nedb/lib/datastore.js

const { EventEmitter } = require('events')
const { callbackify, deprecate } = require('util')
const Cursor = require('./cursor.js')
const customUtils = require('./customUtils.js')
const Executor = require('./executor.js')
const Index = require('./indexes.js')
const model = require('./model.js')
const Persistence = require('./persistence.js')
const { isDate } = require('./utils.js')

/**
 * Callback with no parameter
 * @callback NoParamCallback
 * @param {?Error} err
 */

/**
 * String comparison function.
 * ```
 *   if (a < b) return -1
 *   if (a > b) return 1
 *   return 0
 * ```
 * @callback compareStrings
 * @param {string} a
 * @param {string} b
 * @return {number}
 */

/**
 * Callback that returns an Array of documents
 * @callback MultipleDocumentsCallback
 * @param {?Error} err
 * @param {?document[]} docs
 */

/**
 * Callback that returns a single document
 * @callback SingleDocumentCallback
 * @param {?Error} err
 * @param {?document} docs
 */

/**
 * Generic async function
 * @callback AsyncFunction
 * @param {...*} args
 * @return {Promise<*>}
 */

/**
 * Compaction event. Happens when the Datastore's Persistence has been compacted.
 * It happens when calling `datastore.persistence.compactDatafile`, which is called periodically if you have called
 * `datastore.persistence.setAutocompactionInterval`.
 *
 * @event Datastore#event:"compaction.done"
 * @type {undefined}
 */

/**
 * Generic document in NeDB.
 * It consists of an Object with anything you want inside.
 * @typedef document
 * @property {?string} [_id] Internal `_id` of the document, which can be `null` or undefined at some points (when not
 * inserted yet for example).
 * @type {Object.<string, *>}
 */

/**
 * Nedb query.
 *
 * Each key of a query references a field name, which can use the dot-notation to reference subfields inside nested
 * documents, arrays, arrays of subdocuments and to match a specific element of an array.
 *
 * Each value of a query can be one of the following:
 * - `string`: matches all documents which have this string as value for the referenced field name
 * - `number`: matches all documents which have this number as value for the referenced field name
 * - `Regexp`: matches all documents which have a value that matches the given `Regexp` for the referenced field name
 * - `object`: matches all documents which have this object as deep-value for the referenced field name
 * - Comparison operators: the syntax is `{ field: { $op: value } }` where `$op` is any comparison operator:
 *   - `$lt`, `$lte`: less than, less than or equal
 *   - `$gt`, `$gte`: greater than, greater than or equal
 *   - `$in`: member of. `value` must be an array of values
 *   - `$ne`, `$nin`: not equal, not a member of
 *   - `$stat`: checks whether the document posses the property `field`. `value` should be true or false
 *   - `$regex`: checks whether a string is matched by the regular expression. Contrary to MongoDB, the use of
 *   `$options` with `$regex` is not supported, because it doesn't give you more power than regex flags. Basic
 *   queries are more readable so only use the `$regex` operator when you need to use another operator with it
 *   - `$size`: if the referenced filed is an Array, matches on the size of the array
 *   - `$elemMatch`: matches if at least one array element matches the sub-query entirely
 * - Logical operators: You can combine queries using logical operators:
 *   - For `$or` and `$and`, the syntax is `{ $op: [query1, query2, ...] }`.
 *   - For `$not`, the syntax is `{ $not: query }`
 *   - For `$where`, the syntax is:
 *   ```
 *   { $where: function () {
 *     // object is 'this'
 *     // return a boolean
 *   } }
 *   ```
 * @typedef query
 * @type {Object.<string, *>}
 */

/**
 * Nedb projection.
 *
 * You can give `find` and `findOne` an optional second argument, `projections`.
 * The syntax is the same as MongoDB: `{ a: 1, b: 1 }` to return only the `a`
 * and `b` fields, `{ a: 0, b: 0 }` to omit these two fields. You cannot use both
 * modes at the time, except for `_id` which is by default always returned and
 * which you can choose to omit. You can project on nested documents.
 *
 * To reference subfields, you can use the dot-notation.
 *
 * @typedef projection
 * @type {Object.<string, 0|1>}
 */

/**
 * The `beforeDeserialization`and `afterDeserialization` callbacks should
 * @callback serializationHook
 * @param {string} x
 * @return {string}
 */

