Add partitionAllocGetSize() for determining actual size of allocation

Source/wtf/PartitionAlloc.h

Index: Source/wtf/PartitionAlloc.h
diff --git a/Source/wtf/PartitionAlloc.h b/Source/wtf/PartitionAlloc.h
index b1354777c6686b1566ab0a8cb56722ce9aedc7d8..1500bb5ac79da2f81f308de6dbc3eae07a415b30 100644
--- a/Source/wtf/PartitionAlloc.h
+++ b/Source/wtf/PartitionAlloc.h
@@ -565,6 +565,24 @@ ALWAYS_INLINE void partitionFreeGeneric(PartitionRootGeneric* root, void* ptr)
 #endif
 }
 
+ALWAYS_INLINE size_t partitionAllocGetSize(PartitionRootGeneric* root, void* ptr)
+{
+#if defined(MEMORY_TOOL_REPLACES_ALLOCATOR)
+    ASSERT_NOT_REACHED();

[Chris Evans, 2014/02/18 22:19:59]

I think this definitely will be reached. We want the client to call this API
unconditionally, right?

Hmm, I think the way to resolve this that doesn't add cost (branching, etc.) to
the common case would be to add a new ALWAYS_INLINE API:

partitionAllocSupportsGetSize(). It'd just return a hardwired true / false,
depending on MEMORY_TOOL_REPLACES_ALLOCATOR, allowing the compiler to optimize
fully.

+#else
+    ASSERT(root->initialized);
+    if (UNLIKELY(!ptr))
+        return 0;

[Chris Evans, 2014/02/18 22:19:59]

For simplicity, maybe try not supporting passing in NULL? I don't think the
Vector client will need to do that, for example.

+    ptr = partitionCookieFreePointerAdjust(ptr);
+    ASSERT(partitionPointerIsValid(ptr));
+    PartitionPage* page = partitionPointerToPage(ptr);
+    spinLockLock(&root->lock);

[Chris Evans, 2014/02/18 22:19:59]

No need for locking, as covered separately. But definitely add a small comment
to make it clear that the lack of locking is willful.

+    size_t size = page->bucket->slotSize;
+    spinLockUnlock(&root->lock);
+    return partitionCookieSizeAdjustSubtract(size);
+#endif
+}
+
 // N (or more accurately, N - sizeof(void*)) represents the largest size in
 // bytes that will be handled by a SizeSpecificPartitionAllocator.
 // Attempts to partitionAlloc() more than this amount will fail.