContext: () => this.createEntryContext(entry),
+      onLoadStart: promise => this.trackPendingLoad(promise)
+    })
+  }
+
+  writeComputedValue(entry, value) {
+    const { data, name } = entry
+    if (!equals(value, data[name])) {
+      // Values written while the model settles are derived, see
+      // `takeProcessedDataSnapshot()`:
+      this.derivedValueDataPaths?.add(this.getRelativeDataPath(entry))
+      // Access `data[name]` directly to update the value without calling
+      // `parse()`, see `ValueMixin`:
+      data[name] = value
+    }
+  }
+
+  // Returns the data path of the value of the entry, relative to the model's
+  // data, as in its processed data.
+  getRelativeDataPath({ schema, name, dataPath }) {
+    // The data paths of nested components include their own name.
+    const tokens = isNested(schema)
+      ? parseDataPath(dataPath)
+      : [...parseDataPath(dataPath), name]
+    return tokens.slice(parseDataPath(this.dataPath).length)
+  }
+
+  // Returns the context for `if`, data schemas and options, with the component
+  // that owns the data, which stays the same when components are mounted, so
+  // that data schemas and options don't load again.
+  createEntryContext({ schema, data, name, dataPath }) {
+    return DitoContext.createForSchema(this.component, {
+      schema,
+      name,
+      data,
+      dataPath,
+      rootData: this.rootData
+    })
+  }
+}
+
+// Returns a copy of `processedData` with the values at `dataPaths` taken over
+// from `currentProcessedData`, or removed if they're missing there. Values
+// whose parents are missing in `processedData` are left out, as their parents
+// differ anyway.
+function takeOverValuesAtDataPaths(
+  processedData,
+  currentProcessedData,
+  dataPaths
+) {
+  if (dataPaths.size === 0) {
+    return processedData
+  }
+  const result = clone(processedData)
+  for (const dataPath of dataPaths) {
+    const parentDataPath = dataPath.slice(0, -1)
+    const key = dataPath.at(-1)
+    const parent = getValueAtDataPath(result, parentDataPath, () => null)
+    if (parent && typeof parent === 'object') {
+      const value = getValueAtDataPath(
+        currentProcessedData,
+        dataPath,
+        () => notFound
+      )
+      if (value === notFound) {
+        delete parent[key]
+      } else {
+        parent[key] = clone(value)
+      }
+    }
+  }
+  return result
+}
+
+// Returns a copy of `target` with the values that differ between `before` and
+// `after` taken over from `after`, comparing objects and arrays of the same
+// length entry by entry. Arrays whose length changed are taken over as a whole,
+// as their entries can't be matched, including changes of their entries that
+// were made before.
+function takeOverChangedValues(target, before, after) {
+  if (equals(before, after)) {
+    return target
+  }
+  const isSameShape = (
+    isPlainObject(before) && isPlainObject(after) && isPlainObject(target) ||
+    isArray(before) &&
+    isArray(after) &&
+    isArray(target) &&
+    before.length === after.length &&
+    after.length === target.length
+  )
+  if (!isSameShape) {
+    return clone(after)
+  }
+  const result = isArray(target) ? [...target] : { ...target }
+  for (const key of new Set([...Object.keys(before), ...Object.keys(after)])) {
+    if (!(key in after)) {
+      delete result[key]
+    } else {
+      result[key] = takeOverChangedValues(target[key], before[key], after[key])
+    }
+  }
+  return result
+}
+
+const notFound = Symbol('notFound')
+
+// Returns whether the schema is a source of computed values, through
+// `schema.compute()` or a data schema of the `computed` types.
+function hasComputedValueSource(schema) {
+  return !!schema.compute || hasValueFromDataSchema(schema)
+}