On 01/18/2010 12:24 PM, YuGr wrote: > I have recently run into a non-portable c-torture test > I have seen something similar in the past appearing on the mailing list and maybe I can say something, briefly. A first consideration is that in GCC we have, in general, regression testsuites, not C/C++/whatever Standard conformance testsuites. Thus, in particular, I don't think we are aiming for full generality and portability outside GCC. For example, for C++ and its library, which I know better than C, we are always implicitly assuming the multi-vendor ABI additionally to the Standard, and where it applies, Posix.
Pragmatically, I don't think people are really happy to start including limits.h everywhere or even conditionalizing tests for the most exotic combination of the limits themselves if those circumstances do not exist for the targets actually supported by GCC. But I may be wrong about many details, seems a good occasion to clarify some of those issues for the benefit of everybody. Paolo.
