Establish distributivity for type union & intersection

Establish distributivity for type union & intersection

This requires introducing proper bounds on all leaf types, so that intersection between bitsets and these types can be accurately represented. Extending a union also becomes more involved.

(On the upside, the modified union/intersect algorithm would now allow support for proper variance for function types.)

Not sure if it is worth landing this. Distributivity isn't really a crucial property for our use cases. It seems fine if intersection is slightly lossy.

R=bmeurer@chromium.org, jarin@chromium.org
BUG=

Committed: https://code.google.com/p/v8/source/detail?r=21528

[rossberg, 2014-05-22 13:58:24]

Benedikt, wanna have a look?

[Benedikt Meurer, 2014-05-23 05:08:52]

I like this change, although it's a pretty big CL. Just two comments, otherwise
LGTM.

[Benedikt Meurer, 2014-05-23 05:08:59]

https://codereview.chromium.org/251753005/diff/20001/src/types.cc
File src/types.cc (right):

https://codereview.chromium.org/251753005/diff/20001/src/types.cc#newcode468
src/types.cc:468: ASSERT(this->Length() >= 2);
This function always returns true, so the return type should be changed to void
or the ASSERTs should be changed to return false.

https://codereview.chromium.org/251753005/diff/20001/src/types.h
File src/types.h (right):

https://codereview.chromium.org/251753005/diff/20001/src/types.h#newcode374
src/types.h:374: TypeImpl* unhandle() { return this; }
Nit: Can we come up with a better name? "unhandle" is horrible.

[rossberg, 2014-05-26 12:46:30]

PTAL. I fixed one remaining case of lossiness: the glb should retain
representation information. Needed some follow-up changes in the union/intersect
algorithms.

I also renamed Get/SetHeap to Get/SetValue.

[Benedikt Meurer, 2014-05-27 09:29:47]

Still LGTM

https://codereview.chromium.org/251753005/diff/40001/test/cctest/test-types.cc
File test/cctest/test-types.cc (right):

https://codereview.chromium.org/251753005/diff/40001/test/cctest/test-types.c...
test/cctest/test-types.cc:30: #define private public  /* To test private methods
:) */
Urghs... how about friends?

[rossberg1, 2014-05-27 09:42:58]

https://codereview.chromium.org/251753005/diff/40001/test/cctest/test-types.cc
File test/cctest/test-types.cc (right):

https://codereview.chromium.org/251753005/diff/40001/test/cctest/test-types.c...
test/cctest/test-types.cc:30: #define private public  /* To test private methods
:) */
On 2014/05/27 09:29:47, Benedikt Meurer wrote:
> Urghs... how about friends?

Because I don't want to pollute the code proper with friend declarations for
tests (which I would need to repeat in several classes).

[rossberg, 2014-05-27 13:52:39]

Committed patchset #3 manually as r21528 (presubmit successful).