/**
 * The `Datastore` class is the main class of NeDB.
 * @extends EventEmitter
 */
class Datastore extends EventEmitter {
  /**
   * Create a new collection, either persistent or in-memory.
   *
   * If you use a persistent datastore without the `autoload` option, you need to call `loadDatabase` manually. This
   * function fetches the data from datafile and prepares the database. **Don't forget it!** If you use a persistent
   * datastore, no command (insert, find, update, remove) will be executed before `loadDatabase` is called, so make sure
   * to call it yourself or use the `autoload` option.
   *
   * @param {object|string} options Can be an object or a string. If options is a string, the behavior is the same as in
   * v0.6: it will be interpreted as `options.filename`. **Giving a string is deprecated, and will be removed in the
   * next major version.**
   * @param {string} [options.filename = null] Path to the file where the data is persisted. If left blank, the datastore is
   * automatically considered in-memory only. It cannot end with a `~` which is used in the temporary files NeDB uses to
   * perform crash-safe writes.
   * @param {boolean} [options.inMemoryOnly = false] If set to true, no data will be written in storage.
   * @param {boolean} [options.timestampData = false] If set to true, createdAt and updatedAt will be created and
   * populated automatically (if not specified by user)
   * @param {boolean} [options.autoload = false] If used, the database will automatically be loaded from the datafile
   * upon creation (you don't need to call `loadDatabase`). Any command issued before load is finished is buffered and
   * will be executed when load is done. When autoloading is done, you can either use the `onload` callback, or you can
   * use `this.autoloadPromise` which resolves (or rejects) when autloading is done.
   * @param {function} [options.onload] If you use autoloading, this is the handler called after the `loadDatabase`. It
   * takes one `error` argument. If you use autoloading without specifying this handler, and an error happens during
   * load, an error will be thrown.
   * @param {function} [options.beforeDeserialization] Hook you can use to transform data after it was serialized and
   * before it is written to disk. Can be used for example to encrypt data before writing database to disk. This
   * function takes a string as parameter (one line of an NeDB data file) and outputs the transformed string, **which
   * must absolutely not contain a `\n` character** (or data will be lost).
   * @param {function} [options.afterSerialization] Inverse of `afterSerialization`. Make sure to include both and not
   * just one, or you risk data loss. For the same reason, make sure both functions are inverses of one another. Some
   * failsafe mechanisms are in place to prevent data loss if you misuse the serialization hooks: NeDB checks that never
   * one is declared without the other, and checks that they are reverse of one another by testing on random strings of
   * various lengths. In addition, if too much data is detected as corrupt, NeDB will refuse to start as it could mean
   * you're not using the deserialization hook corresponding to the serialization hook used before.
   * @param {number} [options.corruptAlertThreshold = 0.1] Between 0 and 1, defaults to 10%. NeDB will refuse to start
   * if more than this percentage of the datafile is corrupt. 0 means you don't tolerate any corruption, 1 means you
   * don't care.
   * @param {compareStrings} [options.compareStrings] If specified, it overrides default string comparison which is not
   * well adapted to non-US characters in particular accented letters. Native `localCompare` will most of the time be
   * the right choice.
   * @param {string} [options.nodeWebkitAppName] **Deprecated:** if you are using NeDB from whithin a Node Webkit app,
   * specify its name (the same one you use in the `package.json`) in this field and the `filename` will be relative to
   * the directory Node Webkit uses to store the rest of the application's data (local storage etc.). It works on Linux,
   * OS X and Windows. Now that you can use `require('nw.gui').App.dataPath` in Node Webkit to get the path to the data
   * directory for your application, you should not use this option anymore and it will be removed.
   *
   * @fires Datastore#event:"compaction.done"
   */
  constructor (options) {
    super()
    let filename

    // Retrocompatibility with v0.6 and before
    if (typeof options === 'string') {
      deprecate(() => {
        filename = options
        this.inMemoryOnly = false // Default
      }, 'Giving a string to the Datastore constructor is deprecated and will be removed in the next version. Please use an options object with an argument \'filename\'.')()
    } else {
      options = options || {}
      filename = options.filename
      /**
       * Determines if the `Datastore` keeps data in-memory, or if it saves it in storage. Is not read after
       * instanciation.
       * @type {boolean}
       * @protected
       */
      this.inMemoryOnly = options.inMemoryOnly || false
      /**
       * Determines if the `Datastore` should autoload the database upon instantiation. Is not read after instanciation.
       * @type {boolean}
       * @protected
       */
      this.autoload = options.autoload || false
      /**
       * Determines if the `Datastore` should add `createdAt` and `updatedAt` fields automatically if not set by the user.
       * @type {boolean}
       * @protected
       */
      this.timestampData = options.timestampData || false
    }

    // Determine whether in memory or persistent
    if (!filename || typeof filename !== 'string' || filename.length === 0) {
      /**
       * If null, it means `inMemoryOnly` is `true`. The `filename` is the name given to the storage module. Is not read
       * after instanciation.
       * @type {?string}
       * @protected
       */
      this.filename = null
      this.inMemoryOnly = true
    } else {
      this.filename = filename
    }

    // String comparison function
    /**
     * Overrides default string comparison which is not well adapted to non-US characters in particular accented
     * letters. Native `localCompare` will most of the time be the right choice
     * @type {compareStrings}
     * @function
     * @protected
     */
    this.compareStrings = options.compareStrings

    // Persistence handling
    /**
     * The `Persistence` instance for this `Datastore`.
     * @type {Persistence}
     */
    this.persistence = new Persistence({
      db: this,
      nodeWebkitAppName: options.nodeWebkitAppName,
      afterSerialization: options.afterSerialization,
      beforeDeserialization: options.beforeDeserialization,
      corruptAlertThreshold: options.corruptAlertThreshold
    })

    // This new executor is ready if we don't use persistence
    // If we do, it will only be ready once loadDatabase is called
    /**
     * The `Executor` instance for this `Datastore`. It is used in all methods exposed by the `Datastore`, any `Cursor`
     * produced by the `Datastore` and by `this.persistence.compactDataFile` & `this.persistence.compactDataFileAsync`
     * to ensure operations are performed sequentially in the database.
     * @type {Executor}
     * @protected
     */
    this.executor = new Executor()
    if (this.inMemoryOnly) this.executor.ready = true

