Fix -Wshift-negative-value transgression in tcmalloc.

third_party/tcmalloc/chromium/src/base/basictypes.h

 // Copyright (c) 2005, Google Inc.
 // All rights reserved.
 // 
 // Redistribution and use in source and binary forms, with or without
 // modification, are permitted provided that the following conditions are
 // met:
 // 
 //     * Redistributions of source code must retain the above copyright
 // notice, this list of conditions and the following disclaimer.
 //     * Redistributions in binary form must reproduce the above
@@ skipping 64 matching lines @@
 const uint64 kuint64max = ( (((uint64) kuint32max) << 32) | kuint32max );
 
 const  int8  kint8max   = (   (  int8) 0x7F);
 const  int16 kint16max  = (   ( int16) 0x7FFF);
 const  int32 kint32max  = (   ( int32) 0x7FFFFFFF);
 const  int64 kint64max =  ( ((( int64) kint32max) << 32) | kuint32max );
 
 const  int8  kint8min   = (   (  int8) 0x80);
 const  int16 kint16min  = (   ( int16) 0x8000);
 const  int32 kint32min  = (   ( int32) 0x80000000);
-const  int64 kint64min =  ( ((( int64) kint32min) << 32) | 0 );
+const  int64 kint64min =  (-kint64max - 1LL);

[Primiano Tucci (use gerrit), 2017/04/02 16:49:44]

took my a while to figure out why this was defined like this (which, after some
googling, seems to match most #define INT64_MIN out there).
It's impressive how these days we have very sophisticated features like lambdas
with captures, rvalue references, thread-safe statics, but still, integers are
so frighteningly easy to get wrong.

Instead, I have to admit I still can't figure out the kint64max above ? Why
isn't that just 0x7fff(enough f's)LL ? (and, to be clear, I am not implying we
should change it, but just curious at this point)

[Nico, 2017/04/02 18:17:54]

0xffff is an int literal. 0xffffLL should work, but that wasn't standardized
until I think c++11, and tcmalloc predates that. I agree that would be better,
but the current way fits the file better. (There's a lot of stuff in here we
could delete with c++11.)

[Lei Zhang, 2017/04/03 20:51:07]

Do we want to file a bug to refresh tcmalloc with C++11?

 
 // Define the "portable" printf and scanf macros, if they're not
 // already there (via the inttypes.h we #included above, hopefully).
 // Mostly it's old systems that don't support inttypes.h, so we assume
 // they're 32 bit.
 #ifndef PRIx64
 #define PRIx64 "llx"
 #endif
 #ifndef SCNx64
 #define SCNx64 "llx"
@@ skipping 252 matching lines @@
 // zeroes (which the loader does for static variables), AND the destructor also
 // does nothing to the storage, then a constructor declared as
 //       explicit MyClass(base::LinkerInitialized x) {}
 // and invoked as
 //       static MyClass my_variable_name(base::LINKER_INITIALIZED);
 namespace base {
 enum LinkerInitialized { LINKER_INITIALIZED };
 }
 
 #endif  // _BASICTYPES_H_