Let bionic use the correct definition of ssize_t.

Bug: 8253769
Change-Id: I50c7cc20828fc089b83580e039ce9153a6c5a8cc

tests/stdio_test.cpp

@@ -171,19 +171,15 @@ TEST(stdio, getline_invalid) {
 }
 
 TEST(stdio, printf_ssize_t) {
-#if __BIONIC__
   // http://b/8253769
-  ASSERT_EQ(sizeof(__kernel_ssize_t), sizeof(long int));
   ASSERT_EQ(sizeof(ssize_t), sizeof(long int));
-#else
-  // TODO: add a .c file so we can test this for bionic --- our C ssize_t is fine.
-  // For our 32-bit C++ ABI, we have a ssize_t definition that confuses GCC into saying:
+  ASSERT_EQ(sizeof(ssize_t), sizeof(size_t));
+  // For our 32-bit ABI, we had a ssize_t definition that confuses GCC into saying:
   // error: format '%zd' expects argument of type 'signed size_t',
   //     but argument 4 has type 'ssize_t {aka long int}' [-Werror=format]
   ssize_t v = 1;
   char buf[32];
   snprintf(buf, sizeof(buf), "%zd", v);
-#endif
 }
 
 TEST(stdio, popen) {