Repository navigation
time.process_time() constant / erratic on Windows #82040
Description
| BPO | 37859 |
|---|---|
| Nosy | @pfmoore, @vstinner, @tjguk, @zware, @eryksun, @zooba |
Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.
Show more details
GitHub fields:
assignee = None
closed_at = <Date 2019-08-14.20:48:14.849>
created_at = <Date 2019-08-14.20:32:41.783>
labels = ['type-bug', '3.8', 'invalid', '3.7', 'library', 'OS-windows']
title = 'time.process_time() constant / erratic on Windows'
updated_at = <Date 2019-08-14.21:54:45.041>
user = 'https://bugs.python.org/RedGlyph'bugs.python.org fields:
activity = <Date 2019-08-14.21:54:45.041>
actor = 'vstinner'
assignee = 'none'
closed = True
closed_date = <Date 2019-08-14.20:48:14.849>
closer = 'zach.ware'
components = ['Library (Lib)', 'Windows']
creation = <Date 2019-08-14.20:32:41.783>
creator = 'Red Glyph'
dependencies = []
files = []
hgrepos = []
issue_num = 37859
keywords = []
message_count = 6.0
messages = ['349745', '349746', '349748', '349759', '349762', '349763']
nosy_count = 7.0
nosy_names = ['paul.moore', 'vstinner', 'tim.golden', 'zach.ware', 'eryksun', 'steve.dower', 'Red Glyph']
pr_nums = []
priority = 'normal'
resolution = 'not a bug'
stage = 'resolved'
status = 'closed'
superseder = None
type = 'behavior'
url = 'https://bugs.python.org/issue37859'
versions = ['Python 3.7', 'Python 3.8']
Activity
RedGlyph commentedon Aug 14, 2019
Tested with
on Windows 7 x64 Professional
time.process_time() always returns the same value. If I check the value of time.process_time_ns(), sometimes it is constant, sometimes I observe a few changes, then it becomes constant too.
Here is a log of an example session, I'm waiting at least 1-2 seconds between each command:
Python 3.7.4 (tags/v3.7.4:e09359112e, Jul 8 2019, 20:34:20) [MSC v.1916 64 bit (AMD64)] on win32 Type "help", "copyright", "credits" or "license()" for more information. >>> import time >>> time.process_time() 0.1092007 >>> time.process_time() 0.1092007 >>> time.process_time_ns() 109200700 >>> time.process_time_ns() 124800800 >>> time.process_time_ns() 124800800 >>> time.process_time() 0.1248008 >>> time.process_time() 0.1248008 >>> time.process_time() 0.1248008 >>> time.process_time_ns() 124800800 >>> time.process_time_ns() 124800800 >>> time.process_time_ns() 124800800 >>> time.process_time_ns() 124800800 >>> time.clock() Warning (from warnings module): File "__main__", line 1 DeprecationWarning: time.clock has been deprecated in Python 3.3 and will be removed from Python 3.8: use time.perf_counter or time.process_time instead 77.006126126 >>> time.clock() 79.245575778 >>> time.clock() 80.801103036 >>> time.process_time() 0.1248008 >>>zware commentedon Aug 14, 2019
Try this:
Or:
Basically, process_time() (and process_time_ns()) measure CPU time used by the process, and you aren't making the CPU do anything between your checks.
RedGlyph commentedon Aug 14, 2019
I see that now. The behaviour was different in Linux, though, I suppose it may benefit from a more precise counter, but since in Windows it also has a precise counter with time.perf_counter_ns(), I was expecting to see that value change, but it was mainly a confusion with the older time.clock().
Thanks!
eryksun commentedon Aug 14, 2019
Don't read too much into the clock info here:
>>> time.get_clock_info('process_time').resolution 1e-07Process times [1] are stored as a 64-bit integer in units of 100 ns (1e-7). But the kernel schedules threads based on a timer that ticks every 15.625 ms by default. It can be lowered to about 0.5 ms, but this degrades battery life.
[1] https://docs.microsoft.com/en-us/windows/win32/api/processthreadsapi/nf-processthreadsapi-getprocesstimes
vstinner commentedon Aug 14, 2019
Patches are welcome to enhance time.get_clock_info() :-)
Don't rely too much on time.get_clock_info() on Linux neither: it basically always say 1 ns, even if the effective resolution is way worse.
See the PEP-418 for some numbers:
https://www.python.org/dev/peps/pep-0418/#process-time
I wrote these programs to write this PEP ;-)
https://github.com/python/peps/tree/master/pep-0418
vstinner commentedon Aug 14, 2019
On Windows, time.clock() was implemented with QueryPerformanceCounter(). This function became time.perf_counter() in Python 3.4. time.clock() was removed. Use time.get_clock_info('perf_counter') ;-)
The PEP-418 introduces new well defined clocks, since time.clock() was not portable.
perf_counter and process_time have very different properties. process_time is stopped when the process sleeps, for example.
https://docs.python.org/dev/library/time.html#time.perf_counter
https://docs.python.org/dev/library/time.html#time.process_time