Posted Jul 6, 2010 10:46 UTC (Tue) by markh (subscriber, #33984)
Parent article: xstat() and fxstat()
> unsigned long long tv_sec, tv_nsec;
Existing filesystems support timestamps prior to the epoch (i.e. negative tv_sec), so shouldn't at least tv_sec be signed?
As for tv_nsec, I assume that its size was increased solely for alignment purposes since its value must be in the range [0,999 999 999]. Would it not be better to save the extra 32 bits in a separate field for possible future use (e.g. for indicating timestamp granularity)?