svelte/src/internal/client/reactivity/batch.js

Cybernetically enhanced web apps

/** @import { Fork } from 'svelte' */
/** @import { Derived, Effect, Reaction, Source, Value } from '#client' */
import {
	BLOCK_EFFECT,
	BRANCH_EFFECT,
	CLEAN,
	DESTROYED,
	DIRTY,
	EFFECT,
	ASYNC,
	INERT,
	RENDER_EFFECT,
	ROOT_EFFECT,
	MAYBE_DIRTY,
	DERIVED,
	EAGER_EFFECT,
	HEAD_EFFECT,
	ERROR_VALUE,
	MANAGED_EFFECT,
	REACTION_RAN
} from '#client/constants';
import { async_mode_flag } from '../../flags/index.js';
import { deferred, define_property, includes } from '../../shared/utils.js';
import {
	active_effect,
	get,
	increment_write_version,
	is_dirty,
	update_effect
} from '../runtime.js';
import * as e from '../errors.js';
import { flush_tasks, queue_micro_task } from '../dom/task.js';
import { DEV } from 'esm-env';
import { invoke_error_boundary } from '../error-handling.js';
import { flush_eager_effects, old_values, set_eager_effects, source, update } from './sources.js';
import { eager_effect, unlink_effect } from './effects.js';
import { defer_effect } from './utils.js';
import { UNINITIALIZED } from '../../../constants.js';
import { set_signal_status } from './status.js';

/** @type {Set<Batch>} */
const batches = new Set();

/** @type {Batch | null} */
export let current_batch = null;

/**
 * This is needed to avoid overwriting inputs in non-async mode
 * TODO 6.0 remove this, as non-async mode will go away
 * @type {Batch | null}
 */
export let previous_batch = null;

/**
 * When time travelling (i.e. working in one batch, while other batches
 * still have ongoing work), we ignore the real values of affected
 * signals in favour of their values within the batch
 * @type {Map<Value, any> | null}
 */
export let batch_values = null;

// TODO this should really be a property of `batch`
/** @type {Effect[]} */
let queued_root_effects = [];

/** @type {Effect | null} */
let last_scheduled_effect = null;

export let is_flushing_sync = false;

/**
 * During traversal, this is an array. Newly created effects are (if not immediately
 * executed) pushed to this array, rather than going through the scheduling
 * rigamarole that would cause another turn of the flush loop.
 * @type {Effect[] | null}
 */
export let collected_effects = null;

export class Batch {
	/**
	 * The current values of any sources that are updated in this batch
	 * They keys of this map are identical to `this.#previous`
	 * @type {Map<Source, any>}
	 */
	current = new Map();

	/**
	 * The values of any sources that are updated in this batch _before_ those updates took place.
	 * They keys of this map are identical to `this.#current`
	 * @type {Map<Source, any>}
	 */
	previous = new Map();

	/**
	 * When the batch is committed (and the DOM is updated), we need to remove old branches
	 * and append new ones by calling the functions added inside (if/each/key/etc) blocks
	 * @type {Set<(batch: Batch) => void>}
	 */
	#commit_callbacks = new Set();

	/**
	 * If a fork is discarded, we need to destroy any effects that are no longer needed
	 * @type {Set<(batch: Batch) => void>}
	 */
	#discard_callbacks = new Set();

	/**
	 * The number of async effects that are currently in flight
	 */
	#pending = 0;

	/**
	 * The number of async effects that are currently in flight, _not_ inside a pending boundary
	 */
	#blocking_pending = 0;

	/**
	 * A deferred that resolves when the batch is committed, used with `settled()`
	 * TODO replace with Promise.withResolvers once supported widely enough
	 * @type {{ promise: Promise<void>, resolve: (value?: any) => void, reject: (reason: unknown) => void } | null}
	 */
	#deferred = null;

	/**
	 * Deferred effects (which run after async work has completed) that are DIRTY
	 * @type {Set<Effect>}
	 */
	#dirty_effects = new Set();

