Why is `(foo) = “bar”` legal in JavaScript?

Question

In Node.js's REPL (tested in SpiderMonkey as well) the sequence

var foo = null;
(foo) = "bar";

is valid, with foo subsequently equal to "bar" as opposed to null.

This seems counterintuitive because one would think the parenthesis would at least dereference bar and throw Invalid left-hand side in assignment`.

Understandably, when you do anything interesting it does fail in aforementioned way.

(foo, bar) = 4
(true ? bar : foo) = 4

According to ECMA-262 on LeftHandExpressions (so far as I can interpret) are no valid non-terminals that would lead to a parenthetical being accepted.

Is there something I'm not seeing?

Answer 1

It's valid indeed. You're allowed to wrap any simple assignment target in parenthesis.

The left hand part of the = operation is a LeftHandSideExpression as you correctly identified. This can be tracked down through the various precendence levels (NewExpression, MemberExpression) to a PrimaryExpression, which in turn might be a Cover­Parenthesized­Expression­And­Arrow­Parameter­List:

( Expression[In, ?Yield])

(actually, when parsed with target PrimaryExpression, it's a ParenthesizedExpression).

So it's valid by the grammar, at least. Whether it's actually valid JS syntax is determined by another factor: early error static semantics. Those are basically prose or algorithmic rules that make some production expansions invalid (syntax errors) in certain cases. This for example allowed the authors to reuse the array and object initialiser grammars for destructuring, but only applying certain rules. In the early errors for assignment expressions we find

It is an early Reference Error if LeftHandSideExpression is neither an ObjectLiteral nor an ArrayLiteral and IsValidSimpleAssignmentTarget of LeftHandSideExpression is false.

We can also see this distinction in the evaluation of assignment expressions, where simple assignment targets are evaluated to a reference that can be assigned to, instead of getting the destructuring pattern stuff like object and array literals.

So what does that IsValidSimpleAssignmentTarget do to LeftHandSideExpressions? Basically it allows assignments to property accesses and disallows assignments to call expressions. It doesn't state anything about plain PrimaryExpressions which have their own IsValidSimpleAssignmentTarget rule. All it does is to extract the Expression between the parentheses through the Covered­Parenthesized­Expression operation, and then again check IsValidSimpleAssignmentTarget of that. In short: (…) = … is valid when … = … is valid. It'll yield true only for Identifiers (like in your example) and properties.

Answer 2

As per @dlatikay's suggestion, following an existing hunch, research into CoveredParenthesizedExpression yielded a better understanding of what's happening here.

Apparently, the reason why a non-terminal cannot be found in the spec, to explain why (foo) is acceptable as a LeftHandExpression, is surprisingly simple. I'm assuming you understand how parsers work, and that they operate in two separate stages: Lexing and Parsing.

What I've learned from this little research tangent is that the construct (foo) is not technically being delivered to parser, and therefor the engine, as you might think.

Consider the following

var foo = (((bar)));

As we all know, something like this is perfectly legal. Why? Well you visually can just ignore the parenthesis while the statement remains making perfect sense.

Similarly, here is another valid example, even from a human readability perspective, because the parentheses only explicate what PEMDAS already makes implicit.

(3 + ((4 * 5) / 2)) === 3 + 4 * 5 / 2
>> true

One key observation can be loosely derived from this, given an understanding of how parsers already work. (remember, Javascript still is being parsed (read: compiled) and then run) So in a direct sense, these parentheses are "stating the obvious".

So all that being said, what exactly is going on?

Basically, parentheses (with the exception of function parameters) are collapsed into proper groupings of their containing symbols. IANAL but, in lay man's terms, that means that parentheses are only interpreted to guide the parser how to group what it reads. If the context of the parentheses is already "in order", and thus does not require any tweaking of the emitted AST, then the (machine) code is emitted as if those parentheses did not exist at all.

The parser is more or less being lazy, assuming the parens are impertinent. (which in this edge-case, is not true)

Okay, and where exactly is this happening?

According to 12.2.1.5 Static Semantics: IsValidSimpleAssignmentTarget in the spec,

PrimaryExpression: (CoverParenthesizedExpressionAndArrowParameterList)

1. Let expr be CoveredParenthesizedExpression of CoverParenthesizedExpressionAndArrowParameterList.
2. Return IsValidSimpleAssignmentTarget of expr.

I.E. If expecting primaryExpression return whatever is inside the parenthesis and use that.

Because of that, in this scenario, it does not convert (foo) into CoveredParenthesizedExpression{ inner: "foo" }, it converts it simply into foo which preserves the fact that it is an Identifier and thus syntactically, while not necessarily lexically, valid.

TL;DR

It's wat.

Want a little more insight?

Check out @Bergi's answer.