http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52718
Manuel López-Ibáñez <manu at gcc dot gnu.org> changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |manu at gcc dot gnu.org --- Comment #18 from Manuel López-Ibáñez <manu at gcc dot gnu.org> 2012-11-01 23:00:51 UTC --- (In reply to comment #15) > This is what I meant when I said that the issue is different, and is much more > general than -Wzero-as-null-pointer-constant. Consider, eg, with -Woverflow: > > #pragma GCC system_header > > class Foo > { > public: > Foo(signed char); > }; > > class Bar > { > public: > Bar(Foo = 10000); > }; The issue is that literals don't have locations, so input_location is used. And input_location here is in the main file, so system_header doesn't apply. This is why input_location must die. Clang uses the correct location for warning, so it honors system_header (requires -Wsystem-headers): In file included from pr52718.c:1: ./pr52718.h:11:13: warning: implicit conversion from 'int' to 'signed char' changes value from 10000 to 16 [-Wconstant-conversion] Bar(Foo = 10000); ^~~~~ 1 warning generated. Ultimately, GCC needs to track the locations of literals like clang does. Everything else is a hack around the main issue. This is PR43486. There have been some recent attempts http://gcc.gnu.org/ml/gcc-patches/2012-09/msg01222.html, but I am afraid that work seems stalled for lack of time/help.