traverseda
/
cpython
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

bpo-37205: time.perf_counter() and time.monotonic() are system-wide (GH-23284) 

time.perf_counter() on Windows and time.monotonic() on macOS are now
system-wide. Previously, they used an offset computed at startup to
reduce the precision loss caused by the float type. Use
time.perf_counter_ns() and time.monotonic_ns() added in Python 3.7 to
avoid this precision loss.
This commit is contained in:
Victor Stinner 2020-11-16 13:21:45 +01:00 committed by GitHub
parent aa01011003
commit 3df5c68487
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 43 additions and 21 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

43
Doc/library/time.rst
View File

@ -166,6 +166,9 @@ Functions
Return the time of the specified clock *clk_id*. Refer to
:ref:`time-clock-id-constants` for a list of accepted values for *clk_id*.
Use :func:`clock_gettime_ns` to avoid the precision loss caused by the
:class:`float` type.
.. availability:: Unix.
.. versionadded:: 3.3
@ -185,6 +188,9 @@ Functions
Set the time of the specified clock *clk_id*. Currently,
:data:`CLOCK_REALTIME` is the only accepted value for *clk_id*.
Use :func:`clock_settime_ns` to avoid the precision loss caused by the
:class:`float` type.
.. availability:: Unix.
.. versionadded:: 3.3
@ -273,10 +279,17 @@ Functions
The reference point of the returned value is undefined, so that only the
difference between the results of consecutive calls is valid.
Use :func:`monotonic_ns` to avoid the precision loss caused by the
:class:`float` type.
.. versionadded:: 3.3
.. versionchanged:: 3.5
The function is now always available and always system-wide.
.. versionchanged:: 3.10
On macOS, the function is now system-wide.
.. function:: monotonic_ns() -> int
@ -295,8 +308,14 @@ Functions
point of the returned value is undefined, so that only the difference between
the results of consecutive calls is valid.
Use :func:`perf_counter_ns` to avoid the precision loss caused by the
:class:`float` type.
.. versionadded:: 3.3
.. versionchanged:: 3.10
On Windows, the function is now system-wide.
.. function:: perf_counter_ns() -> int
Similar to :func:`perf_counter`, but return time as nanoseconds.
@ -317,6 +336,9 @@ Functions
returned value is undefined, so that only the difference between the results
of consecutive calls is valid.
Use :func:`process_time_ns` to avoid the precision loss caused by the
:class:`float` type.
.. versionadded:: 3.3
.. function:: process_time_ns() -> int
@ -581,6 +603,17 @@ Functions
:class:`struct_time` object is returned, from which the components
of the calendar date may be accessed as attributes.
Use :func:`time_ns` to avoid the precision loss caused by the :class:`float`
type.
.. function:: time_ns() -> int
Similar to :func:`~time.time` but returns time as an integer number of nanoseconds
since the epoch_.
.. versionadded:: 3.7
.. function:: thread_time() -> float
@ -595,6 +628,9 @@ Functions
returned value is undefined, so that only the difference between the results
of consecutive calls in the same thread is valid.
Use :func:`thread_time_ns` to avoid the precision loss caused by the
:class:`float` type.
.. availability:: Windows, Linux, Unix systems supporting
``CLOCK_THREAD_CPUTIME_ID``.
@ -608,13 +644,6 @@ Functions
.. versionadded:: 3.7
.. function:: time_ns() -> int
Similar to :func:`~time.time` but returns time as an integer number of nanoseconds
since the epoch_.
.. versionadded:: 3.7
.. function:: tzset()
Reset the time conversion rules used by the library routines. The environment

5
Misc/NEWS.d/next/Library/2020-11-14-14-34-32.bpo-37205.iDbHrw.rst Normal file
View File

@ -0,0 +1,5 @@
:func:`time.perf_counter()` on Windows and :func:`time.monotonic()` on macOS
are now system-wide. Previously, they used an offset computed at startup to
reduce the precision loss caused by the float type. Use
:func:`time.perf_counter_ns()` and :func:`time.monotonic_ns()` added in Python
3.7 to avoid this precision loss.

16
Python/pytime.c
View File

@ -6,7 +6,7 @@
#if defined(__APPLE__)
#include <mach/mach_time.h> /* mach_absolute_time(), mach_timebase_info() */
#if defined(__APPLE__) && defined(__has_builtin)
#if defined(__APPLE__) && defined(__has_builtin)
# if __has_builtin(__builtin_available)
# define HAVE_CLOCK_GETTIME_RUNTIME __builtin_available(macOS 10.12, iOS 10.0, tvOS 10.0, watchOS 3.0, *)
# endif
@ -730,7 +730,7 @@ pygettimeofday(_PyTime_t *tp, _Py_clock_info_t *info, int raise)
}
#ifdef HAVE_CLOCK_GETTIME_RUNTIME
} else {
} else {
#endif
#endif
@ -822,7 +822,6 @@ pymonotonic(_PyTime_t *tp, _Py_clock_info_t *info, int raise)
#elif defined(__APPLE__)
static mach_timebase_info_data_t timebase;
static uint64_t t0 = 0;
uint64_t ticks;
if (timebase.denom == 0) {
@ -859,8 +858,6 @@ pymonotonic(_PyTime_t *tp, _Py_clock_info_t *info, int raise)
"mach_timebase_info is too large");
return -1;
}
t0 = mach_absolute_time();
}
if (info) {
@ -871,9 +868,6 @@ pymonotonic(_PyTime_t *tp, _Py_clock_info_t *info, int raise)
}
ticks = mach_absolute_time();
/* Use a "time zero" to reduce precision loss when converting time
to floatting point number, as in time.monotonic(). */
ticks -= t0;
*tp = _PyTime_MulDiv(ticks,
(_PyTime_t)timebase.numer,
(_PyTime_t)timebase.denom);
@ -960,7 +954,6 @@ static int
win_perf_counter(_PyTime_t *tp, _Py_clock_info_t *info)
{
static LONGLONG frequency = 0;
static LONGLONG t0 = 0;
LARGE_INTEGER now;
LONGLONG ticksll;
_PyTime_t ticks;
@ -1000,7 +993,6 @@ win_perf_counter(_PyTime_t *tp, _Py_clock_info_t *info)
}
QueryPerformanceCounter(&now);
t0 = now.QuadPart;
}
if (info) {
@ -1013,10 +1005,6 @@ win_perf_counter(_PyTime_t *tp, _Py_clock_info_t *info)
QueryPerformanceCounter(&now);
ticksll = now.QuadPart;
/* Use a "time zero" to reduce precision loss when converting time
to floatting point number, as in time.perf_counter(). */
ticksll -= t0;
/* Make sure that casting LONGLONG to _PyTime_t cannot overflow,
both types are signed */
Py_BUILD_ASSERT(sizeof(ticksll) <= sizeof(ticks));