mirrors/cpython
0
0
Fork 0
mirror of https://github.com/python/cpython.git synced 2024-11-24 10:24:35 +08:00
Code Issues Packages Projects Releases Wiki Activity
6e22149cb6
cpython/Lib/profile.py
Tim Peters 6e22149cb6 Repair some longstanding comment errors: 
+ The last index in the timing tuple is 4, not 5 (noted by Guido).

+ The poorly named trace_dispatch_i works with float return values too.
2001-10-07 04:02:36 +00:00

519 lines
19 KiB
Python
Executable File
Raw Blame History

#! /usr/bin/env python
#
# Class for profiling python code. rev 1.0 6/2/94
#
# Based on prior profile module by Sjoerd Mullender...
# which was hacked somewhat by: Guido van Rossum
#
# See profile.doc for more information
"""Class for profiling Python code."""
# Copyright 1994, by InfoSeek Corporation, all rights reserved.
# Written by James Roskind
#
# Permission to use, copy, modify, and distribute this Python software
# and its associated documentation for any purpose (subject to the
# restriction in the following sentence) without fee is hereby granted,
# provided that the above copyright notice appears in all copies, and
# that both that copyright notice and this permission notice appear in
# supporting documentation, and that the name of InfoSeek not be used in
# advertising or publicity pertaining to distribution of the software
# without specific, written prior permission. This permission is
# explicitly restricted to the copying and modification of the software
# to remain in Python, compiled Python, or other languages (such as C)
# wherein the modified or derived code is exclusively imported into a
# Python module.
#
# INFOSEEK CORPORATION DISCLAIMS ALL WARRANTIES WITH REGARD TO THIS
# SOFTWARE, INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND
# FITNESS. IN NO EVENT SHALL INFOSEEK CORPORATION BE LIABLE FOR ANY
# SPECIAL, INDIRECT OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER
# RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF
# CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN
# CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
import sys
import os
import time
import marshal
__all__ = ["run","help","Profile"]
# Sample timer for use with
#i_count = 0
#def integer_timer():
# global i_count
# i_count = i_count + 1
# return i_count
#itimes = integer_timer # replace with C coded timer returning integers
#**************************************************************************
# The following are the static member functions for the profiler class
# Note that an instance of Profile() is *not* needed to call them.
#**************************************************************************
def run(statement, filename=None):
"""Run statement under profiler optionally saving results in filename
This function takes a single argument that can be passed to the
"exec" statement, and an optional file name. In all cases this
routine attempts to "exec" its first argument and gather profiling
statistics from the execution. If no file name is present, then this
function automatically prints a simple profiling report, sorted by the
standard name string (file/line/function-name) that is presented in
each line.
"""
prof = Profile()
try:
prof = prof.run(statement)
except SystemExit:
pass
if filename is not None:
prof.dump_stats(filename)
else:
return prof.print_stats()
# print help
def help():
for dirname in sys.path:
fullname = os.path.join(dirname, 'profile.doc')
if os.path.exists(fullname):
sts = os.system('${PAGER-more} '+fullname)
if sts: print '*** Pager exit status:', sts
break
else:
print 'Sorry, can\'t find the help file "profile.doc"',
print 'along the Python search path'
if os.name == "mac":
import MacOS
def _get_time_mac(timer=MacOS.GetTicks):
return timer() / 60.0
if hasattr(os, "times"):
def _get_time_times(timer=os.times):
t = timer()
return t[0] + t[1]
class Profile:
"""Profiler class.
self.cur is always a tuple. Each such tuple corresponds to a stack
frame that is currently active (self.cur[-2]). The following are the
definitions of its members. We use this external "parallel stack" to
avoid contaminating the program that we are profiling. (old profiler
used to write into the frames local dictionary!!) Derived classes
can change the definition of some entries, as long as they leave
[-2:] intact (frame and previous tuple). In case an internal error is
detected, the -3 element is used as the function name.
[ 0] = Time that needs to be charged to the parent frame's function.
It is used so that a function call will not have to access the
timing data for the parent frame.