	/**
	 * Deferred effects that are MAYBE_DIRTY
	 * @type {Set<Effect>}
	 */
	#maybe_dirty_effects = new Set();

	/**
	 * A map of branches that still exist, but will be destroyed when this batch
	 * is committed — we skip over these during `process`.
	 * The value contains child effects that were dirty/maybe_dirty before being reset,
	 * so they can be rescheduled if the branch survives.
	 * @type {Map<Effect, { d: Effect[], m: Effect[] }>}
	 */
	#skipped_branches = new Map();

	is_fork = false;

	#decrement_queued = false;

	#is_deferred() {
		return this.is_fork || this.#blocking_pending > 0;
	}

	/**
	 * Add an effect to the #skipped_branches map and reset its children
	 * @param {Effect} effect
	 */
	skip_effect(effect) {
		if (!this.#skipped_branches.has(effect)) {
			this.#skipped_branches.set(effect, { d: [], m: [] });
		}
	}

	/**
	 * Remove an effect from the #skipped_branches map and reschedule
	 * any tracked dirty/maybe_dirty child effects
	 * @param {Effect} effect
	 */
	unskip_effect(effect) {
		var tracked = this.#skipped_branches.get(effect);
		if (tracked) {
			this.#skipped_branches.delete(effect);

			for (var e of tracked.d) {
				set_signal_status(e, DIRTY);
				schedule_effect(e);
			}

			for (e of tracked.m) {
				set_signal_status(e, MAYBE_DIRTY);
				schedule_effect(e);
			}
		}
	}

	/**
	 *
	 * @param {Effect[]} root_effects
	 */
	process(root_effects) {
		queued_root_effects = [];

		this.apply();

		/** @type {Effect[]} */
		var effects = (collected_effects = []);

		/** @type {Effect[]} */
		var render_effects = [];

		for (const root of root_effects) {
			this.#traverse_effect_tree(root, effects, render_effects);
			// Note: #traverse_effect_tree runs block effects eagerly, which can schedule effects,
			// which means queued_root_effects now may be filled again.

			// Helpful for debugging reactivity loss that has to do with branches being skipped:
			// log_inconsistent_branches(root);
		}

		collected_effects = null;

		if (this.#is_deferred()) {
			this.#defer_effects(render_effects);
			this.#defer_effects(effects);

			for (const [e, t] of this.#skipped_branches) {
				reset_branch(e, t);
			}
		} else {
			// If sources are written to, then work needs to happen in a separate batch, else prior sources would be mixed with
			// newly updated sources, which could lead to infinite loops when effects run over and over again.
			previous_batch = this;
			current_batch = null;

			// append/remove branches
			for (const fn of this.#commit_callbacks) fn(this);
			this.#commit_callbacks.clear();

			if (this.#pending === 0) {
				this.#commit();
			}

			flush_queued_effects(render_effects);
			flush_queued_effects(effects);

			// Clear effects. Those that are still needed will be rescheduled through unskipping the skipped branches.
			this.#dirty_effects.clear();
			this.#maybe_dirty_effects.clear();

			previous_batch = null;

			this.#deferred?.resolve();
		}

		batch_values = null;
	}

	/**
	 * Traverse the effect tree, executing effects or stashing
	 * them for later execution as appropriate
	 * @param {Effect} root
	 * @param {Effect[]} effects
	 * @param {Effect[]} render_effects
	 */
	#traverse_effect_tree(root, effects, render_effects) {
		root.f ^= CLEAN;

		var effect = root.first;

		while (effect !== null) {
			var flags = effect.f;
			var is_branch = (flags & (BRANCH_EFFECT | ROOT_EFFECT)) !== 0;
			var is_skippable_branch = is_branch && (flags & CLEAN) !== 0;

			var skip = is_skippable_branch || (flags & INERT) !== 0 || this.#skipped_branches.has(effect);

