`force_print` argument for output behaves strange

[danlkv]

trafficstars

https://github.com/plasma-umass/scalene/blob/6980cb0c41df6ae724a2ce8d4686398b4b058853/scalene/scalene_json.py#L109C12-L109C15

Shouldn't it be if force_print and not ...? The printing will happen if force_print=False.

[emeryberger]

I agree that this looks unintuitive but that line of code has been in the codebase for 2 years. Can you point to a bug here?

[danlkv]

I think the issue is that there seems to be no way to output only the functions I care about (but maybe there's another way?), only the full file. The output_profile_line never returns a False, which in turn makes output print every line.

[emeryberger]

So I see that there is some code that is always executing now that output_profile_line returns a "null" entry instead of False (which led to some issues, addressed by https://github.com/plasma-umass/scalene/commit/cadb8f463e3fc11d57b20795743c40af5e5d0b69). Are you using @profile to select the functions you care about? You should definitely not be using Scalene's internal functions.

[danlkv]

I see. No, I use scalene_profiler.Scalene.profile to wrap the function in runtime, since adding and removing @profile everywhere is a hassle:) I was under impression that using @profile and Scalene.profile should give the same results. It seems that there is some other mechanism for filtering lines in addition to profile_this_code.

[emeryberger]

It's the same mechanism, but it was never intended to be called directly. I like the idea of listing a bunch of functions to profile without selecting them one by one. Can you give me an example of how you are doing this? Might make sense to expose an API.

[danlkv]

The usecase is that I have profile.py script that imports a library that I want to profile. It's simply calling Scalene.profile on whatever function I want to profile. If I run scalene profile.py it works, but requires the library files to be within working directory tree (can workaround with --profile-only).

  for func in [mylib.func1, mylib.Class1.method1]:
      scalene_profiler.Scalene.profile(func)
  logger.log(10, f"{scalene_profiler.Scalene.__dict__['_Scalene__functions_to_profile']=}")
  scalene_profiler.Scalene.start()
  func_to_run(*args, **kwargs)
  scalene_profiler.Scalene.stop()

Unrelated, but something that I found useful is to start scalene from within a python script with pre-configured parameters. This removes the need to know or remember to run the script as scalene [params] profile.py: you can run python profile.py and get the same result. Example