[ 1] = Total time spent in this frame's function, excluding time in
subfunctions
[ 2] = Cumulative time spent in this frame's function, including time in
all subfunctions to this frame.
[-3] = Name of the function that corresponds to this frame.
[-2] = Actual frame that we correspond to (used to sync exception handling)
[-1] = Our parent 6-tuple (corresponds to frame.f_back)
Timing data for each function is stored as a 5-tuple in the dictionary
self.timings[]. The index is always the name stored in self.cur[-3].
The following are the definitions of the members:
[0] = The number of times this function was called, not counting direct
or indirect recursion,
[1] = Number of times this function appears on the stack, minus one
[2] = Total time spent internal to this function
[3] = Cumulative time that this function was present on the stack. In
non-recursive functions, this is the total execution time from start
to finish of each invocation of a function, including time spent in
all subfunctions.
[4] = A dictionary indicating for each function name, the number of times
it was called by us.
"""
def __init__(self, timer=None):
self.timings = {}
self.cur = None
self.cmd = ""
if not timer:
if os.name == 'mac':
self.timer = MacOS.GetTicks
self.dispatcher = self.trace_dispatch_mac
self.get_time = _get_time_mac
elif hasattr(time, 'clock'):
self.timer = self.get_time = time.clock
self.dispatcher = self.trace_dispatch_i
elif hasattr(os, 'times'):
self.timer = os.times
self.dispatcher = self.trace_dispatch
self.get_time = _get_time_times
else:
self.timer = self.get_time = time.time
self.dispatcher = self.trace_dispatch_i
else:
self.timer = timer
t = self.timer() # test out timer function
try:
length = len(t)
except TypeError:
self.get_time = timer
self.dispatcher = self.trace_dispatch_i
else:
if length == 2:
self.dispatcher = self.trace_dispatch
else:
self.dispatcher = self.trace_dispatch_l
# This get_time() implementation needs to be defined
# here to capture the passed-in timer in the parameter
# list (for performance). Note that we can't assume
# the timer() result contains two values in all
# cases.
import operator
def get_time_timer(timer=timer,
reduce=reduce, reducer=operator.add):
return reduce(reducer, timer(), 0)
self.get_time = get_time_timer
self.t = self.get_time()
self.simulate_call('profiler')
# Heavily optimized dispatch routine for os.times() timer
def trace_dispatch(self, frame, event, arg):
timer = self.timer
t = timer()
t = t[0] + t[1] - self.t # No Calibration constant
# t = t[0] + t[1] - self.t - .00053 # Calibration constant
if self.dispatch[event](self, frame,t):
t = timer()
self.t = t[0] + t[1]
else:
r = timer()
self.t = r[0] + r[1] - t # put back unrecorded delta
return
# Dispatch routine for best timer program (return = scalar, fastest if
# an integer but float works too -- and time.clock() relies on that).
def trace_dispatch_i(self, frame, event, arg):
timer = self.timer
t = timer() - self.t # - 1 # calibration constant
if self.dispatch[event](self, frame,t):
self.t = timer()
else:
self.t = timer() - t # put back unrecorded delta
return
# Dispatch routine for macintosh (timer returns time in ticks of
# 1/60th second)
def trace_dispatch_mac(self, frame, event, arg):
timer = self.timer
t = timer()/60.0 - self.t # - 1 # Integer calibration constant
if self.dispatch[event](self, frame,t):
self.t = timer()/60.0
else:
self.t = timer()/60.0 - t # put back unrecorded delta
return
# SLOW generic dispatch routine for timer returning lists of numbers
def trace_dispatch_l(self, frame, event, arg):
get_time = self.get_time
t = get_time() - self.t
if self.dispatch[event](self, frame,t):
self.t = get_time()
else:
self.t = get_time() - t # put back unrecorded delta
