Surprising negative results from 32-bit bit ops when compiling to JS

[madsager]

main() {
  var a = [ 0x80 ];
  print(a[0] << 24);
  print(0x80 << 24);
}

yields:

-2147483648
2147483648

The inconsistency is confusing.

Maybe we could generate JS code that yields positive numbers for 32-bit bitops?

[sethladd]

Marked this as blocking #210.

This surprising result is per "spec", i.e., dart2js and VM differ how left shift and also modulo/remainder are implemented.
I never liked that difference, but heard that it would be slow to implement "correctly" in dart2js. Maybe we can revisit this.

---

cc @floitschG.

[floitschG]

Modulo/remainder should behave correctly. If not, it's a bug.
I just uploaded a CL that makes all bit-operations positive: https://chromiumcodereview.appspot.com/10343002/
This is still not spec-compliant, but should be less surprising.

---

Added Started label.

[floitschG]

Closing as fixed (with r7278).
Bit operations still don't follow the spec but at least they shouldn't be surprising anymore.

---

Added Fixed label.

[rakudrama]

The surprise for me is not the sign, but that the values differ.

hide(x) => [x][0];
main() {
  var c = 0x81;
  var x = hide(0x81);
  print(c << 30);
  print(x << 30);
  print(0x81 << 30);
}

prints:

138512695296
1073741824
138512695296

Given that the issue is inconsistency between constant folding and evaluation, what was the reasoning behind the decision to change the evaluation (to something less efficient?)
Is there a use case where making the faithful domain [0,2^32) works better than [-2^31,2^31) ?

[floitschG]

The compile-time and runtime differences bug is tracked in issue #2887.

[rakudrama]

I still get negative results in some cases:

main() {
  int y = ([main][0] == main) ? 0 : 1;
  print( (-1) | y);
}
prints: -1

The generated code appears to have a bug with operator precedence.

  $.print(-1 | y >>> 0);

---

Added Triaged label.

[floitschG]

Issue #2908 has been merged into this issue.

[kasperl]

It seems like this needs a bit more work to get the precedence right (comment 7).

---

Set owner to @floitschG.
Added Accepted label.

[floitschG]

Thanks Stephen.
I have uploaded a CL for the precedence bug.
https://chromiumcodereview.appspot.com/10378047/

[floitschG]

Committed with r7414.

---

Added Fixed label.