[turbofan] Weakening of types must handle ranges inside unions.

[turbofan] Weakening of types must weaken ranges inside unions.

BUG=
R=rossberg@chromium.org

Committed: https://chromium.googlesource.com/v8/v8/+/4c1f4b796d1c455fc6a023abe145a5e48c4b7b1f

[Jarin, 2014-11-07 15:56:18]

jarin@chromium.org changed reviewers:
+ rossberg@chromium.org

[Jarin, 2014-11-07 16:00:46]

Could you take a look at the fix?

I am somewhat reluctant to check in something like this because I am not
entirely sure what Type::GetRange actually means and what properties we require
for weakening to work.

Any thoughts?

(FYI, the test case was found by ClusterFuzz, so it is not particularly
meaningful.)

[rossberg, 2014-11-10 13:46:11]

It looks okayish to me, except that I suppose there should be some minimal tests
for GetRange now.

https://codereview.chromium.org/712623002/diff/20001/src/types.cc
File src/types.cc (right):

https://codereview.chromium.org/712623002/diff/20001/src/types.cc#newcode1024
src/types.cc:1024: os << "Range(" << std::fixed << std::setprecision(0)
Why precision 0? And why has to be set twice?

[Jarin, 2014-11-12 14:26:48]

On 2014/11/10 13:46:11, rossberg wrote:
> It looks okayish to me, except that I suppose there should be some minimal
tests
> for GetRange now.

Minimal tests added.

> https://codereview.chromium.org/712623002/diff/20001/src/types.cc
> File src/types.cc (right):
> 
> https://codereview.chromium.org/712623002/diff/20001/src/types.cc#newcode1024
> src/types.cc:1024: os << "Range(" << std::fixed << std::setprecision(0)
> Why precision 0? And why has to be set twice?

Done. It is set just once now. I have changed formatting so that the ranges are
printed as integers (0 decimal places precision). We should now get 1000000000
instead of 1e+9 or 1000000000.00.

[rossberg, 2014-11-12 14:51:15]

LGTM, modulo nit.

https://codereview.chromium.org/712623002/diff/40001/test/cctest/test-types.cc
File test/cctest/test-types.cc (right):

https://codereview.chromium.org/712623002/diff/40001/test/cctest/test-types.c...
test/cctest/test-types.cc:1834: TypeHandle RangeToHandle(typename
Type::RangeType* range) {
That seems a bit hacky. Can we instead have a proper generic handlify function
inherited from the respective Rep classes?

[rossberg, 2014-11-12 14:51:15]

LGTM, modulo nit.

https://codereview.chromium.org/712623002/diff/40001/test/cctest/test-types.cc
File test/cctest/test-types.cc (right):

https://codereview.chromium.org/712623002/diff/40001/test/cctest/test-types.c...
test/cctest/test-types.cc:1834: TypeHandle RangeToHandle(typename
Type::RangeType* range) {
That seems a bit hacky. Can we instead have a proper generic handlify function
inherited from the respective Rep classes?

[Jarin, 2014-11-13 05:31:57]

Committed patchset #3 (id:40001) manually as
4c1f4b796d1c455fc6a023abe145a5e48c4b7b1f (presubmit successful).