[tracing] Add SkResourceCache as suballocation to avoid double counting.

base/memory/discardable_shared_memory.h

Index: base/memory/discardable_shared_memory.h
diff --git a/base/memory/discardable_shared_memory.h b/base/memory/discardable_shared_memory.h
index 74bbe8e95f427ba347d696e6c6e9c408e166e167..83c128af532def6d6ba9488dba6f440b5a2135f4 100644
--- a/base/memory/discardable_shared_memory.h
+++ b/base/memory/discardable_shared_memory.h
@@ -29,6 +29,13 @@ namespace base {
 // access to an instance of this class.
 class BASE_EXPORT DiscardableSharedMemory {
  public:
+  // Used by other allocators to expess sub-allocation from discardable memory
+  // heap for tracing. To expess any suballocation, the dump providers need to
+  // create a sub-allocation edge from this node.
+  // E.g.: ProcessMemoryDump::AddSuballocation(skia_dump_guid,
+  //                                           kAllocatedObjectsDumpName);
+  static const char kAllocatedObjectsDumpName[];

[reveman, 2015/07/29 18:32:11]

I don't think we want this here. We shouldn't assume that DiscardableMemory
instances are backed by DiscardableSharedMemory.

[ssid, 2015/07/29 19:36:22]

Do you think DiscardableMemoryAllocator is a better place for this? I initially
placed it there, but moved it here since the kAllocatedObjectsDumpName is used
only if shared memory allocator is used.

[reveman, 2015/07/30 16:00:59]

Yes, DiscardableMemoryAllocator is better. That name we add there will be
correct independent of the underlying discardable memory system. Shared memory
based or not.

Also, can we make the name consistent with the name for system allocator dumps
and wrap it in a function?
https://code.google.com/p/chromium/codesearch#chromium/src/base/trace_event/m...

GetDiscardableAllocatorMemoryPoolName() maybe?

[ssid, 2015/07/30 20:05:15]

Thanks I din't know this existed.

+
   enum LockResult { SUCCESS, PURGED, FAILED };
 
   DiscardableSharedMemory();