Have PrismDerivation no longer extend AbstractDerivation

packages/dataverse/src/derivations/prism/prism.ts

@@ -1,25 +1,267 @@
 import Box from '../../Box'
+import type Ticker from '../../Ticker'
 import type {$IntentionalAny, VoidFn} from '../../types'
 import Stack from '../../utils/Stack'
-import AbstractDerivation from '../AbstractDerivation'
+import type Tappable from '../../utils/Tappable'
+import DerivationEmitter from '../DerivationEmitter'
+import DerivationValuelessEmitter from '../DerivationValuelessEmitter'
 import type {IDerivation} from '../IDerivation'
+import {isDerivation} from '../IDerivation'
 import {
   startIgnoringDependencies,
   stopIgnoringDependencies,
   pushCollector,
   popCollector,
+  reportResolutionStart,
+  reportResolutionEnd,
 } from './discoveryMechanism'
 
+type IDependent = (msgComingFrom: IDerivation<$IntentionalAny>) => void
+
 const voidFn = () => {}
 
-export class PrismDerivation<V> extends AbstractDerivation<V> {
+export class PrismDerivation<V> implements IDerivation<V> {
   protected _cacheOfDendencyValues: Map<IDerivation<unknown>, unknown> =
     new Map()
   protected _possiblyStaleDeps = new Set<IDerivation<unknown>>()
   private _prismScope = new PrismScope()
 
