Fix mojo double and int64_t alignment expectations

Previously, alignof(int64_t) was 8 on i686, but now it is 4. This is
true in GCC 8+ and the latest version of clang. These types have always
only been aligned to 4 bytes in struct layout, so this should not have
wide ranging implications for ABI stability.

Bug: 900406
Change-Id: Ia2f0d87300679148d7ca653b7cc52fe9cf910bbc
Reviewed-on: https://chromium-review.googlesource.com/c/1327606Reviewed-by: Ken Rockot <rockot@google.com>
Commit-Queue: Reid Kleckner <rnk@chromium.org>
Cr-Commit-Position: refs/heads/master@{#606622}

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

@@ -34,8 +34,11 @@ TEST(MacrosTest, Alignof) {
   // all the platforms we currently support.
   EXPECT_EQ(1u, MOJO_ALIGNOF(char));
   EXPECT_EQ(4u, MOJO_ALIGNOF(int32_t));
-  EXPECT_EQ(8u, MOJO_ALIGNOF(int64_t));
-  EXPECT_EQ(8u, MOJO_ALIGNOF(double));
+
+  // 'double' and 'int64_t' are usually 8 byte aligned, except some 32-bit x86
+  // platforms, or when -malign-double is passed.
+  EXPECT_TRUE(8u == MOJO_ALIGNOF(int64_t) || 4u == MOJO_ALIGNOF(int64_t));
+  EXPECT_TRUE(8u == MOJO_ALIGNOF(double) || 4u == MOJO_ALIGNOF(double));
 }
 
 // These structs are used in the Alignas test. Define them globally to avoid