自动分号插入：NodeJS与浏览器之间的替代转换行为？

Question

I have noticed different behavior in NodeJS vs. Browser while screwing around with automatic semicolon insertion and type casting.

Browser:

> {}+{}
NaN
> {}+{};
NaN
> ({}+{})
"[object Object][object Object]"
> ({}+{});
"[object Object][object Object]"

NodeJS:

> {}+{}
'[object Object][object Object]'
> {}+{};
NaN
> ({}+{})
'[object Object][object Object]'
> ({}+{});
'[object Object][object Object]'

A. Why is casting interpreted differently with/without a semicolon or in parenthesis?

B. Which is more compliant to the standard? Or is this not addressed in the standard?

UPDATE: I found that it only does this different behavior in NodeJS. I previously thought this was V8 vs SpiderMonkey.

Answer 1

I'm going to guess that you're entering this into different debug consoles. One debug console treats that as a Program , and the other as an Expression .

---

{}+{}

When interpreted as a Program this is equivalent to

{
  // empty block
}  // No semicolon inserted because statement is a block.
(+ Number({}))  // Prefix + operator coerces its argument to a number.

When interpreted as an Expression this is equivalent to

// Infix + operator concatenates if arguments are not both numeric
String({}) + String({});

which coerces each object to a string and concatenates those two strings.

---

Which interpreter is more compliant to the standard? Or is this not addressed in the standard?

If, as I assume, this occurs in a debug console, then no standard addresses what a debug console has to do -- it can bypass the JavaScript interpreter entirely and interpret its input as COBOL.

If you pass it into eval , the first interpretation must win because the argument to eval is a string that is parsed as a Program .