Better work-around for weird GCC 4.8 warning breaking tests.

c++/src/kj/parse/char-test.c++

@@ -21,14 +21,6 @@
 // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
 // SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
-#if !__clang__ && __GNUC__ == 4 && __GNUC_MINOR__ == 8
-// GCC 4.8 warns that ConstResult_<T, Tuple<>>::operator() returns an uninitialized value from
-// the second branch.  This is bogus AFAICT.  Moreover, the warning comes from the use of the
-// `number` parser.  GCC does not complain when this parser is used in the Cap'n Proto compiler;
-// only when it is used in this test.
-#pragma GCC diagnostic ignored "-Wmaybe-uninitialized"
-#endif
-
 #include "char.h"
 #include "../string.h"
 #include <gtest/gtest.h>

super-test.sh

@@ -193,6 +193,16 @@ case ${CXX:-g++} in
     # There's an unused private field in gtest.
     export CXXFLAGS="$CXXFLAGS -Wno-unused-private-field"
     ;;
+  *g++* )
+    if (${CXX:-g++} --version | grep -q ' 4[.]8'); then
+      # GCC 4.8 emits a weird uninitialized warning in kj/parse/char-test, deep in one of the parser
+      # combinators.  It could be a real bug but there is just not enough information to figure out
+      # where the problem is coming from, because GCC does not provide any description of the inlining
+      # that has occurred.  Since I have not observed any actual problem (tests pass, etc.), I'm
+      # muting it for now.
+      CXXFLAGS="$CXXFLAGS -Wno-maybe-uninitialized"
+    fi
+    ;;
 esac
 
 cd c++