Experimental: fix a frame-height issue with eval on ARM.

Experimental: fix a frame-height issue with eval on ARM.  Also, added
frame-height assertions for ARM to the code generator's Visit member
functions.

Committed: http://code.google.com/p/v8/source/detail?r=1328

[Kevin Millikin (Chromium), 2009-02-20 12:46:03]

Now with added assertions.  All over the place.

[William Hesse, 2009-02-20 13:06:20]

LGTM.

http://codereview.chromium.org/20533/diff/4/1003
File src/codegen-arm.cc (right):

http://codereview.chromium.org/20533/diff/4/1003#newcode1573
Line 1573: ASSERT(!has_valid_frame() || frame_->height() == original_height);
How could there not be a valid frame here, if the function is called with a
valid frame?  Doesn't control flow have to exit the switch statement at the end?

[Kevin Millikin (Chromium), 2009-02-20 13:12:48]

http://codereview.chromium.org/20533/diff/4/1003
File src/codegen-arm.cc (right):

http://codereview.chromium.org/20533/diff/4/1003#newcode1573
Line 1573: ASSERT(!has_valid_frame() || frame_->height() == original_height);
On 2009/02/20 13:06:20, William Hesse wrote:
> How could there not be a valid frame here, if the function is called with a
> valid frame?  Doesn't control flow have to exit the switch statement at the
end?

If the last case we compile returns and none of the other ones break, then there
is no frame at the end.

[Mads Ager (chromium), 2009-02-20 13:28:17]

LGTM