On Tue, Dec 3, 2013 at 1:03 PM, Matthew Dempsky <matt...@dempsky.org> wrote: > On Tue, Dec 3, 2013 at 12:37 PM, Philip Guenther <guent...@gmail.com> wrote: >> <string.h> needs to stay until FD_ZERO() and FD_COPY() are changed to >> not use memset()/memcpy(). > > Good point. > > Would something like this work? > > #define FD_COPY(f, t) (*(fd_set *)(t) = *(const fd_set *)(f))
What problem does the casts solve? (And actually, as noted in naddy's cite from FreeBSD, FD_COPY() is a non-standard BSDism, so it can just continue to assume the caller pulled in <string.h>) > static const fd_set __fd_zero_set; > #define FD_ZERO(p) FD_COPY(&__fd_zero_set, p) That'll put a __fd_zero_set in each .o that references FD_ZERO(), ick. I think I would prefer either the FreeBSD inline bzero() routine (although zeroing from the end seems an odd choice for caching reasons, no?), or adding a strong alias to libc, _memset == memset, and then having sys/select.h say void *_memset(void *, int, size_t); #define FD_ZERO(p) (void )_memset(p, 0, sizeof(*(p))) > Or we can just declare memset/memcpy in sys/select.h too? Magic POSIX 8-ball says "no". Philip Guenther
