[PATCH] Correct timespec implementation [BZ #26232]

[Paul Eggert]

On 7/14/20 6:17 AM, H.J. Lu via Libc-alpha wrote:
> No, it doesn't work on i686.

Could you give more details about why it doesn't work? Is it a GCC bug (and if 
so, which version and what flags did you use)? Is it a bug in 
INT_MULTIPLY_WRAPV? in INT_ADD_WRAPV? What code is generated and why is it 
wrong? That sort of thing. If there's a bug in INT_*_WRAPV I'd like to fix it.

An advantage to using 'long long' over 'time_t' is that insulates the code from 
possible future changes where time_t is unsigned (which POSIX allows). 
Admittedly this is unlikely.