switch == invocation

[DartBot]

This issue was originally filed by kresten.kra...@gmail.com

---

When doing

   switch(v) { case e: ... }

The spec says that it tests v==e,

  the vm does e.operator ==(v)
  the frog compiles it to JS: "switch(v) { case e: ... }"

Neither of which correspond to the spec. Which is right?

Attached test case tests the spec semantics.

---

Attachment:
SwitchEqualsTest.dart (826 Bytes)

[floitschG]

I'll let Gilad be the final judge on this, but our intention was e == v (i.e. the VM behavior). This gives more predictable behavior and can be optimized more easily.

---

Added Area-Language, Triaged labels.

[DartBot]

This comment was originally written by kresten...@gmail.com

---

For what it is worth, It seems that v==e would have (potentially) better performance, because the virtual method '==' can be extracted once, and then applied repeatedly without the vtable indirection.

[floitschG]

But the type of e is almost always known at compilation time (in most cases it's a literal).
For example:
switch (v) {
  case 1: ...
  case 2: ...
  case 3: ...
  ...
}

This could be efficiently compiled to a jump-table or a tree after first checking that the input is a number. (In the case of a double a little bit more trickery is necessary).
If the ==operator is invoked on v, then the compiler doesn't have any information and must run through each case.

[gbracha]

Fixed in spec version 0.07 to match VM behavior, which was desired intent.

---

Set owner to @gbracha.
Added Done label.