unchecked_malloc() for Skia on OSX.

base/process_util.h

Index: base/process_util.h
diff --git a/base/process_util.h b/base/process_util.h
index 25d24202272c9cb3413a61db2658e75da2ba6d32..5efc25cc8b0931322140fb5cec25667fbd998fbd 100644
--- a/base/process_util.h
+++ b/base/process_util.h
@@ -856,6 +856,16 @@ BASE_EXPORT void RaiseProcessToHighPriority();
 void RestoreDefaultExceptionHandler();
 #endif  // defined(OS_MACOSX)
 
+#if defined(OS_MACOSX)
+// Very large images or svg canvases can cause huge mallocs.  Skia
+// does tricks on tcmalloc-based systems to allow malloc to fail with
+// a NULL rather than hit the oom crasher.  This replicates that for
+// OSX.
+// TODO(shess): Weird place to put it, but this is where the OOM
+// killer currently lives.
+BASE_EXPORT void* oom_safe_malloc(size_t size);

[Avi (use Gerrit), 2012/09/13 19:45:54]

I don't like the name "oom safe". While it's safe from the oom killer, it's a
very _unsafe_ call overall. I'd lean towards "unsafe_malloc" or the like.

[Scott Hess - ex-Googler, 2012/09/13 20:04:00]

Well, it's not _unsafe_, per se.  Just against general policy. 
unchecked_malloc()?  original_malloc() or system_malloc() seem kind
implementation-driven, but since this is explicitly a case about avoiding a bit
of our implementation, seems fair.

I suppose it would also be worth putting in a scary comment, like "If you use
this without running it by a few of the OSX developers, your code will be
reverted."

+#endif  // defined(OS_MACOSX)
+
 }  // namespace base
 
 #endif  // BASE_PROCESS_UTIL_H_