			if (!skip && effect.fn !== null) {
				if (is_branch) {
					effect.f ^= CLEAN;
				} else if ((flags & EFFECT) !== 0) {
					effects.push(effect);
				} else if (async_mode_flag && (flags & (RENDER_EFFECT | MANAGED_EFFECT)) !== 0) {
					render_effects.push(effect);
				} else if (is_dirty(effect)) {
					if ((flags & BLOCK_EFFECT) !== 0) this.#maybe_dirty_effects.add(effect);
					update_effect(effect);
				}

				var child = effect.first;

				if (child !== null) {
					effect = child;
					continue;
				}
			}

			while (effect !== null) {
				var next = effect.next;

				if (next !== null) {
					effect = next;
					break;
				}

				effect = effect.parent;
			}
		}
	}

	/**
	 * @param {Effect[]} effects
	 */
	#defer_effects(effects) {
		for (var i = 0; i < effects.length; i += 1) {
			defer_effect(effects[i], this.#dirty_effects, this.#maybe_dirty_effects);
		}
	}

	/**
	 * Associate a change to a given source with the current
	 * batch, noting its previous and current values
	 * @param {Source} source
	 * @param {any} value
	 */
	capture(source, value) {
		if (value !== UNINITIALIZED && !this.previous.has(source)) {
			this.previous.set(source, value);
		}

		// Don't save errors in `batch_values`, or they won't be thrown in `runtime.js#get`
		if ((source.f & ERROR_VALUE) === 0) {
			this.current.set(source, source.v);
			batch_values?.set(source, source.v);
		}
	}

	activate() {
		current_batch = this;
		this.apply();
	}

	deactivate() {
		// If we're not the current batch, don't deactivate,
		// else we could create zombie batches that are never flushed
		if (current_batch !== this) return;

		current_batch = null;
		batch_values = null;
	}

	flush() {
		if (queued_root_effects.length > 0) {
			current_batch = this;
			flush_effects();
		} else if (this.#pending === 0 && !this.is_fork) {
			// append/remove branches
			for (const fn of this.#commit_callbacks) fn(this);
			this.#commit_callbacks.clear();

			this.#commit();
			this.#deferred?.resolve();
		}

		this.deactivate();
	}

	discard() {
		for (const fn of this.#discard_callbacks) fn(this);
		this.#discard_callbacks.clear();
	}

	#commit() {
		// If there are other pending batches, they now need to be 'rebased' —
		// in other words, we re-run block/async effects with the newly
		// committed state, unless the batch in question has a more
		// recent value for a given source
		if (batches.size > 1) {
			this.previous.clear();

			var previous_batch = current_batch;
			var previous_batch_values = batch_values;
			var is_earlier = true;

			for (const batch of batches) {
				if (batch === this) {
					is_earlier = false;
					continue;
				}

				/** @type {Source[]} */
				const sources = [];

				for (const [source, value] of this.current) {
					if (batch.current.has(source)) {
						if (is_earlier && value !== batch.current.get(source)) {
							// bring the value up to date
							batch.current.set(source, value);
						} else {
							// same value or later batch has more recent value,
							// no need to re-run these effects
							continue;
						}
					}

					sources.push(source);
				}

				if (sources.length === 0) {
					continue;
				}

				// Re-run async/block effects that depend on distinct values changed in both batches
				const others = [...batch.current.keys()].filter((s) => !this.current.has(s));
				if (others.length > 0) {
					// Avoid running queued root effects on the wrong branch
					var prev_queued_root_effects = queued_root_effects;
					queued_root_effects = [];

					/** @type {Set<Value>} */
					const marked = new Set();
					/** @type {Map<Reaction, boolean>} */
					const checked = new Map();
					for (const source of sources) {
						mark_effects(source, others, marked, checked);
					}

					if (queued_root_effects.length > 0) {
						current_batch = batch;
						batch.apply();

						for (const root of queued_root_effects) {
							batch.#traverse_effect_tree(root, [], []);
						}

						// TODO do we need to do anything with the dummy effect arrays?

