After time.wallclock() (issue #10278), let add a time.monotomic() function. It is similar to time.wallclock() (try to get the most accurate clock) but is not available if the system doesn't provide a monotonic clock. It may also fail at runtime if Python cannot find a monotonic clock, whereas time.clock() and time.wallclock() fallback on a wallclock which may go backward on NTP adjust.

The documentation of the patch should be improved :-)

wallclock() tests may be simplified or dropped because they may fail on NTP adjust.

Do we actually yet another function, or could this be covered by adding a parameter such as monotonic=False, perhaps to wallclock().

> Do we actually yet another function, or could this be covered by adding a parameter such as monotonic=False, perhaps to wallclock().

A monotonic is a different clock, it would be surprising that an
argument uses another clock.

I don’t understand why this new function would be useful.  Time-related modules in Python are already complicated.

If you are trying to time something (an interval), having the time go backward can really screw up your data.  And that *will* happen on a system that is running NTP (or even just resets its time).  monotonic clocks were introduced at the OS level for a reason, and it seems reasonable for Python to expose them (when the are available) like it does other system resources.

Thanks, now I see the usefulness.

New changeset 376ce937823c by Victor Stinner in branch 'default':
Issue #13846: Add time.monotonic(), monotonic clock.
http://hg.python.org/cpython/rev/376ce937823c

Victor, I think the doc doesn’t say why the function is useful for people like me who don’t already know it.  David’s explanation could be reused.

Boost has a monotonic time implementation (steady_clock) for OS X:
http://svn.boost.org/svn/boost/trunk/boost/chrono/detail/inlined/mac/chrono.hpp

New changeset 50b1f2d07011 by Victor Stinner in branch 'default':
Issue #13846: Enhance time.monotonic() documentation
http://hg.python.org/cpython/rev/50b1f2d07011

> Boost has a monotonic time implementation (steady_clock) for OS X

I opened a new issue for Mac OS X: #14104.

--

@Éric: I enhanced the doc. Feel free to complete or rewrite it if you have better information about such clocks.