[Python-ideas] Small improvements to the profile/cProfile API

[Tim Mitchell]

Hi Ben,

Mostly I just print to stdout, I imagine more flexibility would be needed
in general.

This is for python 2.7 - don't know if it works for 3.


def profile(sort='time', restriction=(), callers=None, callees=None,
filename=None):
    def _profileDecorator(func):
        "print profile stats for decorated function"
        def wrapper(*args, **kwargs):
            print 'Profile for:', func.__name__

            prof = cProfile.Profile()
            result = prof.runcall(func, *args, **kwargs)
            _, statsFileName = tempfile.mkstemp()
            prof.dump_stats(statsFileName)
            if filename is None:
                stats = pstats.Stats(statsFileName)
            else:
                stats = pstats.Stats(statsFileName, stream=open(filename, 'w'))
            if isinstance(sort, basestring):
                stats.sort_stats(sort)
            else:
                stats.sort_stats(*sort)
            if isinstance(restriction, (tuple, list)):
                stats.print_stats(*restriction)
            else:
                stats.print_stats(restriction)
            if callers is not None:
                if isinstance(callers, basestring):
                    stats.print_callers(callers)
                else:
                    stats.print_callers(*callers)
            if callees is not None:
                if isinstance(callees, basestring):
                    stats.print_callees(callees)
                else:
                    stats.print_callees(*callees)
            return result
        return wrapper
    return _profileDecorator

Cheers

Tim


On 3 November 2016 at 09:58, Ben Hoyt <benhoyt at gmail.com> wrote:

> Okay, got it, that sounds fair enough. With your @profile decorator how do
> you tell it when and where to print the output? Can you post the source for
> your decorator?
>
> On Wed, Nov 2, 2016 at 4:52 PM, Tim Mitchell <tim.mitchell at leapfrog3d.com>
> wrote:
>
>> I use an @profile() decorator for almost all my profiling.  If you want
>> to profile function foo you just decorate it and re-run the program.
>> With a with block you have to find the places where foo is called and put
>> with statements around the calls.
>> I think both approaches are equally valid and useful.
>>
>>> conduct/ <http://python.org/psf/codeofconduct/>
>>>
>>
>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/python-ideas/attachments/20161103/92571e37/attachment.html>