						batch.deactivate();
					}

					queued_root_effects = prev_queued_root_effects;
				}
			}

			current_batch = previous_batch;
			batch_values = previous_batch_values;
		}

		this.#skipped_branches.clear();
		batches.delete(this);
	}

	/**
	 *
	 * @param {boolean} blocking
	 */
	increment(blocking) {
		this.#pending += 1;
		if (blocking) this.#blocking_pending += 1;
	}

	/**
	 *
	 * @param {boolean} blocking
	 */
	decrement(blocking) {
		this.#pending -= 1;
		if (blocking) this.#blocking_pending -= 1;

		if (this.#decrement_queued) return;
		this.#decrement_queued = true;

		queue_micro_task(() => {
			this.#decrement_queued = false;

			if (!this.#is_deferred()) {
				// we only reschedule previously-deferred effects if we expect
				// to be able to run them after processing the batch
				this.revive();
			} else if (queued_root_effects.length > 0) {
				// if other effects are scheduled, process the batch _without_
				// rescheduling the previously-deferred effects
				this.flush();
			}
		});
	}

	revive() {
		for (const e of this.#dirty_effects) {
			this.#maybe_dirty_effects.delete(e);
			set_signal_status(e, DIRTY);
			schedule_effect(e);
		}

		for (const e of this.#maybe_dirty_effects) {
			set_signal_status(e, MAYBE_DIRTY);
			schedule_effect(e);
		}

		this.flush();
	}

	/** @param {(batch: Batch) => void} fn */
	oncommit(fn) {
		this.#commit_callbacks.add(fn);
	}

	/** @param {(batch: Batch) => void} fn */
	ondiscard(fn) {
		this.#discard_callbacks.add(fn);
	}

	settled() {
		return (this.#deferred ??= deferred()).promise;
	}

	static ensure() {
		if (current_batch === null) {
			const batch = (current_batch = new Batch());
			batches.add(current_batch);

			if (!is_flushing_sync) {
				queue_micro_task(() => {
					if (current_batch !== batch) {
						// a flushSync happened in the meantime
						return;
					}

					batch.flush();
				});
			}
		}

		return current_batch;
	}

	apply() {
		if (!async_mode_flag || (!this.is_fork && batches.size === 1)) return;

		// if there are multiple batches, we are 'time travelling' —
		// we need to override values with the ones in this batch...
		batch_values = new Map(this.current);

		// ...and undo changes belonging to other batches
		for (const batch of batches) {
			if (batch === this) continue;

			for (const [source, previous] of batch.previous) {
				if (!batch_values.has(source)) {
					batch_values.set(source, previous);
				}
			}
		}
	}
}

/**
 * Synchronously flush any pending updates.
 * Returns void if no callback is provided, otherwise returns the result of calling the callback.
 * @template [T=void]
 * @param {(() => T) | undefined} [fn]
 * @returns {T}
 */
export function flushSync(fn) {
	var was_flushing_sync = is_flushing_sync;
	is_flushing_sync = true;

	try {
		var result;

		if (fn) {
			if (current_batch !== null) {
				flush_effects();
			}

			result = fn();
		}

		while (true) {
			flush_tasks();

			if (queued_root_effects.length === 0) {
				current_batch?.flush();

				// we need to check again, in case we just updated an `$effect.pending()`
				if (queued_root_effects.length === 0) {
					// this would be reset in `flush_effects()` but since we are early returning here,
					// we need to reset it here as well in case the first time there's 0 queued root effects
					last_scheduled_effect = null;

					return /** @type {T} */ (result);
				}
			}

			flush_effects();
		}
	} finally {
		is_flushing_sync = was_flushing_sync;
	}
}

function flush_effects() {
	var source_stacks = DEV ? new Set() : null;

	try {
		var flush_count = 0;

		while (queued_root_effects.length > 0) {
			var batch = Batch.ensure();

			if (flush_count++ > 1000) {
				if (DEV) {
					var updates = new Map();

					for (const source of batch.current.keys()) {
						for (const [stack, update] of source.updated ?? []) {
							var entry = updates.get(stack);

