4 ms·

Good observation. My personal POV is that the best way to optimize your Python code is to use native code (whether as libraries or through pure Python code that

Good observation. My personal POV is that the best way to optimize your Python code is to use native code (whether as libraries or through pure Python code that is essentially a thin wrapper over C) rather than living in the interpreter. I want to see the parts of the program that are spending that time in the interpreter. In short, a profiler that tells me that a program is spending a lot of time in C is not generally providing me particularly actionable information. (In any event, the top-line report is that the Python part of the program only accounts for 33.66% of the execution time, which looks just about right.)

bawr_hszm 7y ago
I can understand this approach, but I fundamentally disagree with it - the first duty of a profiling tool is not to mislead the user. In this example, the program spends a third of the time just sleeping / blocked, a third of the time on CPU but at the C level, and the remainder just evaluating Python loops. Unless you already know how it's implemented, that 33.66% is not easy to interpret, and the docs don't mention what exactly is being profiled. Specifically, these samples aren't a % of CPU time, they're a % of real time that we happened to have been able to do a Python-level interrupt, and even that isn't a great explanation. I think most users would still expect line 22 to get traced properly. I very much do want to know about C time, too, because it's very much actionable for most of the optimizations I end up making in production systems. That said, I don't think this line of discussion is super productive for either of us, we seem to have different goals in mind, which is fine. ;) So I'll close by saying that I was impressed by your LD_PRELOAD hacks for memory profiling, which isn't an approach that I've ever seen in other Python profilers.
- emeryberger 7y ago
  Re: being impressed by the memory profiling hacks, thanks! :) And you persuaded me. New version to be uploaded momentarily that attributes time to line 22. % python3 -m scalene sample-sample.py f1 1.8081 312499987500000 f2 1.7910 2812499962500000 f3 1.5034 1.5 sample-sample.py: % of CPU time = 70.53% out of 5.10s. Line | CPU % | [sample-sample.py] 1 | | #!/usr/bin/env python 2 | | 3 | | import time 4 | | 5 | | def timed(f): 6 | | def f_timed(*args, **kwargs): 7 | | t = time.time() 8 | | r = f(*args, **kwargs) 9 | 0.32% | t = time.time() - t 10 | | print('%s %.4f %s' % (f.__name__, t, r)) 11 | | return f_timed 12 | | 13 | | @timed 14 | | def f1(n): 15 | | s = 0 16 | 8.58% | for i in range(n): 17 | 41.34% | s += i 18 | | return s 19 | | 20 | | @timed 21 | | def f2(n): 22 | 49.76% | return sum(range(n)) 23 | | 24 | | @timed 25 | | def f3(t): 26 | | time.sleep(t) 27 | | return t 28 | | 29 | | if __name__ == '__main__': 30 | | f1(25000000) 31 | | f2(75000000) 32 | | f3(1.5)
  - emeryberger 7y ago
    This discussion inspired me to refine Scalene's output further. Scalene now separates out CPU time spent in Python from time spent in C (that is, not executing in the interpreter). I think this is the best of both worlds: all CPU time is accounted for, and it's easy for users to see what code is running where. Thanks! sample-sample.py: % of CPU time = 99.58% out of 3.34s. | CPU % | CPU % | Line | (Python) | (C) | [sample-sample.py] -------------------------------------------------------------------------------- 1 | | | #!/usr/bin/env python 2 | | | 3 | | | import time 4 | | | 5 | | | def timed(f): 6 | | | def f_timed(*args, **kwargs): 7 | | | t = time.time() 8 | | | r = f(*args, **kwargs) 9 | | | t = time.time() - t 10 | | | print('%s %.4f %s' % (f.__name__, t, r)) 11 | | | return f_timed 12 | | | 13 | | | @timed 14 | | | def f1(n): 15 | | | s = 0 16 | 6.61% | 0.41% | for i in range(n): 17 | 35.17% | 3.97% | s += i 18 | | | return s 19 | | | 20 | | | @timed 21 | | | def f2(n): 22 | 0.30% | 53.53% | return sum(range(n)) 23 | | | 24 | | | @timed 25 | | | def f3(t): 26 | | | time.sleep(t) 27 | | | return t 28 | | | 29 | | | if __name__ == '__main__': 30 | | | f1(25000000) 31 | | | f2(75000000) 32 | | | f3(1.5)
    - gshulegaard 7y ago
      Just wanted to take a moment to appreciate the awesome response to feedback and new added feature.
      - kinow 7y ago