    /**
     * Indexed by field name, dot notation can be used.
     * _id is always indexed and since _ids are generated randomly the underlying binary search tree is always well-balanced
     * @type {Object.<string, Index>}
     * @protected
     */
    this.indexes = {}
    this.indexes._id = new Index({ fieldName: '_id', unique: true })
    /**
     * Stores the time to live (TTL) of the indexes created. The key represents the field name, the value the number of
     * seconds after which data with this index field should be removed.
     * @type {Object.<string, number>}
     * @protected
     */
    this.ttlIndexes = {}

    // Queue a load of the database right away and call the onload handler
    // By default (no onload handler), if there is an error there, no operation will be possible so warn the user by throwing an exception
    if (this.autoload) {
      /**
       * A Promise that resolves when the autoload has finished.
       *
       * The onload callback is not awaited by this Promise, it is started immediately after that.
       * @type {?Promise}
       */
      this.autoloadPromise = this.loadDatabaseAsync()
      this.autoloadPromise
        .then(() => {
          if (options.onload) options.onload()
        }, err => {
          if (options.onload) options.onload(err)
          else throw err
        })
    }
  }

  /**
   * Load the database from the datafile, and trigger the execution of buffered commands if any.
   * @param {function} callback
   */
  loadDatabase (callback) {
    this.executor.push({ this: this.persistence, fn: this.persistence.loadDatabase, arguments: [callback] }, true)
  }

  /**
   * Async version of {@link Datastore#loadDatabase}.
   * @async
   * @return {Promise}
   * @see Datastore#loadDatabase
   */
  loadDatabaseAsync () {
    return this.executor.pushAsync(() => this.persistence.loadDatabaseAsync(), true)
  }

  /**
   * Get an array of all the data in the database.
   * @return {document[]}
   */
  getAllData () {
    return this.indexes._id.getAll()
  }

  /**
   * Reset all currently defined indexes.
   * @param {?document|?document[]} newData
   */
  resetIndexes (newData) {
    for (const index of Object.values(this.indexes)) {
      index.reset(newData)
    }
  }

  /**
   * Ensure an index is kept for this field. Same parameters as lib/indexes
   * This function acts synchronously on the indexes, however the persistence of the indexes is deferred with the
   * executor.
   * Previous versions said explicitly the callback was optional, it is now recommended setting one.
   * @param {object} options
   * @param {string} options.fieldName Name of the field to index. Use the dot notation to index a field in a nested document.
   * @param {boolean} [options.unique = false] Enforce field uniqueness. Note that a unique index will raise an error if you try to index two documents for which the field is not defined.
   * @param {boolean} [options.sparse = false] don't index documents for which the field is not defined. Use this option along with "unique" if you want to accept multiple documents for which it is not defined.
   * @param {number} [options.expireAfterSeconds] - if set, the created index is a TTL (time to live) index, that will automatically remove documents when the system date becomes larger than the date on the indexed field plus `expireAfterSeconds`. Documents where the indexed field is not specified or not a `Date` object are ignored
   * @param {NoParamCallback} callback Callback, signature: err
   */
  // TODO: contrary to what is said in the JSDoc, this function should probably be called through the executor, it persists a new state
  ensureIndex (options = {}, callback = () => {}) {
    callbackify(this.ensureIndexAsync.bind(this))(options, callback)
  }

  /**
   * Async version of {@link Datastore#ensureIndex}.
   * @param {object} options
   * @param {string} options.fieldName Name of the field to index. Use the dot notation to index a field in a nested document.
   * @param {boolean} [options.unique = false] Enforce field uniqueness. Note that a unique index will raise an error if you try to index two documents for which the field is not defined.
   * @param {boolean} [options.sparse = false] Don't index documents for which the field is not defined. Use this option along with "unique" if you want to accept multiple documents for which it is not defined.
   * @param {number} [options.expireAfterSeconds] - If set, the created index is a TTL (time to live) index, that will automatically remove documents when the system date becomes larger than the date on the indexed field plus `expireAfterSeconds`. Documents where the indexed field is not specified or not a `Date` object are ignored
   * @return {Promise<void>}
   * @see Datastore#ensureIndex
   */
  // TODO: contrary to what is said in the JSDoc, this function should probably be called through the executor, it persists a new state
  async ensureIndexAsync (options = {}) {
    if (!options.fieldName) {
      const err = new Error('Cannot create an index without a fieldName')
      err.missingFieldName = true
      throw err
    }
    if (this.indexes[options.fieldName]) return

    this.indexes[options.fieldName] = new Index(options)
    if (options.expireAfterSeconds !== undefined) this.ttlIndexes[options.fieldName] = options.expireAfterSeconds // With this implementation index creation is not necessary to ensure TTL but we stick with MongoDB's API here

    try {
      this.indexes[options.fieldName].insert(this.getAllData())
    } catch (e) {
      delete this.indexes[options.fieldName]
      throw e
    }

    // We may want to force all options to be persisted including defaults, not just the ones passed the index creation function
    await this.persistence.persistNewStateAsync([{ $$indexCreated: options }])
  }

  /**
   * Remove an index
   * Previous versions said explicitly the callback was optional, it is now recommended setting one.
   * @param {string} fieldName Field name of the index to remove. Use the dot notation to remove an index referring to a
   * field in a nested document.
   * @param {NoParamCallback} callback Optional callback, signature: err
   */
  // TODO: contrary to what is said in the JSDoc, this function should probably be called through the executor, it persists a new state
  removeIndex (fieldName, callback = () => {}) {
    callbackify(this.removeIndexAsync.bind(this))(fieldName, callback)
  }

