[Mac] Capture -dealloc backtrace to log with CrZombie messages.

base/debug/stack_trace.h

Index: base/debug/stack_trace.h
diff --git a/base/debug/stack_trace.h b/base/debug/stack_trace.h
index 02cb01f2c5bb30de5036648ecdb82f51b5c046f0..be68c75e26a2bc2050f42a1de52b109d763c8270 100644
--- a/base/debug/stack_trace.h
+++ b/base/debug/stack_trace.h
@@ -26,6 +26,10 @@ class BASE_EXPORT StackTrace {
   // Creates a stacktrace from the current location.
   StackTrace();
 
+  // Creates a stacktrace from an existing buffer gotten elsewhere.
+  // Useful for sharing PrintBacktrace().
+  StackTrace(const void* const* trace, int count);

[Mark Mentovai, 2011/08/27 00:29:02]

Since this is part of the public interface now, you kind of need to say what the
trace argument is all about. I can see from the implementation file that it’s
dumped right into trace_, but that doesn’t really help someone who doesn’t know
any better because the contents of trace_ isn’t defined below other than to say
that it’s an array of up to kMaxTraces pointers.

[Scott Hess - ex-Googler, 2011/08/27 05:15:40]

OK, I'll document it as accepting an array of instruction pointers such as
returned by Addresses().

Given that, would it make sense to have the zombie code generate an instance and
store results from Addresses(), or to continue using backtrace() directly?  I'm
a bit paranoid about overhead, but I find that a hard decision (it's called a
lot, arguing for low overhead, and this class of code is hardly abstract in the
first place).

+
 #if defined(OS_WIN)
   // Creates a stacktrace for an exception.
   // Note: this function will throw an import not found (StackWalk64) exception