Make --enable-heap-profiling=native "work" on Windows.

base/debug/stack_trace_unittest.cc

Index: base/debug/stack_trace_unittest.cc
diff --git a/base/debug/stack_trace_unittest.cc b/base/debug/stack_trace_unittest.cc
index 560dc1ddde8dc5ee7dd42e62b0e0e296ad5a5ac4..652bdb15ab468cde8cd24e48dbd7b89d218a78c1 100644
--- a/base/debug/stack_trace_unittest.cc
+++ b/base/debug/stack_trace_unittest.cc
@@ -254,7 +254,9 @@ TEST_F(StackTraceTest, itoa_r) {
 }
 #endif  // defined(OS_POSIX) && !defined(OS_ANDROID)
 
-#if HAVE_TRACE_STACK_FRAME_POINTERS
+#if HAVE_TRACE_STACK_FRAME_POINTERS && !defined(OS_WIN)
+// MSVC (Windows) doesn't generate frame pointers and can't take the address

[dcheng, 2017/03/13 02:11:21]

I'm a bit confused about the comments vs the GN changes: the GN changes mean we
generate frame pointers, but this comments says we don't...

[Wez, 2017/03/14 05:22:39]

Originally we planned to use frame pointers; this comment is correct for Win64
builds, which don't support frame pointers at all - updated it to correctly
specify why we're skipping these tests.

+// of labels.
 
 template <size_t Depth>
 void NOINLINE ExpectStackFramePointers(const void** frames,
@@ -313,7 +315,7 @@ TEST_F(StackTraceTest, MAYBE_StackEnd) {
   EXPECT_NE(0u, GetStackEnd());
 }
 
-#endif  // HAVE_TRACE_STACK_FRAME_POINTERS
+#endif  // HAVE_TRACE_STACK_FRAME_POINTERS && !defined(OS_WIN)
 
 }  // namespace debug
 }  // namespace base