  /**
   * Async version of {@link Datastore#removeIndex}.
   * @param {string} fieldName Field name of the index to remove. Use the dot notation to remove an index referring to a
   * field in a nested document.
   * @return {Promise<void>}
   * @see Datastore#removeIndex
   */
  // TODO: contrary to what is said in the JSDoc, this function should probably be called through the executor, it persists a new state
  async removeIndexAsync (fieldName) {
    delete this.indexes[fieldName]

    await this.persistence.persistNewStateAsync([{ $$indexRemoved: fieldName }])
  }

  /**
   * Add one or several document(s) to all indexes.
   *
   * This is an internal function.
   * @param {document} doc
   * @protected
   */
  addToIndexes (doc) {
    let failingIndex
    let error
    const keys = Object.keys(this.indexes)

    for (let i = 0; i < keys.length; i += 1) {
      try {
        this.indexes[keys[i]].insert(doc)
      } catch (e) {
        failingIndex = i
        error = e
        break
      }
    }

    // If an error happened, we need to rollback the insert on all other indexes
    if (error) {
      for (let i = 0; i < failingIndex; i += 1) {
        this.indexes[keys[i]].remove(doc)
      }

      throw error
    }
  }

  /**
   * Remove one or several document(s) from all indexes.
   *
   * This is an internal function.
   * @param {document} doc
   * @protected
   */
  removeFromIndexes (doc) {
    for (const index of Object.values(this.indexes)) {
      index.remove(doc)
    }
  }

  /**
   * Update one or several documents in all indexes.
   *
   * To update multiple documents, oldDoc must be an array of { oldDoc, newDoc } pairs.
   *
   * If one update violates a constraint, all changes are rolled back.
   *
   * This is an internal function.
   * @param {document|Array.<{oldDoc: document, newDoc: document}>} oldDoc Document to update, or an `Array` of
   * `{oldDoc, newDoc}` pairs.
   * @param {document} [newDoc] Document to replace the oldDoc with. If the first argument is an `Array` of
   * `{oldDoc, newDoc}` pairs, this second argument is ignored.
   */
  updateIndexes (oldDoc, newDoc) {
    let failingIndex
    let error
    const keys = Object.keys(this.indexes)

    for (let i = 0; i < keys.length; i += 1) {
      try {
        this.indexes[keys[i]].update(oldDoc, newDoc)
      } catch (e) {
        failingIndex = i
        error = e
        break
      }
    }

    // If an error happened, we need to rollback the update on all other indexes
    if (error) {
      for (let i = 0; i < failingIndex; i += 1) {
        this.indexes[keys[i]].revertUpdate(oldDoc, newDoc)
      }

      throw error
    }
  }

  /**
   * Get all candidate documents matching the query, regardless of their expiry status.
   * @param {query} query
   * @return {document[]}
   *
   * @private
   */
  _getCandidates (query) {
    const indexNames = Object.keys(this.indexes)
    // STEP 1: get candidates list by checking indexes from most to least frequent usecase
    // For a basic match
    let usableQuery
    usableQuery = Object.entries(query)
      .filter(([k, v]) =>
        !!(typeof v === 'string' || typeof v === 'number' || typeof v === 'boolean' || isDate(v) || v === null) &&
        indexNames.includes(k)
      )
      .pop()
    if (usableQuery) return this.indexes[usableQuery[0]].getMatching(usableQuery[1])
    // For a $in match
    usableQuery = Object.entries(query)
      .filter(([k, v]) =>
        !!(query[k] && Object.prototype.hasOwnProperty.call(query[k], '$in')) &&
        indexNames.includes(k)
      )
      .pop()
    if (usableQuery) return this.indexes[usableQuery[0]].getMatching(usableQuery[1].$in)
    // For a comparison match
    usableQuery = Object.entries(query)
      .filter(([k, v]) =>
        !!(query[k] && (Object.prototype.hasOwnProperty.call(query[k], '$lt') || Object.prototype.hasOwnProperty.call(query[k], '$lte') || Object.prototype.hasOwnProperty.call(query[k], '$gt') || Object.prototype.hasOwnProperty.call(query[k], '$gte'))) &&
        indexNames.includes(k)
      )
      .pop()
    if (usableQuery) return this.indexes[usableQuery[0]].getBetweenBounds(usableQuery[1])
    // By default, return all the DB data
    return this.getAllData()
  }

  /**
   * Return the list of candidates for a given query
   * Crude implementation for now, we return the candidates given by the first usable index if any
   * We try the following query types, in this order: basic match, $in match, comparison match
   * One way to make it better would be to enable the use of multiple indexes if the first usable index
   * returns too much data. I may do it in the future.
   *
   * Returned candidates will be scanned to find and remove all expired documents
   *
   * This is an internal function.
   * @param {query} query
   * @param {boolean|function} [dontExpireStaleDocs = false] If true don't remove stale docs. Useful for the remove
   * function which shouldn't be impacted by expirations. If argument is not given, it is used as the callback.
   * @param {MultipleDocumentsCallback} callback Signature err, candidates
   *
   * @protected
   */
  getCandidates (query, dontExpireStaleDocs, callback) {
    if (typeof dontExpireStaleDocs === 'function') {
      callback = dontExpireStaleDocs
      dontExpireStaleDocs = false
    }

    callbackify(this.getCandidatesAsync.bind(this))(query, dontExpireStaleDocs, callback)
  }