-  constructor(readonly _fn: () => V) {
-    super()
+  /**
+   * Whether the object is a derivation.
+   */
+  readonly isDerivation: true = true
+  private _didMarkDependentsAsStale: boolean = false
+  private _isHot: boolean = false
+
+  private _isFresh: boolean = false
+
+  /**
+   * @internal
+   */
+  protected _lastValue: undefined | V = undefined
+
+  /**
+   * @internal
+   */
+  protected _dependents: Set<IDependent> = new Set()
+
+  /**
+   * @internal
+   */
+  protected _dependencies: Set<IDerivation<$IntentionalAny>> = new Set()
+
+  constructor(readonly _fn: () => V) {}
+
+  /**
+   * Whether the derivation is hot.
+   */
+  get isHot(): boolean {
+    return this._isHot
+  }
+
+  /**
+   * @internal
+   */
+  protected _addDependency(d: IDerivation<$IntentionalAny>) {
+    if (this._dependencies.has(d)) return
+    this._dependencies.add(d)
+    if (this._isHot) d.addDependent(this._markAsStale)
+  }
+
+  /**
+   * @internal
+   */
+  protected _removeDependency(d: IDerivation<$IntentionalAny>) {
+    if (!this._dependencies.has(d)) return
+    this._dependencies.delete(d)
+    if (this._isHot) d.removeDependent(this._markAsStale)
+  }
+
+  /**
+   * Returns a `Tappable` of the changes of this derivation.
+   */
+  changes(ticker: Ticker): Tappable<V> {
+    return new DerivationEmitter(this, ticker).tappable()
+  }
+
+  /**
+   * Like {@link AbstractDerivation.changes} but with a different performance model. `changesWithoutValues` returns a `Tappable` that
+   * updates every time the derivation is updated, even if the value didn't change, and the callback is called without
+   * the value. The advantage of this is that you have control over when the derivation is freshened, it won't
+   * automatically be kept fresh.
+   */
+  changesWithoutValues(): Tappable<void> {
+    return new DerivationValuelessEmitter(this).tappable()
+  }
+
+  /**
+   * Keep the derivation hot, even if there are no tappers (subscribers).
+   */
+  keepHot() {
+    return this.changesWithoutValues().tap(() => {})
+  }
+
+  /**
+   * Convenience method that taps (subscribes to) the derivation using `this.changes(ticker).tap(fn)` and immediately calls
+   * the callback with the current value.
+   *
+   * @param ticker - The ticker to use for batching.
+   * @param fn - The callback to call on update.
+   *
+   * @see changes
+   */
+  tapImmediate(ticker: Ticker, fn: (cb: V) => void): VoidFn {
+    const untap = this.changes(ticker).tap(fn)
+    fn(this.getValue())
+    return untap
+  }
+
+  /**
+   * Add a derivation as a dependent of this derivation.
+   *
+   * @param d - The derivation to be made a dependent of this derivation.
+   *
+   * @see removeDependent
+   */
+  // TODO: document this better, what are dependents?
+  addDependent(d: IDependent) {
+    const hadDepsBefore = this._dependents.size > 0
+    this._dependents.add(d)
+    const hasDepsNow = this._dependents.size > 0
+    if (hadDepsBefore !== hasDepsNow) {
+      this._reactToNumberOfDependentsChange()
+    }
+  }
+
+  /**
+   * Remove a derivation as a dependent of this derivation.
+   *
+   * @param d - The derivation to be removed from as a dependent of this derivation.
+   *
+   * @see addDependent
+   */
+  removeDependent(d: IDependent) {
+    const hadDepsBefore = this._dependents.size > 0
+    this._dependents.delete(d)
+    const hasDepsNow = this._dependents.size > 0
+    if (hadDepsBefore !== hasDepsNow) {
+      this._reactToNumberOfDependentsChange()
+    }
+  }
+
+  protected _markAsStale = (which: IDerivation<$IntentionalAny>) => {
+    this._reactToDependencyBecomingStale(which)
+
+    if (this._didMarkDependentsAsStale) return
+
+    this._didMarkDependentsAsStale = true
+    this._isFresh = false
+
+    for (const dependent of this._dependents) {
+      dependent(this)
+    }
+  }
+
+  /**
+   * Gets the current value of the derivation. If the value is stale, it causes the derivation to freshen.
+   */
+  getValue(): V {
+    /**
+     * TODO We should prevent (or warn about) a common mistake users make, which is reading the value of
+     * a derivation in the body of a react component (e.g. `der.getValue()` (often via `val()`) instead of `useVal()`
+     * or `uesPrism()`).
+     *
+     * Although that's the most common example of this mistake, you can also find it outside of react components.
+     * Basically the user runs `der.getValue()` assuming the read is detected by a wrapping prism when it's not.
+     *
+     * Sometiems the derivation isn't even hot when the user assumes it is.
+     *
+     * We can fix this type of mistake by:
+     * 1. Warning the user when they call `getValue()` on a cold derivation.
+     * 2. Warning the user about calling `getValue()` on a hot-but-stale derivation
+     *    if `getValue()` isn't called by a known mechanism like a `DerivationEmitter`.
+     *
+     * Design constraints:
+     * - This fix should not have a perf-penalty in production. Perhaps use a global flag + `process.env.NODE_ENV !== 'production'`
+     *   to enable it.
+     * - In the case of `DerivationValuelessEmitter`, we don't control when the user calls
+     *   `getValue()` (as opposed to `DerivationEmitter` which calls `getValue()` directly).
+     *   Perhaps we can disable the check in that case.
+     * - Probably the best place to add this check is right here in this method plus some changes to `reportResulutionStart()`,
+     *   which would have to be changed to let the caller know if there is an actual collector (a prism)
+     *   present in its stack.
+     */
+    reportResolutionStart(this)
+
+    if (!this._isFresh) {
+      const newValue = this._recalculate()
+      this._lastValue = newValue
+      if (this._isHot) {
+        this._isFresh = true
+        this._didMarkDependentsAsStale = false
+      }
+    }
+
+    reportResolutionEnd(this)
+    return this._lastValue!
+  }
+
+  private _reactToNumberOfDependentsChange() {
+    const shouldBecomeHot = this._dependents.size > 0
+
+    if (shouldBecomeHot === this._isHot) return
+
+    this._isHot = shouldBecomeHot
+    this._didMarkDependentsAsStale = false
+    this._isFresh = false
+    if (shouldBecomeHot) {
+      for (const d of this._dependencies) {
+        d.addDependent(this._markAsStale)
+      }
+      this._keepHot()
+    } else {
+      for (const d of this._dependencies) {
+        d.removeDependent(this._markAsStale)
+      }
+      this._becomeCold()
+    }
+  }
+
+  /**
+   * A simple mapping function similar to Array.map()
+   *
+   * @deprecated This is a remnant of the old monadic api. Now it's functionally equal to `prism(() => fn(der.getValue()))`, so use that instead.
+   */
+  map<T>(fn: (v: V) => T): IDerivation<T> {
+    console.log('map')
+
+    return prism(() => fn(this.getValue()))
+  }
+
+  /**
+   * Same as {@link AbstractDerivation.map}, but the mapping function can also return a derivation, in which case the derivation returned
+   * by `flatMap` takes the value of that derivation.
+   *
+   * @deprecated This is a remnant of the old monadic api. Now it's functionally equal to `prism(() => val(fn(val(der))))`
+   *
+   * @example
+   * ```ts
+   * // Simply using map() here would return the inner derivation when we call getValue()
+   * new Box(3).derivation.map((value) => new Box(value).derivation).getValue()
+   *
+   * // Using flatMap() eliminates the inner derivation
+   * new Box(3).derivation.flatMap((value) => new Box(value).derivation).getValue()
+   * ```
+   *
+   * @param fn - The mapping function to use. Note: it accepts a plain value, not a derivation.
+   */
+  flatMap<R>(
+    fn: (v: V) => R,
+  ): IDerivation<R extends IDerivation<infer T> ? T : R> {
+    console.log('flatMap')
+    return prism(() => {
+      return possibleDerivationToValue(fn(this.getValue()))
+    })
   }
 
   _recalculate() {
@@ -376,6 +618,18 @@ function inPrism(): boolean {
   return !!hookScopeStack.peek()
 }
 
+const possibleDerivationToValue = <
+  P extends IDerivation<$IntentionalAny> | unknown,
+>(
+  input: P,
+): P extends IDerivation<infer T> ? T : P => {
+  if (isDerivation(input)) {
+    return input.getValue() as $IntentionalAny
+  } else {
+    return input as $IntentionalAny
+  }
+}
+
 type IPrismFn = {
   <T>(fn: () => T): IDerivation<T>
   ref: typeof ref