bawr_hszm 7y ago

I can understand this approach, but I fundamentally disagree with it - the first duty of a profiling tool is not to mislead the user. In this example, the program spends a third of the time just sleeping / blocked, a third of the time on CPU but at the C level, and the remainder just evaluating Python loops. Unless you already know how it's implemented, that 33.66% is not easy to interpret, and the docs don't mention what exactly is being profiled. Specifically, these samples aren't a % of CPU time, they're a % of real time that we happened to have been able to do a Python-level interrupt, and even that isn't a great explanation. I think most users would still expect line 22 to get traced properly. I very much do want to know about C time, too, because it's very much actionable for most of the optimizations I end up making in production systems. That said, I don't think this line of discussion is super productive for either of us, we seem to have different goals in mind, which is fine. ;) So I'll close by saying that I was impressed by your LD_PRELOAD hacks for memory profiling, which isn't an approach that I've ever seen in other Python profilers.

emeryberger 7y ago
Re: being impressed by the memory profiling hacks, thanks! :) And you persuaded me. New version to be uploaded momentarily that attributes time to line 22. % python3 -m scalene sample-sample.py f1 1.8081 312499987500000 f2 1.7910 2812499962500000 f3 1.5034 1.5 sample-sample.py: % of CPU time = 70.53% out of 5.10s. Line | CPU % | [sample-sample.py] 1 | | #!/usr/bin/env python 2 | | 3 | | import time 4 | | 5 | | def timed(f): 6 | | def f_timed(*args, **kwargs): 7 | | t = time.time() 8 | | r = f(*args, **kwargs) 9 | 0.32% | t = time.time() - t 10 | | print('%s %.4f %s' % (f.__name__, t, r)) 11 | | return f_timed 12 | | 13 | | @timed 14 | | def f1(n): 15 | | s = 0 16 | 8.58% | for i in range(n): 17 | 41.34% | s += i 18 | | return s 19 | | 20 | | @timed 21 | | def f2(n): 22 | 49.76% | return sum(range(n)) 23 | | 24 | | @timed 25 | | def f3(t): 26 | | time.sleep(t) 27 | | return t 28 | | 29 | | if __name__ == '__main__': 30 | | f1(25000000) 31 | | f2(75000000) 32 | | f3(1.5)
- emeryberger 7y ago
  This discussion inspired me to refine Scalene's output further. Scalene now separates out CPU time spent in Python from time spent in C (that is, not executing in the interpreter). I think this is the best of both worlds: all CPU time is accounted for, and it's easy for users to see what code is running where. Thanks! sample-sample.py: % of CPU time = 99.58% out of 3.34s. | CPU % | CPU % | Line | (Python) | (C) | [sample-sample.py] -------------------------------------------------------------------------------- 1 | | | #!/usr/bin/env python 2 | | | 3 | | | import time 4 | | | 5 | | | def timed(f): 6 | | | def f_timed(*args, **kwargs): 7 | | | t = time.time() 8 | | | r = f(*args, **kwargs) 9 | | | t = time.time() - t 10 | | | print('%s %.4f %s' % (f.__name__, t, r)) 11 | | | return f_timed 12 | | | 13 | | | @timed 14 | | | def f1(n): 15 | | | s = 0 16 | 6.61% | 0.41% | for i in range(n): 17 | 35.17% | 3.97% | s += i 18 | | | return s 19 | | | 20 | | | @timed 21 | | | def f2(n): 22 | 0.30% | 53.53% | return sum(range(n)) 23 | | | 24 | | | @timed 25 | | | def f3(t): 26 | | | time.sleep(t) 27 | | | return t 28 | | | 29 | | | if __name__ == '__main__': 30 | | | f1(25000000) 31 | | | f2(75000000) 32 | | | f3(1.5)
  - gshulegaard 7y ago
    Just wanted to take a moment to appreciate the awesome response to feedback and new added feature.
    - kinow 7y ago