  /**
   * Async version of {@link Datastore#getCandidates}.
   *
   * This is an internal function.
   * @param {query} query
   * @param {boolean} [dontExpireStaleDocs = false] If true don't remove stale docs. Useful for the remove function
   * which shouldn't be impacted by expirations.
   * @return {Promise<document[]>} candidates
   * @see Datastore#getCandidates
   * @protected
   */
  async getCandidatesAsync (query, dontExpireStaleDocs = false) {
    const validDocs = []

    // STEP 1: get candidates list by checking indexes from most to least frequent usecase
    const docs = this._getCandidates(query)
    // STEP 2: remove all expired documents
    if (!dontExpireStaleDocs) {
      const expiredDocsIds = []
      const ttlIndexesFieldNames = Object.keys(this.ttlIndexes)

      docs.forEach(doc => {
        if (ttlIndexesFieldNames.every(i => !(doc[i] !== undefined && isDate(doc[i]) && Date.now() > doc[i].getTime() + this.ttlIndexes[i] * 1000))) validDocs.push(doc)
        else expiredDocsIds.push(doc._id)
      })
      for (const _id of expiredDocsIds) {
        await this._removeAsync({ _id: _id }, {})
      }
    } else validDocs.push(...docs)
    return validDocs
  }

  /**
   * Insert a new document
   * This is an internal function, use {@link Datastore#insert} which has the same signature.
   * @param {document|document[]} newDoc
   * @param {SingleDocumentCallback} callback
   *
   * @private
   */
  _insert (newDoc, callback) {
    return callbackify(this._insertAsync.bind(this))(newDoc, callback)
  }

  /**
   * Async version of {@link Datastore#_insert}.
   * @param {document|document[]} newDoc
   * @return {Promise<document|document[]>}
   * @private
   * @see Datastore#_insert
   */
  async _insertAsync (newDoc) {
    const preparedDoc = this._prepareDocumentForInsertion(newDoc)
    this._insertInCache(preparedDoc)

    await this.persistence.persistNewStateAsync(Array.isArray(preparedDoc) ? preparedDoc : [preparedDoc])
    return model.deepCopy(preparedDoc)
  }

  /**
   * Create a new _id that's not already in use
   * @return {string} id
   * @private
   */
  _createNewId () {
    let attemptId = customUtils.uid(16)
    // Try as many times as needed to get an unused _id. As explained in customUtils, the probability of this ever happening is extremely small, so this is O(1)
    if (this.indexes._id.getMatching(attemptId).length > 0) attemptId = this._createNewId()
    return attemptId
  }

  /**
   * Prepare a document (or array of documents) to be inserted in a database
   * Meaning adds _id and timestamps if necessary on a copy of newDoc to avoid any side effect on user input
   * @param {document|document[]} newDoc document, or Array of documents, to prepare
   * @return {document|document[]} prepared document, or Array of prepared documents
   * @private
   */
  _prepareDocumentForInsertion (newDoc) {
    let preparedDoc

    if (Array.isArray(newDoc)) {
      preparedDoc = []
      newDoc.forEach(doc => { preparedDoc.push(this._prepareDocumentForInsertion(doc)) })
    } else {
      preparedDoc = model.deepCopy(newDoc)
      if (preparedDoc._id === undefined) preparedDoc._id = this._createNewId()
      const now = new Date()
      if (this.timestampData && preparedDoc.createdAt === undefined) preparedDoc.createdAt = now
      if (this.timestampData && preparedDoc.updatedAt === undefined) preparedDoc.updatedAt = now
      model.checkObject(preparedDoc)
    }

    return preparedDoc
  }

  /**
   * If newDoc is an array of documents, this will insert all documents in the cache
   * @param {document|document[]} preparedDoc
   * @private
   */
  _insertInCache (preparedDoc) {
    if (Array.isArray(preparedDoc)) this._insertMultipleDocsInCache(preparedDoc)
    else this.addToIndexes(preparedDoc)
  }

  /**
   * If one insertion fails (e.g. because of a unique constraint), roll back all previous
   * inserts and throws the error
   * @param {document[]} preparedDocs
   * @private
   */
  _insertMultipleDocsInCache (preparedDocs) {
    let failingIndex
    let error

    for (let i = 0; i < preparedDocs.length; i += 1) {
      try {
        this.addToIndexes(preparedDocs[i])
      } catch (e) {
        error = e
        failingIndex = i
        break
      }
    }

    if (error) {
      for (let i = 0; i < failingIndex; i += 1) {
        this.removeFromIndexes(preparedDocs[i])
      }

      throw error
    }
  }

  /**
   * Insert a new document.
   * @param {document|document[]} newDoc
   * @param {SingleDocumentCallback} [callback = () => {}] Optional callback, signature: err, insertedDoc
   *
   * @private
   */
  insert (newDoc, callback = () => {}) {
    this.executor.push({ this: this, fn: this._insert, arguments: [newDoc, callback] })
  }

  /**
   * Async version of {@link Datastore#insert}.
   * @param {document|document[]} newDoc
   * @return {Promise<document>}
   * @async
   */
  insertAsync (newDoc) {
    return this.executor.pushAsync(() => this._insertAsync(newDoc))
  }

  /**
   * @callback Datastore~countCallback
   * @param {?Error} err
   * @param {?number} count
   */