							if (!entry) {
								entry = { error: update.error, count: 0 };
								updates.set(stack, entry);
							}

							entry.count += update.count;
						}
					}

					for (const update of updates.values()) {
						if (update.error) {
							// eslint-disable-next-line no-console
							console.error(update.error);
						}
					}
				}

				infinite_loop_guard();
			}

			batch.process(queued_root_effects);
			old_values.clear();

			if (DEV) {
				for (const source of batch.current.keys()) {
					/** @type {Set<Source>} */ (source_stacks).add(source);
				}
			}
		}
	} finally {
		queued_root_effects = [];

		last_scheduled_effect = null;
		collected_effects = null;

		if (DEV) {
			for (const source of /** @type {Set<Source>} */ (source_stacks)) {
				source.updated = null;
			}
		}
	}
}

function infinite_loop_guard() {
	try {
		e.effect_update_depth_exceeded();
	} catch (error) {
		if (DEV) {
			// stack contains no useful information, replace it
			define_property(error, 'stack', { value: '' });
		}

		// Best effort: invoke the boundary nearest the most recent
		// effect and hope that it's relevant to the infinite loop
		invoke_error_boundary(error, last_scheduled_effect);
	}
}

/** @type {Set<Effect> | null} */
export let eager_block_effects = null;

/**
 * @param {Array<Effect>} effects
 * @returns {void}
 */
function flush_queued_effects(effects) {
	var length = effects.length;
	if (length === 0) return;

	var i = 0;

	while (i < length) {
		var effect = effects[i++];

		if ((effect.f & (DESTROYED | INERT)) === 0 && is_dirty(effect)) {
			eager_block_effects = new Set();

			update_effect(effect);

			// Effects with no dependencies or teardown do not get added to the effect tree.
			// Deferred effects (e.g. `$effect(...)`) _are_ added to the tree because we
			// don't know if we need to keep them until they are executed. Doing the check
			// here (rather than in `update_effect`) allows us to skip the work for
			// immediate effects.
			if (
				effect.deps === null &&
				effect.first === null &&
				effect.nodes === null &&
				effect.teardown === null &&
				effect.ac === null
			) {
				// remove this effect from the graph
				unlink_effect(effect);
			}

			// If update_effect() has a flushSync() in it, we may have flushed another flush_queued_effects(),
			// which already handled this logic and did set eager_block_effects to null.
			if (eager_block_effects?.size > 0) {
				old_values.clear();

				for (const e of eager_block_effects) {
					// Skip eager effects that have already been unmounted
					if ((e.f & (DESTROYED | INERT)) !== 0) continue;

					// Run effects in order from ancestor to descendant, else we could run into nullpointers
					/** @type {Effect[]} */
					const ordered_effects = [e];
					let ancestor = e.parent;
					while (ancestor !== null) {
						if (eager_block_effects.has(ancestor)) {
							eager_block_effects.delete(ancestor);
							ordered_effects.push(ancestor);
						}
						ancestor = ancestor.parent;
					}

					for (let j = ordered_effects.length - 1; j >= 0; j--) {
						const e = ordered_effects[j];
						// Skip eager effects that have already been unmounted
						if ((e.f & (DESTROYED | INERT)) !== 0) continue;
						update_effect(e);
					}
				}

				eager_block_effects.clear();
			}
		}
	}

	eager_block_effects = null;
}

/**
 * This is similar to `mark_reactions`, but it only marks async/block effects
 * depending on `value` and at least one of the other `sources`, so that
 * these effects can re-run after another batch has been committed
 * @param {Value} value
 * @param {Source[]} sources
 * @param {Set<Value>} marked
 * @param {Map<Reaction, boolean>} checked
 */
function mark_effects(value, sources, marked, checked) {
	if (marked.has(value)) return;
	marked.add(value);

	if (value.reactions !== null) {
		for (const reaction of value.reactions) {
			const flags = reaction.f;

			if ((flags & DERIVED) !== 0) {
				mark_effects(/** @type {Derived} */ (reaction), sources, marked, checked);
			} else if (
				(flags & (ASYNC | BLOCK_EFFECT)) !== 0 &&
				(flags & DIRTY) === 0 &&
				depends_on(reaction, sources, checked)
			) {
				set_signal_status(reaction, DIRTY);
				schedule_effect(/** @type {Effect} */ (reaction));
			}
		}
	}
}

