Lazily initialize the target map list in IC.

Lazily initialize the target map list in IC.

This makes sure that we keep the target map list alive in the handle scope during IC computation.

BUG=
R=verwaest@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=20426

[ulan, 2014-04-01 13:44:05]

PTAL. This is needed for weak ICs.

[Toon Verwaest, 2014-04-01 15:22:05]

https://codereview.chromium.org/220923003/diff/1/src/ic.cc
File src/ic.cc (right):

https://codereview.chromium.org/220923003/diff/1/src/ic.cc#newcode1426
src/ic.cc:1426: if (target_receiver_maps.length() == 0) {
length() == 0 isn't the same as UNINITIALIZED || PREMONOMORPHIC. We could also
have length() == 0 if for some reason there's an instance type check rather than
a map-check. With your change we'd go monomorphic where otherwise we would've
gone generic. (See the comment below that you removed).

I presume we'd actually need to introduce something similar for load, rather
than remove it for the store.

[ulan, 2014-04-01 16:14:07]

https://codereview.chromium.org/220923003/diff/1/src/ic.cc
File src/ic.cc (right):

https://codereview.chromium.org/220923003/diff/1/src/ic.cc#newcode1426
src/ic.cc:1426: if (target_receiver_maps.length() == 0) {
On 2014/04/01 15:22:05, Toon Verwaest wrote:
> length() == 0 isn't the same as UNINITIALIZED || PREMONOMORPHIC. We could also
> have length() == 0 if for some reason there's an instance type check rather
than
> a map-check. With your change we'd go monomorphic where otherwise we would've
> gone generic. (See the comment below that you removed).
> 
> I presume we'd actually need to introduce something similar for load, rather
> than remove it for the store.
The comment that I removed seems to be out-dated: keyed stores in dictionary
mode use KeyedStoreIC::GenerateSlow(masm) which tail calls to runtime function
KeyedStoreIC_Slow, so can never trigger IC miss. I also ran tests with
UNREACHABLE in the removed branch, and it never triggered.

Keyed loads in dictionary mode are guarded by map checks.

[Toon Verwaest, 2014-04-01 17:11:31]

lgtm

[ulan, 2014-04-02 08:00:00]

Committed patchset #1 manually as r20426 (presubmit successful).