  /**
   * Count all documents matching the query.
   * @param {query} query MongoDB-style query
   * @param {Datastore~countCallback} [callback] If given, the function will return undefined, otherwise it will return the Cursor.
   * @return {Cursor<number>|undefined}
   */
  count (query, callback) {
    const cursor = this.countAsync(query)

    if (typeof callback === 'function') callbackify(cursor.execAsync.bind(cursor))(callback)
    else return cursor
  }

  /**
   * Async version of {@link Datastore#count}.
   * @param {query} query MongoDB-style query
   * @return {Cursor<number>} count
   * @async
   */
  countAsync (query) {
    return new Cursor(this, query, async docs => docs.length, false)
  }

  /**
   * Find all documents matching the query
   * If no callback is passed, we return the cursor so that user can limit, skip and finally exec
   * @param {query} query MongoDB-style query
   * @param {projection|MultipleDocumentsCallback} [projection = {}] MongoDB-style projection. If not given, will be
   * interpreted as the callback.
   * @param {MultipleDocumentsCallback} [callback] Optional callback, signature: err, docs
   * @return {Cursor<document[]>|undefined}
   */
  find (query, projection, callback) {
    if (arguments.length === 1) {
      projection = {}
      // callback is undefined, will return a cursor
    } else if (arguments.length === 2) {
      if (typeof projection === 'function') {
        callback = projection
        projection = {}
      } // If not assume projection is an object and callback undefined
    }

    const cursor = this.findAsync(query, projection)

    if (typeof callback === 'function') callbackify(cursor.execAsync.bind(cursor))(callback)
    else return cursor
  }

  /**
   * Async version of {@link Datastore#find}.
   * @param {query} query MongoDB-style query
   * @param {projection} [projection = {}] MongoDB-style projection
   * @return {Cursor<document[]>}
   * @async
   */
  findAsync (query, projection = {}) {
    const cursor = new Cursor(this, query, docs => docs.map(doc => model.deepCopy(doc)), false)

    cursor.projection(projection)
    return cursor
  }

  /**
   * @callback Datastore~findOneCallback
   * @param {?Error} err
   * @param {document} doc
   */

  /**
   * Find one document matching the query.
   * @param {query} query MongoDB-style query
   * @param {projection|SingleDocumentCallback} [projection = {}] MongoDB-style projection
   * @param {SingleDocumentCallback} [callback] Optional callback, signature: err, doc
   * @return {Cursor<document>|undefined}
   */
  findOne (query, projection, callback) {
    if (arguments.length === 1) {
      projection = {}
      // callback is undefined, will return a cursor
    } else if (arguments.length === 2) {
      if (typeof projection === 'function') {
        callback = projection
        projection = {}
      } // If not assume projection is an object and callback undefined
    }

    const cursor = this.findOneAsync(query, projection)

    if (typeof callback === 'function') callbackify(cursor.execAsync.bind(cursor))(callback)
    else return cursor
  }

  /**
   * Async version of {@link Datastore#findOne}.
   * @param {query} query MongoDB-style query
   * @param {projection} projection MongoDB-style projection
   * @return {Cursor<document>}
   * @see Datastore#findOne
   */
  findOneAsync (query, projection = {}) {
    const cursor = new Cursor(this, query, docs => docs.length === 1 ? model.deepCopy(docs[0]) : null, false)

    cursor.projection(projection).limit(1)
    return cursor
  }

  /**
   * If update was an upsert, `upsert` flag is set to true, `affectedDocuments` can be one of the following:
   * - For an upsert, the upserted document
   * - For an update with returnUpdatedDocs option false, null
   * - For an update with returnUpdatedDocs true and multi false, the updated document
   * - For an update with returnUpdatedDocs true and multi true, the array of updated documents
   *
   * **WARNING:** The API was changed between v1.7.4 and v1.8, for consistency and readability reasons. Prior and
   * including to v1.7.4, the callback signature was (err, numAffected, updated) where updated was the updated document
   * in case of an upsert or the array of updated documents for an update if the returnUpdatedDocs option was true. That
   * meant that the type of affectedDocuments in a non multi update depended on whether there was an upsert or not,
   * leaving only two ways for the user to check whether an upsert had occured: checking the type of affectedDocuments
   * or running another find query on the whole dataset to check its size. Both options being ugly, the breaking change
   * was necessary.
   * @callback Datastore~updateCallback
   * @param {?Error} err
   * @param {?number} numAffected
   * @param {?document[]|?document} affectedDocuments
   * @param {?boolean} upsert
   */

  /**
   * Update all docs matching query.
   *
   * Use {@link Datastore#update} which has the same signature.
   * @param {query} query is the same kind of finding query you use with `find` and `findOne`
   * @param {document|update} update specifies how the documents should be modified. It is either a new document or a
   * set of modifiers (you cannot use both together, it doesn't make sense!):
   *  - A new document will replace the matched docs
   *  - The modifiers create the fields they need to modify if they don't exist, and you can apply them to subdocs.
   *  Available field modifiers are `$set` to change a field's value, `$unset` to delete a field, `$inc` to increment a
   *  field's value and `$min`/`$max` to change field's value, only if provided value is less/greater than current
   *  value. To work on arrays, you have `$push`, `$pop`, `$addToSet`, `$pull`, and the special `$each` and `$slice`.
   * @param {object} [options] Optional options. If not given, is interpreted as the callback.
   * @param {boolean} [options.multi = false] If true, can update multiple documents
   * @param {boolean} [options.upsert = false] If true, can insert a new document corresponding to the `update` rules if
   * your `query` doesn't match anything. If your `update` is a simple object with no modifiers, it is the inserted
   * document. In the other case, the `query` is stripped from all operator recursively, and the `update` is applied to
   * it.
   * @param {boolean} [options.returnUpdatedDocs = false] (not Mongo-DB compatible) If true and update is not an upsert,
   * will return the array of documents matched by the find query and updated. Updated documents will be returned even
   * if the update did not actually modify them.
   * @param {Datastore~updateCallback} callback
   *
   * @private
   */
  _update (query, update, options, callback) {
    const _callback = (err, res = {}) => {
      callback(err, res.numAffected, res.affectedDocuments, res.upsert)
    }
    callbackify(this._updateAsync.bind(this))(query, update, options, _callback)
  }