/**
 * When committing a fork, we need to trigger eager effects so that
 * any `$state.eager(...)` expressions update immediately. This
 * function allows us to discover them
 * @param {Value} value
 * @param {Set<Effect>} effects
 */
function mark_eager_effects(value, effects) {
	if (value.reactions === null) return;

	for (const reaction of value.reactions) {
		const flags = reaction.f;

		if ((flags & DERIVED) !== 0) {
			mark_eager_effects(/** @type {Derived} */ (reaction), effects);
		} else if ((flags & EAGER_EFFECT) !== 0) {
			set_signal_status(reaction, DIRTY);
			effects.add(/** @type {Effect} */ (reaction));
		}
	}
}

/**
 * @param {Reaction} reaction
 * @param {Source[]} sources
 * @param {Map<Reaction, boolean>} checked
 */
function depends_on(reaction, sources, checked) {
	const depends = checked.get(reaction);
	if (depends !== undefined) return depends;

	if (reaction.deps !== null) {
		for (const dep of reaction.deps) {
			if (includes.call(sources, dep)) {
				return true;
			}

			if ((dep.f & DERIVED) !== 0 && depends_on(/** @type {Derived} */ (dep), sources, checked)) {
				checked.set(/** @type {Derived} */ (dep), true);
				return true;
			}
		}
	}

	checked.set(reaction, false);

	return false;
}

/**
 * @param {Effect} signal
 * @returns {void}
 */
export function schedule_effect(signal) {
	var effect = (last_scheduled_effect = signal);

	var boundary = effect.b;

	// defer render effects inside a pending boundary
	// TODO the `REACTION_RAN` check is only necessary because of legacy `$:` effects AFAICT — we can remove later
	if (
		boundary?.is_pending &&
		(signal.f & (EFFECT | RENDER_EFFECT | MANAGED_EFFECT)) !== 0 &&
		(signal.f & REACTION_RAN) === 0
	) {
		boundary.defer_effect(signal);
		return;
	}

	while (effect.parent !== null) {
		effect = effect.parent;
		var flags = effect.f;

		// if the effect is being scheduled because a parent (each/await/etc) block
		// updated an internal source, or because a branch is being unskipped,
		// bail out or we'll cause a second flush
		if (collected_effects !== null && effect === active_effect) {
			// in sync mode, render effects run during traversal. in an extreme edge case
			// they can be made dirty after they have already been visited, in which
			// case we shouldn't bail out
			if (async_mode_flag || (signal.f & RENDER_EFFECT) === 0) {
				return;
			}
		}

		if ((flags & (ROOT_EFFECT | BRANCH_EFFECT)) !== 0) {
			if ((flags & CLEAN) === 0) {
				// branch is already dirty, bail
				return;
			}

			effect.f ^= CLEAN;
		}
	}

	queued_root_effects.push(effect);
}

/** @type {Source<number>[]} */
let eager_versions = [];

function eager_flush() {
	try {
		flushSync(() => {
			for (const version of eager_versions) {
				update(version);
			}
		});
	} finally {
		eager_versions = [];
	}
}

