I was very dissatisfied with the output of existing Python profilers, as they aggregate all calls to the same function (losing context information), and output a big flat list of functions that I find very hard to use in order to profile a program. I believed a better approach is to view the program as a tree of function calls - so I started a small project we call pytracer. pytracer lets you run a program similarly to cProfile (Currently, the performance hit is worse than that of cProfile, but that can probably be improved), while creating a profile.out result file. Then, a Gtk+ based viewer can be used to view the profile.out, exposing an expandable tree of function invocations - each with timing information (real, user, system time). You can find the project at: http://code.google.com/p/pythontracer And the downloads at: http://code.google.com/p/pythontracer/downloads/list Project authors: Eyal Lotem Noam Lewis Simon Yoffe
participants (1)
-
Eyal Lotem