  /**
   * Async version of {@link Datastore#_update}.
   *
   * Use {@link Datastore#updateAsync} which has the same signature.
   * @param {query} query is the same kind of finding query you use with `find` and `findOne`
   * @param {document|update} update specifies how the documents should be modified. It is either a new document or a
   * set of modifiers (you cannot use both together, it doesn't make sense!):
   *  - A new document will replace the matched docs
   *  - The modifiers create the fields they need to modify if they don't exist, and you can apply them to subdocs.
   *  Available field modifiers are `$set` to change a field's value, `$unset` to delete a field, `$inc` to increment a
   *  field's value and `$min`/`$max` to change field's value, only if provided value is less/greater than current
   *  value. To work on arrays, you have `$push`, `$pop`, `$addToSet`, `$pull`, and the special `$each` and `$slice`.
   * @param {Object} options options
   * @param {boolean} [options.multi = false] If true, can update multiple documents
   * @param {boolean} [options.upsert = false] If true, can insert a new document corresponding to the `update` rules if
   * your `query` doesn't match anything. If your `update` is a simple object with no modifiers, it is the inserted
   * document. In the other case, the `query` is stripped from all operator recursively, and the `update` is applied to
   * it.
   * @param {boolean} [options.returnUpdatedDocs = false] (not Mongo-DB compatible) If true and update is not an upsert,
   * will return the array of documents matched by the find query and updated. Updated documents will be returned even
   * if the update did not actually modify them.
   *
   * @return {Promise<{numAffected: number, affectedDocuments: document[]|document|null, upsert: boolean}>}
   * @see Datastore#_update
   * @private
   */
  async _updateAsync (query, update, options) {
    const multi = options.multi !== undefined ? options.multi : false
    const upsert = options.upsert !== undefined ? options.upsert : false

    // If upsert option is set, check whether we need to insert the doc
    if (upsert) {
      const cursor = new Cursor(this, query, x => x, false)

      // Need to use an internal function not tied to the executor to avoid deadlock
      const docs = await cursor.limit(1)._execAsync()

      if (docs.length !== 1) {
        let toBeInserted

        try {
          model.checkObject(update)
          // updateQuery is a simple object with no modifier, use it as the document to insert
          toBeInserted = update
        } catch (e) {
          // updateQuery contains modifiers, use the find query as the base,
          // strip it from all operators and update it according to updateQuery
          toBeInserted = model.modify(model.deepCopy(query, true), update)
        }
        const newDoc = await this._insertAsync(toBeInserted)
        return { numAffected: 1, affectedDocuments: newDoc, upsert: true }
      }
    }
    // Perform the update
    let numReplaced = 0
    let modifiedDoc
    const modifications = []
    let createdAt

    const candidates = await this.getCandidatesAsync(query)
    // Preparing update (if an error is thrown here neither the datafile nor
    // the in-memory indexes are affected)
    for (const candidate of candidates) {
      if (model.match(candidate, query) && (multi || numReplaced === 0)) {
        numReplaced += 1
        if (this.timestampData) { createdAt = candidate.createdAt }
        modifiedDoc = model.modify(candidate, update)
        if (this.timestampData) {
          modifiedDoc.createdAt = createdAt
          modifiedDoc.updatedAt = new Date()
        }
        modifications.push({ oldDoc: candidate, newDoc: modifiedDoc })
      }
    }

    // Change the docs in memory
    this.updateIndexes(modifications)

    // Update the datafile
    const updatedDocs = modifications.map(x => x.newDoc)
    await this.persistence.persistNewStateAsync(updatedDocs)
    if (!options.returnUpdatedDocs) return { numAffected: numReplaced, upsert: false, affectedDocuments: null }
    else {
      let updatedDocsDC = []
      updatedDocs.forEach(doc => { updatedDocsDC.push(model.deepCopy(doc)) })
      if (!multi) updatedDocsDC = updatedDocsDC[0]
      return { numAffected: numReplaced, affectedDocuments: updatedDocsDC, upsert: false }
    }
  }

