Refactor RangeBoundary +/- infinity to be distinct RangeBoundary kinds.

- Refactor RangeBoundary +/- infinity to be distinct RangeBoundary kinds.
- Add tests for RangeBoundary infinities.

R=vegorov@google.com

Committed: https://code.google.com/p/dart/source/detail?r=37124

[Vyacheslav Egorov (Google), 2014-06-08 06:58:22]

lgtm

[Vyacheslav Egorov (Google), 2014-06-09 07:43:16]

One more comment (for the future, not for this CL)

https://codereview.chromium.org/321593004/diff/1/runtime/vm/intermediate_lang...
File runtime/vm/intermediate_language_test.cc (right):

https://codereview.chromium.org/321593004/diff/1/runtime/vm/intermediate_lang...
runtime/vm/intermediate_language_test.cc:68: TEST_CASE(RangeTestsInfinity) {
I was thinking that maybe we should move all arithmetic (binary) operations onto
the Range itself (leaving only interpretations of the results in the
BinaryOpInstr). This would also allow us to test that things like 

(0, +Infty) + (-Infty, 0) = (-Infty, +Infty)

and

((0, +Infty) + (0, +Infty))->IsPositive()

etc

This is something for the future CLs.

It would be especially important to have this tests when we start distinguishing
Smi and x64 ranges, because addition of two smis always produces a smi +
overflow flag so when converted to mint we should see the right range there not
(-Infty, +Infty)

[Cutch, 2014-06-09 17:58:38]

Committed patchset #2 manually as r37124 (presubmit successful).

[Florian Schneider, 2014-06-10 10:43:20]

https://codereview.chromium.org/321593004/diff/20001/runtime/vm/intermediate_...
File runtime/vm/intermediate_language.h (right):

https://codereview.chromium.org/321593004/diff/20001/runtime/vm/intermediate_...
runtime/vm/intermediate_language.h:2645: bool IsPositive() const;
I find IsPositive and IsNegative more confusing than explicitly using IsWithin,
and Overlaps:

IsPositive does not imply !IsNegative, and vice versa. There was a bug because
of that.