Eric Lemings wrote: > > >This is part of the <type_traits> header: > > $TOPDIR/include/type_traits: > ... > 30 #ifndef _RWSTD_TYPE_TRAITS_INCLUDED > 31 #define _RWSTD_TYPE_TRAITS_INCLUDED > 32 > 33 #include <rw/_defs.h> > 34 > 35 #ifdef _RWSTD_NO_EXT_CXX_0X > 36 # error _RWSTD_NO_EXT_CXX_0X defined and C++0x header >included > 37 #endif > ... > >Shouldn't lines 35-37 come before line 30 so the header isn't even >parsed if there is an include error? In other words: > > ... > #ifdef _RWSTD_NO_EXT_CXX_0X > # error _RWSTD_NO_EXT_CXX_0X defined and C++0x header included > #endif > > #ifndef _RWSTD_TYPE_TRAITS_INCLUDED > #define _RWSTD_TYPE_TRAITS_INCLUDED > > #include <rw/_defs.h> > ... >
You have to include rw/_defs.h before you do the _RWSTD_NO_EXT_CXX_0X check because that macro might be defined in one of the files included from there. So that means moving the rw/_defs.h include outside of the header guard. This would require an additional file open [or a lookup at the very least] as well as a condition check every time the file was included from client code. >It doesn't look conventional but, technically, it is slightly more >efficient. I'd like to see some evidence that this change would have any noticeable effect in terms of compile times. > >Brad. >
