Fix for off-by-one when initializing a constant or function...

Fix for off-by-one when initializing a constant or function
declaration that was not a slot.

Committed: http://code.google.com/p/v8/source/detail?r=1203

[Kevin Millikin (Chromium), 2009-02-01 08:53:24]

This is a fix for issue 220.  Discarding the value of SetValue on a reference
and discarding the reference itself (which preserves top-of-stack) only commute
when the reference is zero-sized.

We should be able to construct a regression test for this, but it's tricky.

We should also double check that our behavior in this case is what we want---it
is only triggered by constant declarations and function declarations in
statement position, neither of which is ECMA-262.

[iposva, 2009-02-02 07:12:53]

http://codereview.chromium.org/19745/diff/1/2
File src/codegen-arm.cc (right):

http://codereview.chromium.org/19745/diff/1/2#newcode1154
Line 1154: frame_->Pop();
What is it that we are popping here? If it is the value that has been loaded
with "Load(val)" then aren't you doing things in the wrong order? In any case
this code is so tricky that added comments are definitely going to be a plus
when revisiting.

[Kevin Millikin (Chromium), 2009-02-02 10:12:40]

Added regression test.  Take a quick look, please.

[iposva, 2009-02-02 17:11:10]

LGTM as well, especially once a clarifying comment is put in.

-Ivan