Re: What is supposed to replace clock_t?

[Zack Weinberg]

Linus Torvalds wrote:
...
> Many of the binary interfaces are perfectly fine. In fact, there are
> very few binary interfaces that are fundamentally broken, the
> obvious one being the "times()" system call that nobody actually
> uses any more.

Er, no; people do still use times().  It is (as far as I know) the
only way for a process to determine how much wall-clock, user CPU, and
system CPU time it has consumed, all at once.  If you use getrusage()
instead, you also have to call gettimeofday() to get the wall-clock
time, which at least doubles the overhead.  [Both getrusage and
gettimeofday are somewhat more expensive than times, but I'm pretty
sure the trap cost dominates.]  Profiling code tries very hard to have
as little overhead as possible.

As an application programmer, I would be perfectly happy to use an
extended getrusage() that gave me wall-clock time as a struct timeval.
But please do provide such an interface before deprecating times().

[While we're at it, it would be nice if the kernel would provide
useful values for more of the fields of struct rusage.]

zw