GC branch: Some changes to make V8 work with different page sizes.

src/spaces.cc

Index: src/spaces.cc
===================================================================
--- src/spaces.cc	(revision 8270)
+++ src/spaces.cc	(working copy)
@@ -728,14 +728,6 @@
 }
 
 
-void PagedSpace::SetAllocationInfo(Address top, Address limit) {
-  Free(allocation_info_.top, allocation_info_.limit - allocation_info_.top);
-  allocation_info_.top = top;
-  allocation_info_.limit = limit;
-  ASSERT(allocation_info_.VerifyPagedAllocation());
-}
-
-
 bool PagedSpace::Expand() {
   ASSERT(max_capacity_ % Page::kObjectAreaSize == 0);
   ASSERT(Capacity() % Page::kObjectAreaSize == 0);
@@ -1057,6 +1049,7 @@
       // Next page should be valid.
       CHECK(!page->is_anchor());
       current = page->body();
+      continue;

[Vyacheslav Egorov (Chromium), 2011/06/14 10:59:09]

I do not think this is necessary.

[Erik Corry, 2011/06/14 18:51:21]

It made a bug go away for me.

If the last object on the previous page goes right up to the limit and there are
no objects on the second page then it goes wrong without this line.

[Lasse Reichstein, 2011/06/15 06:56:27]

It should be there.
There is nothing preventing top() from being either at the end of one page or at
the beginning of the next page, effectively representing the same location. This
tests both.

It's likely that both cases don't occur at the same places, e.g., that we
advance to the next page when possible, but on the very last page of the space,
we stay at the end of the page. But there is no invariant saying what we do, so
we should handle both possibilities in all cases.

     }
     // The allocation pointer should not be in the middle of an object.
     CHECK(!page->ContainsLimit(top()) || current < top());