Before a scavenge collection in debug builds with ENABLE_SLOW_ASSERTS,...

Before a scavenge collection in debug builds with ENABLE_SLOW_ASSERTS,
we verify that there are no pointers to new space from the code space.
Add the old data space to this verification.


Committed: http://code.google.com/p/v8/source/detail?r=1897

[Kevin Millikin (Chromium), 2009-05-07 09:37:13]

I bet this will make some tests time out, because it roughly doubles scavenge
time with ENABLE_SLOW_ASSERTS.  I'll sort that out and add it to the change list
before submitting.

[Erik Corry, 2009-05-07 09:53:01]

LGTM

http://codereview.chromium.org/113095/diff/1001/1002
File src/heap.cc (right):

http://codereview.chromium.org/113095/diff/1001/1002#newcode559
Line 559: static void VerifyNonPointerSpacePointers() {
If you move the flag test into this you and put an empty implementation in an
#else you can get rid of an intra-function ifdef at the cost of adding an
extra-function #else.  That seems like a clear style win to me.

[Kevin Millikin (Chromium), 2009-05-07 10:24:02]

http://codereview.chromium.org/113095/diff/1001/1002
File src/heap.cc (right):

http://codereview.chromium.org/113095/diff/1001/1002#newcode559
Line 559: static void VerifyNonPointerSpacePointers() {
On 2009/05/07 09:53:01, Erik Corry wrote:
> If you move the flag test into this you and put an empty implementation in an
> #else you can get rid of an intra-function ifdef at the cost of adding an
> extra-function #else.  That seems like a clear style win to me.

Not a clear win.  I actually prefer seeing at the call site that nothing happens
unless in DEBUG and with --enable-slow-asserts.