Object rest and function scope

[magic-akari]

Playground

https://esbuild.github.io/try/#dAAwLjIwLjIALS10YXJnZXQ9Y2hyb21lNTkAZnVuY3Rpb24gYSgpIHsKICBjb25zb2xlLmxvZygiYSIpOwogIHJldHVybiAiYSI7Cn0KCmZ1bmN0aW9uIGIoKSB7CiAgY29uc29sZS5sb2coImIiKTsKICByZXR1cm4gImIiOwp9CgpmdW5jdGlvbiBjKCkgewogIGNvbnNvbGUubG9nKCJjIik7CiAgcmV0dXJuICJjIjsKfQoKZnVuY3Rpb24gZm9vKHsgYWEsIC4uLmJiIH0sIGNjID0gYWEoKSkgewogIGNvbnNvbGUubG9nKGFhLCBiYiwgY2MpCn0KCgoKZm9vKHsgYWE6IGEsIGJiOiBiLCBjYzogYyB9KQ

Input

function a() {
  console.log("a");
  return "a";
}

function b() {
  console.log("b");
  return "b";
}

function c() {
  console.log("c");
  return "c";
}

function foo({ aa, ...bb }, cc = aa()) {
  console.log(aa, bb, cc)
}



foo({ aa: a, bb: b, cc: c })

with --target=chrome59

Expected Result

print some function and objects.

Actual Result

Uncaught ReferenceError: aa is not defined

Relates issue

babel/babel#16470

[evanw]

This comment is relevant:

esbuild/internal/js_parser/js_parser_lower.go

Lines 284 to 313 in 51f67d4

	// Lower each argument individually instead of lowering all arguments
	// together. There is a correctness tradeoff here around default values
	// for function arguments, with no right answer.
	//
	// Lowering all arguments together will preserve the order of side effects
	// for default values, but will mess up their scope:
	//
	// // Side effect order: a(), b(), c()
	// function foo([{[a()]: w, ...x}, y = b()], z = c()) {}
	//
	// // Side effect order is correct but scope is wrong
	// function foo(_a, _b) {
	// var [[{[a()]: w, ...x}, y = b()], z = c()] = [_a, _b]
	// }
	//
	// Lowering each argument individually will preserve the scope for default
	// values that don't contain object rest binding patterns, but will mess up
	// the side effect order:
	//
	// // Side effect order: a(), b(), c()
	// function foo([{[a()]: w, ...x}, y = b()], z = c()) {}
	//
	// // Side effect order is wrong but scope for c() is correct
	// function foo(_a, z = c()) {
	// var [{[a()]: w, ...x}, y = b()] = _a
	// }
	//
	// This transform chooses to lower each argument individually with the
	// thinking that perhaps scope matters more in real-world code than side
	// effect order.

[magic-akari]

esbuild considers scope to be more important than side effect order.
However, in this example, the reference to aa is still incorrect.

[magic-akari]

Thank you for the clear comments in esbuild.
This issue was actually discovered when I was reading the esbuild source code.
However, I feel that it might not be worth solving, or it might be unsolvable. I'm just bringing it up here for the record.

[evanw]

Yes, makes sense. Thanks for the report. When I get the time to think about this, I'll have to load this back in my head to figure out what to do about it, if anything. That comment (and any relevant surrounding commit logs from then) will be a good starting point for me.