1. Enable large object pointer offset check in release build.

base/allocator/allocator.gyp

 # Copyright (c) 2012 The Chromium Authors. All rights reserved.
 # Use of this source code is governed by a BSD-style license that can be
 # found in the LICENSE file.
 
 {
   'variables': {
     'jemalloc_dir': '../../third_party/jemalloc/chromium',
     'tcmalloc_dir': '../../third_party/tcmalloc/chromium',
   },
   'targets': [
@@ skipping 183 matching lines @@
         # jemalloc files
         '<(jemalloc_dir)/jemalloc.c',
         '<(jemalloc_dir)/jemalloc.h',
         '<(jemalloc_dir)/ql.h',
         '<(jemalloc_dir)/qr.h',
         '<(jemalloc_dir)/rb.h',
 
         'allocator_shim.cc',
         'allocator_shim.h',
         'generic_allocators.cc',
-        'win_allocator.cc',        
+        'win_allocator.cc',
       ],
       # sources! means that these are not compiled directly.
       'sources!': [
         # Included by allocator_shim.cc for maximal inlining.
         'generic_allocators.cc',
         'win_allocator.cc',
 
         # We simply don't use these, but list them above so that IDE
         # users can view the full available source for reference, etc.
         '<(tcmalloc_dir)/src/addressmap-inl.h',
@@ skipping 206 matching lines @@
             '<(tcmalloc_dir)/src/heap-checker.cc',
           ],
           # Disable the heap checker in tcmalloc.
           'defines': [
             'NO_HEAP_CHECK',
           ],
         }],
       ],
     },
     {
-      # This library is linked in to libbase and allocator_unittests.
-      # It can't depend on either and nothing else should depend on it -
+      # This library is linked in to libbase, allocator_unittests and TCMalloc

[rvargas (doing something else), 2012/05/19 01:08:04]

nit: libbase -> base

[kaiwang, 2012/05/21 18:11:42]

Done.

+      # tests. It can't depend on either and nothing else should depend on it -

[rvargas (doing something else), 2012/05/19 01:08:04]

If you want to refer to this as tcmalloc tests, the target should be named
tcmalloc_tests

[kaiwang, 2012/05/21 18:11:42]

Done, modified comment and renamed to tcmalloc_unittests

       # all other code should use the interfaced provided by libbase.
       'target_name': 'allocator_extension_thunks',
       'type': 'static_library',
       'sources': [
         'allocator_extension_thunks.cc',
         'allocator_extension_thunks.h',
       ],
       'toolsets': ['host', 'target'],
       'include_dirs': [
         '../../'
@@ skipping 53 matching lines @@
           'toolsets': ['host', 'target'],
           'include_dirs': [
             '../../'
           ],
           'configurations': {
             'Common_Base': {
               'msvs_target_platform': 'x64',
             },
           },
         },
+      {
+        'target_name': 'tcmalloc_free_check_test',
+        'type': 'executable',
+        'sources': [
+          'tcmalloc_free_check_test.cc',
+
+          # Part of chromium code(instead of TCMalloc). We don't include the
+          # TCMalloc version(atomicops-internals-x86.cc) in allocator library,
+          # so need this for replacement.
+          '../atomicops_internals_x86_gcc.cc',
+
+          # Code allocator_shim.cc depends on.
+          '../profiler/alternate_timer.cc',
+          '../profiler/alternate_timer.h',
+        ],
+        'include_dirs': [
+          # This is tricky.. We add this include dir so alternate_timer.cc will
+          # use TCMalloc's logging.h, instead of the chromium base/logging.h as

[rvargas (doing something else), 2012/05/19 01:08:04]

say that we don't want to depend on base. (why not?)

[kaiwang, 2012/05/21 18:11:42]

Maybe I'm wrong. base is defined in base.gypi, I thought the only way to depend
on it is to include it into this gyp? They all other targets here will depend on
base.

[rvargas (doing something else), 2012/05/21 18:44:30]

I'm not sure what you mean by "include it into this gyp". Each target can have
different requirements than other targets on the same file. If you want to
depend on base, just add another line to the dependencies section (just like
this project depends on gtest):

'..\base.gyp:base'

(note the use of base.gyp, as opposed to gypi).

I thought you were trying to avoid that dependency for some reason.

[kaiwang, 2012/05/21 22:21:42]

You are right. This makes this target a lot more simpler!

+          # in other build targets.
+          '<(tcmalloc_dir)/src',
+

[rvargas (doing something else), 2012/05/19 01:08:04]

remove line

[kaiwang, 2012/05/21 18:11:42]

The above comment is only targeting one line of the includes. I think adding
this newline will help clarify this?

[rvargas (doing something else), 2012/05/21 18:44:30]

That's a good point, I missed that. But then, it is better to have this line
before the others, especially because the list should be alpha ordered in any
case :) ... and that points me to the other lists of this target: they are not
correctly ordered.

[kaiwang, 2012/05/21 22:21:42]

Putting tcmalloc/src before ../.. is because tcmalloc dir must be checked first
for header files, or building will fail.

But now I depend on base, all these tricky things are not necessary :)

+          '../..',
+        ],
+        'dependencies': [
+          'allocator',
+          'allocator_extension_thunks',
+          '../../testing/gtest.gyp:gtest',
+        ],
+      },
       ],
     }],
   ],
 }