Enable MSVC warning for unused locals.

mojo/public/cpp/system/tests/macros_unittest.cc

Index: mojo/public/cpp/system/tests/macros_unittest.cc
diff --git a/mojo/public/cpp/system/tests/macros_unittest.cc b/mojo/public/cpp/system/tests/macros_unittest.cc
index 99b158dabe10dd3a39bcef6827f3c28b35b0c553..f884baaf1866b409fdb07b6696cf4e354970a0a9 100644
--- a/mojo/public/cpp/system/tests/macros_unittest.cc
+++ b/mojo/public/cpp/system/tests/macros_unittest.cc
@@ -67,6 +67,7 @@ static_assert(MOJO_ARRAYSIZE(kGlobalArray) == 5u,
 
 TEST(MacrosCppTest, ArraySize) {
   double local_array[4] = {6.7, 7.8, 8.9, 9.0};
+  MOJO_ALLOW_UNUSED_LOCAL(local_array);

[jamesr, 2014/11/18 00:44:16]

this local seems to be used on the next line - no? is this a toolchain bug?

[Peter Kasting, 2014/11/18 00:50:30]

MSVC considers something unused if e.g. you only ever take sizeof(local), or
refer to it in typedefs, or the like, and not actually use its value.  I find
this arguable, but it turns out to be fairly rare that we actually want to use a
variable for only such purposes and nothing else, so it's easy enough to just
say this is ALLOW_UNUSED and move on.

The total number of places across the codebase that MSVC's quirks cause us to
have to do something otherwise-ugly, like this, is a small subset of the changes
in this CL -- less than half a dozen total -- and the process of enabling this
warning has turned up some questionable code and potential bugs, so I think the
tradeoff is worth it.

   EXPECT_EQ(4u, MOJO_ARRAYSIZE(local_array));
 }