return
def trace_dispatch_exception(self, frame, t):
rt, rtt, rct, rfn, rframe, rcur = self.cur
if (rframe is not frame) and rcur:
return self.trace_dispatch_return(rframe, t)
self.cur = rt, rtt+t, rct, rfn, rframe, rcur
return 1
def trace_dispatch_call(self, frame, t):
if self.cur and frame.f_back is not self.cur[-2]:
rt, rtt, rct, rfn, rframe, rcur = self.cur
if not isinstance(rframe, Profile.fake_frame):
if rframe.f_back is not frame.f_back:
print rframe, rframe.f_back
print frame, frame.f_back
raise "Bad call", self.cur[-3]
self.trace_dispatch_return(rframe, 0)
if self.cur and frame.f_back is not self.cur[-2]:
raise "Bad call[2]", self.cur[-3]
fcode = frame.f_code
fn = (fcode.co_filename, fcode.co_firstlineno, fcode.co_name)
self.cur = (t, 0, 0, fn, frame, self.cur)
timings = self.timings
if timings.has_key(fn):
cc, ns, tt, ct, callers = timings[fn]
timings[fn] = cc, ns + 1, tt, ct, callers
else:
timings[fn] = 0, 0, 0, 0, {}
return 1
def trace_dispatch_return(self, frame, t):
if frame is not self.cur[-2]:
if frame is self.cur[-2].f_back:
self.trace_dispatch_return(self.cur[-2], 0)
else:
raise "Bad return", self.cur[-3]
# Prefix "r" means part of the Returning or exiting frame
# Prefix "p" means part of the Previous or older frame
rt, rtt, rct, rfn, frame, rcur = self.cur
rtt = rtt + t
sft = rtt + rct
pt, ptt, pct, pfn, pframe, pcur = rcur
self.cur = pt, ptt+rt, pct+sft, pfn, pframe, pcur
timings = self.timings
cc, ns, tt, ct, callers = timings[rfn]
if not ns:
ct = ct + sft
cc = cc + 1
if callers.has_key(pfn):
callers[pfn] = callers[pfn] + 1 # hack: gather more
# stats such as the amount of time added to ct courtesy
# of this specific call, and the contribution to cc
# courtesy of this call.
else:
callers[pfn] = 1
timings[rfn] = cc, ns - 1, tt+rtt, ct, callers
return 1
dispatch = {
"call": trace_dispatch_call,
"exception": trace_dispatch_exception,
"return": trace_dispatch_return,
}
# The next few function play with self.cmd. By carefully preloading
# our parallel stack, we can force the profiled result to include
# an arbitrary string as the name of the calling function.
# We use self.cmd as that string, and the resulting stats look
# very nice :-).
def set_cmd(self, cmd):
if self.cur[-1]: return # already set
self.cmd = cmd
self.simulate_call(cmd)
class fake_code:
def __init__(self, filename, line, name):
self.co_filename = filename
self.co_line = line
self.co_name = name
self.co_firstlineno = 0
def __repr__(self):
return repr((self.co_filename, self.co_line, self.co_name))
class fake_frame:
def __init__(self, code, prior):
self.f_code = code
self.f_back = prior
def simulate_call(self, name):
code = self.fake_code('profile', 0, name)
if self.cur:
pframe = self.cur[-2]
else:
pframe = None
frame = self.fake_frame(code, pframe)
a = self.dispatch['call'](self, frame, 0)
return
# collect stats from pending stack, including getting final
# timings for self.cmd frame.
def simulate_cmd_complete(self):
get_time = self.get_time
t = get_time() - self.t
while self.cur[-1]:
# We *can* cause assertion errors here if
# dispatch_trace_return checks for a frame match!
a = self.dispatch['return'](self, self.cur[-2], t)
t = 0
self.t = get_time() - t
def print_stats(self):
import pstats
pstats.Stats(self).strip_dirs().sort_stats(-1). \
print_stats()
def dump_stats(self, file):
f = open(file, 'wb')
self.create_stats()
marshal.dump(self.stats, f)
f.close()
def create_stats(self):
self.simulate_cmd_complete()
self.snapshot_stats()
def snapshot_stats(self):
self.stats = {}
for func in self.timings.keys():
cc, ns, tt, ct, callers = self.timings[func]
callers = callers.copy()
nc = 0
for func_caller in callers.keys():
nc = nc + callers[func_caller]