  /**
   * Update all docs matching query.
   * @param {query} query is the same kind of finding query you use with `find` and `findOne`
   * @param {document|*} update specifies how the documents should be modified. It is either a new document or a
   * set of modifiers (you cannot use both together, it doesn't make sense!):
   *  - A new document will replace the matched docs
   *  - The modifiers create the fields they need to modify if they don't exist, and you can apply them to subdocs.
   *  Available field modifiers are `$set` to change a field's value, `$unset` to delete a field, `$inc` to increment a
   *  field's value and `$min`/`$max` to change field's value, only if provided value is less/greater than current
   *  value. To work on arrays, you have `$push`, `$pop`, `$addToSet`, `$pull`, and the special `$each` and `$slice`.
   * @param {Object|Datastore~updateCallback} [options|] Optional options
   * @param {boolean} [options.multi = false] If true, can update multiple documents
   * @param {boolean} [options.upsert = false] If true, can insert a new document corresponding to the `update` rules if
   * your `query` doesn't match anything. If your `update` is a simple object with no modifiers, it is the inserted
   * document. In the other case, the `query` is stripped from all operator recursively, and the `update` is applied to
   * it.
   * @param {boolean} [options.returnUpdatedDocs = false] (not Mongo-DB compatible) If true and update is not an upsert,
   * will return the array of documents matched by the find query and updated. Updated documents will be returned even
   * if the update did not actually modify them.
   * @param {Datastore~updateCallback} [cb = () => {}] Optional callback
   *
   */
  update (query, update, options, cb) {
    if (typeof options === 'function') {
      cb = options
      options = {}
    }
    const callback = cb || (() => {})
    this.executor.push({ this: this, fn: this._update, arguments: [query, update, options, callback] })
  }

  /**
   * Async version of {@link Datastore#update}.
   * @param {query} query is the same kind of finding query you use with `find` and `findOne`
   * @param {document|*} update specifies how the documents should be modified. It is either a new document or a
   * set of modifiers (you cannot use both together, it doesn't make sense!):
   *  - A new document will replace the matched docs
   *  - The modifiers create the fields they need to modify if they don't exist, and you can apply them to subdocs.
   *  Available field modifiers are `$set` to change a field's value, `$unset` to delete a field, `$inc` to increment a
   *  field's value and `$min`/`$max` to change field's value, only if provided value is less/greater than current
   *  value. To work on arrays, you have `$push`, `$pop`, `$addToSet`, `$pull`, and the special `$each` and `$slice`.
   * @param {Object} [options = {}] Optional options
   * @param {boolean} [options.multi = false] If true, can update multiple documents
   * @param {boolean} [options.upsert = false] If true, can insert a new document corresponding to the `update` rules if
   * your `query` doesn't match anything. If your `update` is a simple object with no modifiers, it is the inserted
   * document. In the other case, the `query` is stripped from all operator recursively, and the `update` is applied to
   * it.
   * @param {boolean} [options.returnUpdatedDocs = false] (not Mongo-DB compatible) If true and update is not an upsert,
   * will return the array of documents matched by the find query and updated. Updated documents will be returned even
   * if the update did not actually modify them.
   * @async
   * @return {Promise<{numAffected: number, affectedDocuments: document[]|document|null, upsert: boolean}>}
   * @see Datastore#update
   */
  updateAsync (query, update, options = {}) {
    return this.executor.pushAsync(() => this._updateAsync(query, update, options))
  }

  /**
   * @callback Datastore~removeCallback
   * @param {?Error} err
   * @param {?number} numRemoved
   */

  /**
   * Remove all docs matching the query.
   *
   * Use {@link Datastore#remove} which has the same signature.
   *
   * For now very naive implementation (similar to update).
   * @param {query} query
   * @param {object} options options
   * @param {boolean} [options.multi = false] If true, can update multiple documents
   * @param {Datastore~removeCallback} callback
   * @see Datastore#remove
   * @private
   */
  _remove (query, options, callback) {
    callbackify(this._removeAsync.bind(this))(query, options, callback)
  }

  /**
   * Async version of {@link Datastore#_remove}.
   *
   * Use {@link Datastore#removeAsync} which has the same signature.
   * @param {query} query
   * @param {object} [options] Optional options
   * @param {boolean} [options.multi = false] If true, can update multiple documents
   * @return {Promise<number>} How many documents were removed
   * @private
   * @see Datastore#_remove
   */
  async _removeAsync (query, options = {}) {
    const multi = options.multi !== undefined ? options.multi : false

    const candidates = await this.getCandidatesAsync(query, true)
    const removedDocs = []
    let numRemoved = 0

    candidates.forEach(d => {
      if (model.match(d, query) && (multi || numRemoved === 0)) {
        numRemoved += 1
        removedDocs.push({ $$deleted: true, _id: d._id })
        this.removeFromIndexes(d)
      }
    })

    await this.persistence.persistNewStateAsync(removedDocs)
    return numRemoved
  }

  /**
   * Remove all docs matching the query.
   * @param {query} query
   * @param {object|Datastore~removeCallback} [options={}] Optional options
   * @param {boolean} [options.multi = false] If true, can update multiple documents
   * @param {Datastore~removeCallback} [cb = () => {}] Optional callback
   */
  remove (query, options, cb) {
    if (typeof options === 'function') {
      cb = options
      options = {}
    }
    const callback = cb || (() => {})
    this.executor.push({ this: this, fn: this._remove, arguments: [query, options, callback] })
  }

  /**
   * Remove all docs matching the query.
   * Use Datastore.removeAsync which has the same signature
   * @param {query} query
   * @param {object} [options={}] Optional options
   * @param {boolean} [options.multi = false] If true, can update multiple documents
   * @return {Promise<number>} How many documents were removed
   * @async
   */
  removeAsync (query, options = {}) {
    return this.executor.pushAsync(() => this._removeAsync(query, options))
  }
}

module.exports = Datastore