Simon Josefsson wrote: > There is size_max, which does: > > # include <limits.h>
This is needed to avoid a "macro redefined" warning e.g. on Solaris 7. > # if HAVE_STDINT_H > # include <stdint.h> > # endif This is needed to avoid a "macro redefined" warning e.g. on glibc systems. > that doesn't seem compatible with the stdint.h module (I'm assuming > that the stdint module doesn't define HAVE_STDINT_H). More explanations please? I cannot follow. If the system does not have a <stdint.h> file but gnulib creates its own, it won't be used - since HAVE_STDINT_H is 0 or undefined. If the system has a <stdint.h>, and gnulib 'stdlib' decides to provide its own nevertheless, then gnulib's will be used at compile time while the system's one was used at configure time; this is not perfect but I don't see how it could lead to an error. > So if there is > no stdint.h, and limits.h doesn't define SIZE_MAX, this module doesn't > work. In this case the size_max.m4 macro will have written a SIZE_MAX definition into config.h. The one in gnulib's stdint.h will not be used then, since it's #ifdefed out. > Is there a reason to keep size_max? We don't want "config.h" to #include any system headers. > The stdint module defines SIZE_MAX too, and appear to be more robust. I doubt that stdint.h is more robust; it's hardly used so far. > Alternatively, I think the size_max.h should contain: > > #ifndef SIZE_MAX > # define SIZE_MAX ((size_t)~(size_t)0) > #endif Then it would need to #include yet another header, such as <stddef.h>. Bruno