self.stats[func] = cc, nc, tt, ct, callers
# The following two methods can be called by clients to use
# a profiler to profile a statement, given as a string.
def run(self, cmd):
import __main__
dict = __main__.__dict__
return self.runctx(cmd, dict, dict)
def runctx(self, cmd, globals, locals):
self.set_cmd(cmd)
sys.setprofile(self.dispatcher)
try:
exec cmd in globals, locals
finally:
sys.setprofile(None)
return self
# This method is more useful to profile a single function call.
def runcall(self, func, *args, **kw):
self.set_cmd(`func`)
sys.setprofile(self.dispatcher)
try:
return apply(func, args, kw)
finally:
sys.setprofile(None)
#******************************************************************
# The following calculates the overhead for using a profiler. The
# problem is that it takes a fair amount of time for the profiler
# to stop the stopwatch (from the time it receives an event).
# Similarly, there is a delay from the time that the profiler
# re-starts the stopwatch before the user's code really gets to
# continue. The following code tries to measure the difference on
# a per-event basis. The result can the be placed in the
# Profile.dispatch_event() routine for the given platform. Note
# that this difference is only significant if there are a lot of
# events, and relatively little user code per event. For example,
# code with small functions will typically benefit from having the
# profiler calibrated for the current platform. This *could* be
# done on the fly during init() time, but it is not worth the
# effort. Also note that if too large a value specified, then
# execution time on some functions will actually appear as a
# negative number. It is *normal* for some functions (with very
# low call counts) to have such negative stats, even if the
# calibration figure is "correct."
#
# One alternative to profile-time calibration adjustments (i.e.,
# adding in the magic little delta during each event) is to track
# more carefully the number of events (and cumulatively, the number
# of events during sub functions) that are seen. If this were
# done, then the arithmetic could be done after the fact (i.e., at
# display time). Currently, we track only call/return events.
# These values can be deduced by examining the callees and callers
# vectors for each functions. Hence we *can* almost correct the
# internal time figure at print time (note that we currently don't
# track exception event processing counts). Unfortunately, there
# is currently no similar information for cumulative sub-function
# time. It would not be hard to "get all this info" at profiler
# time. Specifically, we would have to extend the tuples to keep
# counts of this in each frame, and then extend the defs of timing
# tuples to include the significant two figures. I'm a bit fearful
# that this additional feature will slow the heavily optimized
# event/time ratio (i.e., the profiler would run slower, fur a very
# low "value added" feature.)
#
# Plugging in the calibration constant doesn't slow down the
# profiler very much, and the accuracy goes way up.
#**************************************************************
def calibrate(self, m):
# Modified by Tim Peters
get_time = self.get_time
n = m
s = get_time()
while n:
self.simple()
n = n - 1
f = get_time()
my_simple = f - s
#print "Simple =", my_simple,
n = m
s = get_time()
while n:
self.instrumented()
n = n - 1
f = get_time()
my_inst = f - s
# print "Instrumented =", my_inst
avg_cost = (my_inst - my_simple)/m
#print "Delta/call =", avg_cost, "(profiler fixup constant)"
return avg_cost
# simulate a program with no profiler activity
def simple(self):
a = 1
pass
# simulate a program with call/return event processing
def instrumented(self):
a = 1
self.profiler_simulation(a, a, a)
# simulate an event processing activity (from user's perspective)
def profiler_simulation(self, x, y, z):
t = self.timer()
## t = t[0] + t[1]
self.ut = t
#****************************************************************************
def Stats(*args):
print 'Report generating functions are in the "pstats" module\a'
# When invoked as main program, invoke the profiler on a script
if __name__ == '__main__':
if not sys.argv[1:]:
print "usage: profile.py scriptfile [arg] ..."
sys.exit(2)
filename = sys.argv[1] # Get script filename
del sys.argv[0] # Hide "profile.py" from argument list
# Insert script directory in front of module search path
sys.path.insert(0, os.path.dirname(filename))
run('execfile(' + `filename` + ')')
Reference in New Issue View Git Blame Copy Permalink