Skip to main content

A debugging and profiling tool that can trace and visualize python code

Project description


build pypi

CodeSnap is a deterministic debugging/profiling tool that can trace and visualize python code. The major data CodeSnap displays is FEE(function entry/exit), or equivalently, the call stack.

Unlike traditional flame graph, which is normally generated by sampling profiler, CodeSnap can display every function executed and the corresponding entry/exit time from the beginning of the program to the end, which is helpful for programmers to catch sporatic performance issues.

With CodeSnap, the programmer can intuitively understand what their code is doing and how long each function takes.

You can take a look at the demo result of an example program running recursive merge and quick sort algorithm.

trace viewer is used to display the stand alone html data.

CodeSnap also supports json output that complies with Chrome trace event format, which can be loaded using perfetto


CodeSnap requires python 3.5+. No other package is needed. For now, CodeSnap binary on pip only supports CPython + Linux. However, in theory the source code can build on Windows/MacOS.


The prefered way to install CodeSnap is via pip

pip install codesnap

You can also download the source code and build it yourself.


There are a couple ways to use CodeSnap

Command Line

The easiest way to use CodeSnap it through command line. Assume you have a python script to profile and the normal way to run it is:


You can simply use CodeSnap as

python3 -m codesnap

which will generate a result.html file in the directory you run this command. Open it in browser and there's your result.

If your script needs arguments like

python3 arg1 arg2

Just feed it as it is to CodeSnap

python3 -m codesnap arg1 arg2

You can also specify the tracer to be used in command line by passing --tracer argument. c tracer is the default value, you can use python tracer instead

python3 -m codesnap --tracer c
python3 -m codesnap --tracer python

You can specify the output file using -o or --output_file argument. The default output file is result.html. Two types of files are supported, html and json.

python3 -m codesnap -o other_name.html
python3 -m codesnap -o other_name.json


Sometimes the command line may not work as you expected, or you do not want to profile the whole script. You can manually start/stop the profiling in your script as well.

First of all, you need to import CodeSnap class from the package, and make an object of it.

from codesnap import CodeSnap

snap = CodeSnap()

If your code is executable by exec function, you can simply call"import random;random.randrange(10)")

This will as well generate a result.html file in your current directory. You can pass other file path to the function if you do not like the name result.html"import random; random.randrange(10)", output_file = "better_name.html")

When you need a more delicate profiler, you can manually enable/disable the profile using start() and stop() function.

# Something happens here
snap.stop() # also takes output_file as an optional argument

With this method, you can only record the part that you are interested in

# Some code that I don't care
# Some code I do care
# Some code that I want to skip
# Important code again

It is higly recommended that start() and stop() function should be in the same frame(same level on call stack). Problem might happen if the condition is not met

Display Result

By default, CodeSnap will generate a stand alone HTML file which you can simply open with Chrome(maybe Firefox?). The front-end uses trace-viewer to show all the data.

However, you can generate json file as well, which complies to the chrome trace event format. You can load the json file on perfetto, which will replace the deprecated trace viewer in the future.

At the moment, perfetto did not support locally stand alone HTML file generation, so I'm not able to switch completely to it. The good news is that once you load the perfetto page, you can use it even when you are offline.

Choose Tracer

The default tracer for current version is c tracer, which introduce a relatively small overhead(worst case 2-3x) but only works for CPython on Linux. However, if there's other reason that you would prefer a pure-python tracer, you can use python tracer using tracer argument when you initialize CodeSnap object.

snap = CodeSnap(tracer="python")

Cleanup of c Tracer

The interface for c trace is almost exactly the same as python tracer, except for the fact that c tracer does not support command line run now. However, to achieve lower overhead, some optimization is applied to c tracer so it will withhold the memory it allocates for future use to reduce the time it calls malloc(). If you want the c trace to free all the memory it allocates while collecting trace, use



Overhead is a big consideration when people choose profilers. CodeSnap now has a similar overhead as native cProfiler. It works slightly worse in the worst case(Pure FEE) and better in easier case because even though it collects some extra information than cProfiler, the structure is lighter.

Admittedly, CodeSnap is only focusing on FEE now, so cProfiler also gets other information that CodeSnap does not acquire.

An example run for test_performance with Python 3.8 / Ubuntu 18.04.4 on Github VM

fib       (10336, 10336): 0.000852800 vs 0.013735200(16.11)[py] vs 0.001585900(1.86)[c] vs 0.001628400(1.91)[cProfile]
hanoi     (8192, 8192): 0.000621400 vs 0.012924899(20.80)[py] vs 0.001801800(2.90)[c] vs 0.001292900(2.08)[cProfile]
qsort     (10586, 10676): 0.003457500 vs 0.042572898(12.31)[py] vs 0.005594100(1.62)[c] vs 0.007573200(2.19)[cProfile]
slow_fib  (1508, 1508): 0.033606299 vs 0.038840998(1.16)[py] vs 0.033270399(0.99)[c] vs 0.032577599(0.97)[cProfile]


CodeSnap uses sys.setprofile() for its profiler capabilities, so it will conflict with other profiling tools which also use this function. Be aware of it when using CodeSnap.


Please send bug reports and feature requirements through github issue tracker. CodeSnap is currently under development now and it's open to any constructive suggestions.


Copyright Tian Gao, 2020.

Distributed under the terms of the Apache 2.0 license.

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distributions

No source distribution files available for this release. See tutorial on generating distribution archives.

Built Distributions

Supported by

AWS AWS Cloud computing Datadog Datadog Monitoring Facebook / Instagram Facebook / Instagram PSF Sponsor Fastly Fastly CDN Google Google Object Storage and Download Analytics Huawei Huawei PSF Sponsor Microsoft Microsoft PSF Sponsor NVIDIA NVIDIA PSF Sponsor Pingdom Pingdom Monitoring Salesforce Salesforce PSF Sponsor Sentry Sentry Error logging StatusPage StatusPage Status page