Navigation

$replaceAll (aggregation)

Definition

$replaceAll

New in version 4.4.

Replaces all instances of a search string in an input string with a replacement string.

$replaceAll is both case-sensitive and diacritic-sensitive, and ignores any collation present on a collection.

Syntax

The $replaceAll operator has the following operator expression syntax:

{ $replaceAll: { input: <expression>, find: <expression>, replacement: <expression> } }

Operator Fields

Field Description
input

The string on which you wish to apply the find. Can be any valid expression that resolves to a string or a null. If input refers to a field that is missing, $replaceAll returns null.

find

The string to search for within the given input. Can be any valid expression that resolves to a string or a null. If find refers to a field that is missing, $replaceAll returns null.

replacement

The string to use to replace all matched instances of find in input. Can be any valid expression that resolves to a string or a null.

Behavior

The input, find, and replacement expressions must evaluate to a string or a null, or $replaceAll fails with an error.

$replaceAll and Null Values

If input or find refer to a field that is missing, they return null.

If any one of input, find, or replacement evaluates to a null, the entire $replaceAll expression evaluates to null:

Example Result
{ $replaceAll: { input: null, find: "abc", replacement: "ABC" } } null
{ $replaceAll: { input: "abc", find: null, replacement: "ABC" } } null
{ $replaceAll: { input: "abc", find: "abc", replacement: null } } null

$replaceAll and Collation

String matching for all $replaceAll expressions is always case-sensitive and diacritic-sensitive. Any collation configured on a collection, db.collection.aggregate(), or index is ignored when performing string comparisons with $replaceAll.

For example, create a sample collection with collation strength 1:

db.createCollection( "myColl", { collation: { locale: "fr", strength: 1 } } )

A collation strength of 1 compares base character only and ignores other differences such as case and diacritics.

Next, insert three example documents:

db.myColl.insertMany([
   { _id: 1, name: "cafe" },
   { _id: 2, name: "Cafe" },
   { _id: 3, name: "café" }
])

The following $replaceAll operation tries to find and replace all instances of “Cafe” in the name field:

db.myColl.aggregate([
  {
    $addFields:
      {
        resultObject: { $replaceAll: { input: "$name", find: "Cafe", replacement: "CAFE" } }
      }
  }
])

Because $replaceAll ignores the collation configured for this collection, the operation only matches the instance of “Cafe” in document 2:

{ "_id" : 1, "name" : "cafe", "resultObject" : "cafe" }
{ "_id" : 2, "name" : "Cafe", "resultObject" : "CAFE" }
{ "_id" : 3, "name" : "café", "resultObject" : "café" }

Operators which respect collation, such as $match, would match all three documents when performing a string comparison against “Cafe” due to this collection’s collation strength of 1.

$replaceAll and Unicode Normalization

The $replaceAll aggregation expression does not perform any unicode normalization. This means that string matching for all $replaceAll expressions will consider the number of code points used to represent a character in unicode when attempting a match.

For example, the character é can be represented in unicode using either one code point or two:

Unicode Displays as Code points
\xe9 é 1 ( \xe9 )
e\u0301 é 2 ( e + \u0301 )

Using $replaceAll with a find string where the character é is represented in unicode with one code point will not match any instance of é that uses two code points in the input string.

The following table shows whether a match occurs for a find string of “café” when compared to input strings where é is represented by either one code point or two. The find string in this example uses one code point to represent the é character:

Example Match
{ $replaceAll: { input: "caf\xe9", find: "café", replacement: "CAFE" } } yes
{ $replaceAll: { input: "cafe\u0301", find: "café", replacement: "CAFE" } } no

Because $replaceAll does not perform any unicode normalization, only the first string comparison matches, where both the find and input strings use one code point to represent é.

Example

Create an inventory collection with the following documents:

db.inventory.insertMany([
   { "_id" : 1, "item" : "blue paint" },
   { "_id" : 2, "item" : "blue and green paint" },
   { "_id" : 3, "item" : "blue paint with blue paintbrush" },
   { "_id" : 4, "item" : "blue paint with green paintbrush" },
])

The following example replaces each instance of “blue paint” in the item field with “red paint”:

db.inventory.aggregate([
   {
     $project:
      {
         item: { $replaceAll: { input: "$item", find: "blue paint", replacement: "red paint" } }
      }
   }
])

The operation returns the following results:

{ "_id" : 1, "item" : "red paint" }
{ "_id" : 2, "item" : "blue and green paint" }
{ "_id" : 3, "item" : "red paint with red paintbrush" }
{ "_id" : 4, "item" : "red paint with green paintbrush" }