Spreading undefined in array vs object
Why does spreading undefined in an object return an empty object? {...undefined} // equals {}
:
console.log({...undefined})
And Why does spreading undefined in an array give you an error?
[...undefined] // type error
:
console.log([...undefined])
Solution 1:
As noted in the comments, and summarized by @ftor from #687, object spread is equivalent1 to Object.assign() (issues #687, #45), whereas spread in array literal context is iterable spread.
Quoting Ecma-262 6.0, Object.assign() is defined as:
19.1.2.1 Object.assign ( target, ...sources )
The assign function is used to copy the values of all of the enumerable own properties from one or more source objects to a target object. When the assign function is called, the following steps are taken:
- Let to be ToObject(target).
- ReturnIfAbrupt(to).
- If only one argument was passed, return to.
- Let sources be the List of argument values starting with the second argument.
- For each element nextSource of sources, in ascending index order, do
- If nextSource is undefined or null, let keys be an empty List.
- Else, ...
...followed by the description of copying own properties. The draft of Object Rest/Spread Properties is here. It is not a part of the Ecma-262 6.0.
A SpreadElement in an array literal expression is defined to begin as follows:
SpreadElement : ... AssignmentExpression
- Let spreadRef be the result of evaluating AssignmentExpression.
- Let spreadObj be GetValue(spreadRef).
- Let iterator be GetIterator(spreadObj).
- ReturnIfAbrupt(iterator).
And since undefined
does not have a property with the key @@iterator, a TypeError is thrown, based on the steps of GetIterator. The standard is not an easy read, but if I'm not mistaken, the path to error is GetIterator -> GetMethod -> GetV -> ToObject, which throws a TypeError for undefined and null.
A simple remedy to using variables with possibly undefined value in array initialization is to use a default:
const maybeArray = undefined;
const newArray = [ ...(maybeArray || []) ];
1: There is a difference in how setters are handled.
Solution 2:
Normally, uses of ...x
requires x
to be iterable because the point of ...
is normally to flatten an iterable into its components.
However, {...x}
requires x
to be enumerable because we don't just need values, but keys along with them.
null
is not iterable. It has no components, so it doesn't make sense to iterate over null
. (for (const item of null)
will similarly fail.)
However, null
is enumerable. null
is sometime treated as an object, this is one of those cases, and objects are enumerable because they can have properties. (for (const prop in null)
will similarly succeed.)