Tear-off: expression is not evaluated to object

[sgrekhov]

According Dart spec "16.18.3 General Closurization"

"Evaluation of a property extraction i of the form e#m proceeds as follows:
First, the expression e is evaluated to an object o. Then:..."

But in fact expression is not evaluated to object o. Run the following program

class A {
  int get g => 1;
}

class C {
  A get m => new A();
}

main() { 
  C c = new C();
  var x = c.m#g;
}

Result: error: semicolon expected
var x = c.m#g;

var x = (c.m)#g works well.

Tested on Dart VM version: 1.13.0-dev.2.0 (Tue Sep 8 05:28:05 2015) on "linux_x64"

[lrhn]

The behavior is what the current grammar specifies:

postfixExpression:
  assignableExpression postfixOperator |
  primary (selector* | ( ‘#’ ( (identifier ‘=’?) | operator)))
  ;

This means that you can't put a "#foo" after a selector like .m - you can do either, but not both. It does work when you wrap it in parentheses because a parenthesized expression is a primary.

[sgrekhov]

Isn't it a contradiction with

Evaluation of a property extraction i of the form e#m proceeds as follows:
First, the expression e is evaluated to an object o.?

According this statement getter invocation should be evaluated to the object which is value of the getter

[kmillikin]

It's not a contradiction, it just doesn't stand on its own. You have to read that section along with the grammar as @lrhn did above.

Consider the analogous: Evaluation of a multiplication expression of the form e0 * e1 proceeds as follows: first the expression e0 is evaluated, then the expression e1 is evaluated, then the result is multiplied.

Now, would you expect x * y + z to first evaluate x, then evaluate y + z, then multiply the results? That's not what actually happens, because if the precedence rules encoded in the grammar.

Edit: to be clear, c.m#g is not a property extraction of the form e#m because it can't be parsed that way. (c.m)#g is a property extraction of the form e#m because that's the way it is parsed.

[iposva-google]

Over to Gilad to validate that the VM's approach is correct here.

[sgrekhov]

What about this?

class A {
  int m() => 1;
}

A getA() {
  return new A();
}

main() {
  var f = getA()#m;
}

Result:
error: line 24 pos 17: semicolon expected
var f = getA()#m;
^

[lrhn]

getA()#m is still not valid syntax. The grammar allows either a tear-off operation or a "selector" chain (call or member access, basically) after a primary expression. Here you do getA() which is not a primary expression (getA is a primary expression, getA() has the call selector after it so you can't also have a tear-off). Again you need to write (getA())#m if that is what you want to do.

[sgrekhov]

Ok, thank you for clarification. But I think this grammar is, at least, not obvious at this point and the above is not what Dart user,who is not deeply familiar with the specification and sophisticated syntax like 'primary (selector* | ( ‘#’ ( (identifier ‘=’?) | operator)))', expects.

getA().m works, getA()#m not. Why? Why this limited (limited comparing with '.') syntax is needed at all?

[munificent]

Generalized tear-offs are being removed (#27518) so I don't think we have to worry about this.