Steve Ellcey <steve.ell...@imgtec.com> writes: > I was looking at a bug in my new memcpy in glibc & newlib and traced it > down to this code: > > #if (_MIPS_ISA == _MIPS_ISA_MIPS4) || (_MIPS_ISA == _MIPS_ISA_MIPS5) || \ > (_MIPS_ISA == _MIPS_ISA_MIPS32) || (_MIPS_ISA == _MIPS_ISA_MIPS64) > > The problem is that GCC is defining _MIPS_ISA to something > (_MIPS_ISA_MIPS1 in this case) but it does not define _MIPS_ISA_MIPS1 > or _MIPS_ISA_MIPS4 (or any other _MIPS_ISA_MIPS* macro) to anything. > So if _MIPS_ISA is defined to _MIPS_ISA_MIPS1, this if statement is > considered to be true because _MIPS_ISA expands to _MIPS_ISA_MIPS1 which > is expanded to nothing. And since _MIPS_ISA_MIPS4 (or 5 or 32 or 64) > is also not defined they are expanded to nothing and we get a true > comparison when we did not want one.
The _MIPS_ISA system comes from IRIX, and the _MIPS_ISA_* macros are supposed to be used after: #include <sgidefs.h> Not the nicest interface, but that's the way it worked. IIRC __mips is a GNU thing and I agree it's a better interface. Richard