/**
 * Implementation of `$state.eager(fn())`
 * @template T
 * @param {() => T} fn
 * @returns {T}
 */
export function eager(fn) {
	var version = source(0);
	var initial = true;
	var value = /** @type {T} */ (undefined);

	get(version);

	eager_effect(() => {
		if (initial) {
			// the first time this runs, we create an eager effect
			// that will run eagerly whenever the expression changes
			var previous_batch_values = batch_values;

			try {
				batch_values = null;
				value = fn();
			} finally {
				batch_values = previous_batch_values;
			}

			return;
		}

		// the second time this effect runs, it's to schedule a
		// `version` update. since this will recreate the effect,
		// we don't need to evaluate the expression here
		if (eager_versions.length === 0) {
			queue_micro_task(eager_flush);
		}

		eager_versions.push(version);
	});

	initial = false;

	return value;
}

/**
 * Mark all the effects inside a skipped branch CLEAN, so that
 * they can be correctly rescheduled later. Tracks dirty and maybe_dirty
 * effects so they can be rescheduled if the branch survives.
 * @param {Effect} effect
 * @param {{ d: Effect[], m: Effect[] }} tracked
 */
function reset_branch(effect, tracked) {
	// clean branch = nothing dirty inside, no need to traverse further
	if ((effect.f & BRANCH_EFFECT) !== 0 && (effect.f & CLEAN) !== 0) {
		return;
	}

	if ((effect.f & DIRTY) !== 0) {
		tracked.d.push(effect);
	} else if ((effect.f & MAYBE_DIRTY) !== 0) {
		tracked.m.push(effect);
	}

	set_signal_status(effect, CLEAN);

	var e = effect.first;
	while (e !== null) {
		reset_branch(e, tracked);
		e = e.next;
	}
}

/**
 * Creates a 'fork', in which state changes are evaluated but not applied to the DOM.
 * This is useful for speculatively loading data (for example) when you suspect that
 * the user is about to take some action.
 *
 * Frameworks like SvelteKit can use this to preload data when the user touches or
 * hovers over a link, making any subsequent navigation feel instantaneous.
 *
 * The `fn` parameter is a synchronous function that modifies some state. The
 * state changes will be reverted after the fork is initialised, then reapplied
 * if and when the fork is eventually committed.
 *
 * When it becomes clear that a fork will _not_ be committed (e.g. because the
 * user navigated elsewhere), it must be discarded to avoid leaking memory.
 *
 * @param {() => void} fn
 * @returns {Fork}
 * @since 5.42
 */
export function fork(fn) {
	if (!async_mode_flag) {
		e.experimental_async_required('fork');
	}

	if (current_batch !== null) {
		e.fork_timing();
	}

	var batch = Batch.ensure();
	batch.is_fork = true;
	batch_values = new Map();

	var committed = false;
	var settled = batch.settled();

	flushSync(fn);

	// revert state changes
	for (var [source, value] of batch.previous) {
		source.v = value;
	}

	// make writable deriveds dirty, so they recalculate correctly
	for (source of batch.current.keys()) {
		if ((source.f & DERIVED) !== 0) {
			set_signal_status(source, DIRTY);
		}
	}

	return {
		commit: async () => {
			if (committed) {
				await settled;
				return;
			}

			if (!batches.has(batch)) {
				e.fork_discarded();
			}

			committed = true;

			batch.is_fork = false;

			// apply changes and update write versions so deriveds see the change
			for (var [source, value] of batch.current) {
				source.v = value;
				source.wv = increment_write_version();
			}

			// trigger any `$state.eager(...)` expressions with the new state.
			// eager effects don't get scheduled like other effects, so we
			// can't just encounter them during traversal, we need to
			// proactively flush them
			// TODO maybe there's a better implementation?
			flushSync(() => {
				/** @type {Set<Effect>} */
				var eager_effects = new Set();

				for (var source of batch.current.keys()) {
					mark_eager_effects(source, eager_effects);
				}

				set_eager_effects(eager_effects);
				flush_eager_effects();
			});

			batch.revive();
			await settled;
		},
		discard: () => {
			// cause any MAYBE_DIRTY deriveds to update
			// if they depend on things thath changed
			// inside the discarded fork
			for (var source of batch.current.keys()) {
				source.wv = increment_write_version();
			}

			if (!committed && batches.has(batch)) {
				batches.delete(batch);
				batch.discard();
			}
		}
	};
}

/**
 * Forcibly remove all current batches, to prevent cross-talk between tests
 */
export function clear() {
	batches.clear();
}