This proposal is officially in stage 4 of the TC39 process, and is in the process of being integrated into the spec.
This proposal was formerly for Array.prototype.contains
, but that name is not web-compatible. Per the November 2014 TC39 meeting, the name of both String.prototype.contains
and Array.prototype.contains
was changed to includes
to dodge that bullet.
When using ECMAScript arrays, it is commonly desired to determine if the array includes an element. The prevailing pattern for this is
if (arr.indexOf(el) !== -1) {
...
}
with various other possibilities, e.g. arr.indexOf(el) >= 0
, or even ~arr.indexOf(el)
.
These patterns exhibit two problems:
- They fail to "say what you mean": instead of asking about whether the array includes an element, you ask what the index of the first occurrence of that element in the array is, and then compare it or bit-twiddle it, to determine the answer to your actual question.
- They fail for
NaN
, asindexOf
uses Strict Equality Comparison and thus[NaN].indexOf(NaN) === -1
.
We propose the addition of an Array.prototype.includes
method, such that the above patterns can be rewritten as
if (arr.includes(el)) {
...
}
This has almost the same semantics as the above, except that it uses the SameValueZero comparison algorithm instead of Strict Equality Comparison, thus making [NaN].includes(NaN)
true.
Thus, this proposal solves both problems seen in existing code.
We additionally add a fromIndex
parameter, similar to Array.prototype.indexOf
and String.prototype.includes
, for consistency.
If you survey existing APIs, has
is used for conceptual "keys," whereas includes
is used for conceptual "values." That is:
- Keys inside a key-value map:
Map.prototype.has(key)
,WeakMap.prototype.has(key)
,Reflect.has(target, propertyKey)
- Sets, whose elements are conceptually both keys and values:
Set.prototype.has(value)
,WeakSet.prototype.has(value)
,Reflect.Loader.prototype.has(name)
- Strings, which are conceptually maps from indices to code points:
String.prototype.includes(searchString, position)
The best consistency here is with String
, not with Map
or Set
.
The web has classes like DOMStringList and DOMTokenList which are array-like, and have methods named contains
with the same semantics as our includes
. Unfortunately, meshing with those is not web-compatible, as explained above; we will have to accept this inconsistency.
Yes, that's true. The best way to think about this is that String.prototype.indexOf
and String.prototype.includes
behave like their Array.prototype
counterparts in the special case of a single character. But the string versions can also be used in the more general case of a larger string.
So in this way, the relationship between String.prototype.includes
and Array.prototype.includes
is the same as the relationship between String.prototype.indexOf
and Array.prototype.indexOf
.
There are four equality algorithms in the current ES6 draft:
- Abstract Equality Comparison (
==
) - Strict Equality Comparison (
===
): used byArray.prototype.indexOf
,Array.prototype.lastIndexOf
, andcase
-matching - SameValueZero: used by
%TypedArray%
andArrayBuffer
constructors, as well asMap
andSet
operations - SameValue: used in all other places
(Note however that most places SameValue is used could be replaced by SameValueZero since those places often never compare primitives, or at least never compare numbers.)
Using Abstract Equality Comparison would be bonkers, of course. Using SameValue is not a good idea for the same reasons it is not used by Map
and Set
. (Briefly: -0
s can sneak into your code fairly easily via arithmetic operations, but you almost always desire -0
to be treated the same as +0
, so distinguishing them will just cause spurious failures.) This leaves Strict Equality Comparison and SameValueZero as the two possibilities.
SameValueZero is generally the better choice, as it allows you to detect if an array includes a NaN
. The argument for Strict Equality Comparison boils down to "bug compatibility" with Array.prototype.indexOf
. But one of the purposes of Array.prototype.includes
is to steer users away from creating these sort of bugs.
This introduces a slight refactoring hazard from Array.prototype.indexOf
to Array.prototype.includes
: they will indeed behave differently for arrays containing NaN
s. However, it seems much more likely that code will become less buggy via this refactoring, instead of causing problems. Introducing a new method, and accompanying it with the appropriate messaging around this case, should help.
As with all non-mutating array methods, we also install this method on %TypedArray%.prototype
.
assert([1, 2, 3].includes(2) === true);
assert([1, 2, 3].includes(4) === false);
assert([1, 2, NaN].includes(NaN) === true);
assert([1, 2, -0].includes(+0) === true);
assert([1, 2, +0].includes(-0) === true);
assert(["a", "b", "c"].includes("a") === true);
assert(["a", "b", "c"].includes("a", 1) === false);