homepage

A patch including tests is in attachment.
Example usage:
>>> from cProfile import profile
>>>
>>> @profile
... def factorial(n):
... n = abs(int(n))
... if n < 1: 
... n = 1
... x = 1
... for i in range(1, n+1):
... x = i * x
... return x
... 
>>>
>>> factorial(5)
 Thu Jul 15 20:58:21 2010 /tmp/tmpIDejr5
 4 function calls in 0.000 CPU seconds
 Ordered by: internal time, call count
 ncalls tottime percall cumtime percall filename:lineno(function)
 1 0.000 0.000 0.000 0.000 profiler.py:120(factorial)
 1 0.000 0.000 0.000 0.000 {range}
 1 0.000 0.000 0.000 0.000 {abs}
 1 0.000 0.000 0.000 0.000 {method 'disable' of '_lsprof.Profiler' objects}
120
>>>
Before including doc changes I'd like to hear whether this is actually desirable or not.

+1

I have been working on a similar idea. I just wanted to raise an issue I ran into which might affect this code also. The decorated function ended up with a different function signature to the original inner function. This can be important sometimes. I had to use the decorator.decorator module to give the outer function the same signature as the inner function.

See #8916 for adding standard functionality similar to the decorator module.

Brief comments:
- Please don't call it profile -- we already have a module by that name.
- Please make it so that both the decorator and context manager can specify a file where to dump the raw data -- basically it needs to have the same functionality as the functions run()/runctx()/runcall() (the latter TBD, see issue 17130).
- Please also make Profile object itself the context manager -- all you have to do is add __enter__() and __exit__() that call enable() and disable(). (But this doesn't completely replace the global function, which has more functionality -- it prints the profile or dumps the data).

Ok, will look into this soon.

A preliminary patch for cProfile.py is in attachment. Will make changes to profile.py later.

Ok, here's an updated patch modeled after:
http://hg.python.org/cpython/rev/422169310b7c
It works fine with cProfile.py but not with profile.py where I get this exception when I try to use the context manager (tests can be run in order to reproduce it):
 File "/home/giampaolo/svn/python/3.4-profile/Lib/profile.py", line 339, in trace_dispatch_return
 assert frame is self.cur[-2].f_back, ("Bad return", self.cur[-3])
AssertionError: ('Bad return', ('profile', 0, ''))
I have no clue what this error means.
I wasn't able to add a context manager for profile.Profile for the same reason.
Any clue?

Ping. Is there any reason why this patch doesn't accept?

The original patch is basically blocked by the cryptic assertion error reported above. It's not clear what it means or how to work around it.

giampaolo: it seems that contextmanager will somehow make a bad return in `trace_dispatch_return`:
$ ./python tests.py
# logging.fatal('%r %r' % (frame.f_code.co_name, self.cur[-3]))
CRITICAL:root:'runblock' '' ('profile', 0, '')
CRITICAL:root:'__enter__' '' ('profile', 0, '')
Traceback (most recent call last):
 File "tests.py", line 18, in <module>
 with p.runblock():
 File "/home/grd/Python/cpython/Lib/contextlib.py", line 82, in __enter__
 return next(self.gen)
 File "/home/grd/Python/cpython/Lib/profile.py", line 256, in trace_dispatch_i
 if self.dispatch[event](self, frame, t):
 File "/home/grd/Python/cpython/Lib/profile.py", line 346, in trace_dispatch_return
 raise AssertionError("Bad return", self.cur[-3])
AssertionError: ('Bad return', ('profile', 0, ''))
I make a workaround in GitHub PR, that skip the assert when self.cur[-3] is ('profile', 0, ''), and this work fine with your test cases.
p.s. I'm not very sure about the assertion of bad return, maybe this workaround may cause some side affact I didn't notice.

Original patch still applies. Not sure if we should continue with that one or with your new PR. The original assertion error is still there. I CCed Tim Peters as it appears he's the one who originally added it in 2001 - maybe he has some clue.

Giampaolo, the assertion is still worked good, and no need to remove them. The assertion is to prevent dispatch return too more, to return upper then when the profiler was created.
The problem why profile __enter__ can't work, is because it misses the simulate_call between __enter__ and upper frame.
The original scenes:
pr = profile.Profile() # It will call simulate_call at the end of init
sys.setprofile(pr.dispatcher)
# profile
sys.setprofile(None)
The break scenes:
def profile_helper(pr):
 sys.setprofile(pr.dispatcher)
 # Function will return None, dead here
pr = profile.Profile() # Create simulate_call
# We go into profile_helper, but didn't simulate a call!! (didn't setprofile yet)
profile_helper(pr) 
sys.setprofile(None)
The #30113 issue fix this:
def profile_helper(pr):
 pr._adjust_frame() # call simuate_call here
 sys.setprofile(pr.dispatcher)
pr = profile.Profile() # Create simulate_call
profile_helper(pr) 
sys.setprofile(None)
Creating this simuate_call, then profiler can go back to the upper frame without error.