Allocator shims working on VS2015.

build/common.gypi

Index: build/common.gypi
diff --git a/build/common.gypi b/build/common.gypi
index 7c4e60ede1586920c38b67cd49788b3c20f8dbde..df361db3ff0bb65906c7efb63d642beaece25dc4 100644
--- a/build/common.gypi
+++ b/build/common.gypi
@@ -1353,12 +1353,12 @@
     'android_lint%': 1,
 
     # Although base/allocator lets you select a heap library via an
-    # environment variable, the libcmt shim it uses sometimes gets in
-    # the way.  To disable it entirely, and switch to normal msvcrt, do e.g.
+    # environment variable, the shim it uses sometimes gets in the way.
+    # To disable it entirely, and switch to normal msvcrt, do e.g.
     #  'win_use_allocator_shim': 0,
     #  'win_release_RuntimeLibrary': 2
     # to ~/.gyp/include.gypi, gclient runhooks --force, and do a release build.
-    'win_use_allocator_shim%': 1, # 1 = shim allocator via libcmt; 0 = msvcrt
+    'win_use_allocator_shim%': 1, # 1 = shim allocator; 0 = msvcrt
 
     # Enables the unified allocator shim (experimental) which routes all the
     # alloc calls to base/. Right now is supported on Linux Desktop only.
@@ -2016,11 +2016,7 @@
           },{
             'winsdk_arch%': '<(target_arch)',
           }],
-          ['component=="shared_library" or MSVS_VERSION == "2015"', {
-            # TODO(scottmg): The allocator shimming doesn't work on the 2015 CRT
-            # and we are hoping to be able to remove it if an additional feature
-            # lands in the 2015 CRT API. For now, don't shim and revisit once
-            # VS2015 is RTM: http://crbug.com/481611.
+          ['component=="shared_library"', {
             'win_use_allocator_shim%': 0,
           }],
           ['component=="static_library"', {
@@ -2866,11 +2862,6 @@
       }],  # fieldtrial_testing_like_official_build==0 and branding!="Chrome"
       ['OS=="win"', {
         'defines': ['NO_TCMALLOC'],
-        'conditions': [
-          ['win_use_allocator_shim==1', {
-            'defines': ['ALLOCATOR_SHIM'],
-          }],
-        ],
       }],
       ['asan==1', {
         'defines': [
@@ -3538,8 +3529,10 @@
               'WTF_USE_DYNAMIC_ANNOTATIONS=1',
             ],
           }],
-          ['OS=="win"', {
-            'defines': ['NO_TCMALLOC'],
+          ['OS=="win" and win_use_allocator_shim==1', {
+            'defines': [
+              'ALLOCATOR_SHIM'

[Primiano Tucci (use gerrit), 2016/03/29 18:46:07]

What is the reason for removing NO_TCMALLOC here?
IIRC that drives in build_config the generation of USE_TCMALLOC, which will now
become falsely true on windows.

[Will Harris, 2016/03/29 18:56:21]

bruce asked the same question in chat.

NO_TCMALLOC is defined above on line 2864 so this was always a second duplicate
definition. I fixed the glitch when I was adding the Release specific
configuration.

[Primiano Tucci (use gerrit), 2016/03/29 19:30:03]

Ahh makes sense, thanks.

+            ],
           }],
           # _FORTIFY_SOURCE isn't really supported by Clang now, see
           # http://llvm.org/bugs/show_bug.cgi?id=16821.