/slipcover

Near Zero-Overhead Python Code Coverage

Primary LanguagePythonApache License 2.0Apache-2.0

slipcover

SlipCover: Near Zero-Overhead Python Code Coverage

by Juan Altmayer Pizzorno and Emery Berger at UMass Amherst's PLASMA lab.

license pypi Downloads pyversions tests

About Slipcover

SlipCover is a fast code coverage tool. It tracks a Python program as it runs and reports on the parts that executed and those that didn't. That can help guide your testing (showing code that isn't being tested), debugging, fuzzing or to find "dead" code.

Past code coverage tools can make programs significantly slower; it is not uncommon for them to take twice as long to execute. SlipCover aims to provide the same information with near-zero overhead, often almost as fast as running the original Python program.

How it works

Previous coverage tools like Coverage.py rely on Python's tracing facilities, which add significant overhead. Instead, SlipCover uses just-in-time instrumentation and de-instrumentation. When SlipCover gathers coverage information, it modifies the program's Python byte codes, inserting instructions that let it keep track the lines executed by the program. As the program executes, SlipCover gradually removes instrumentation that is no longer needed, allowing those parts to run at full speed. Care is taken throughout SlipCover to keep things as efficient as possible. On Python 3.12, rather than rewrite bytecode, SlipCover uses the new sys.monitoring API to collect coverage information.

Performance

The first image on the right shows SlipCover's speedup, ranging from 1.1x to 3.4x, in relation to Coverage.py, running on CPython 3.10.5.

The first two benchmarks are the test suites for scikit-learn and Flask; "sudoku" runs Peter Norvig's Sudoku solver while the others were derived from the Python Benchmark Suite.

More "Python-intensive" programs such as sudoku and those from the benchmark suite (with a larger proportion of execution time spent in Python, rather than in native code) generate more tracing events, causing more overhead in Coverage.py. While each program's structure can affect SlipCover's ability to de-instrument, its running time stays relatively close to the original.

On PyPy 3.9, the speedup ranges from 2.1x to 104.9x. Since it is so high for some of the benchmarks, we plot it on a logarithmic scale (see the second image on the right).

In a proof-of-concept integration with a property-based testing package, SlipCover sped up coverage-based testing 22x.

Accuracy

We verified SlipCover's accuracy against Coverage.py and against a simple script of our own that collects coverage using Python tracing. We found SlipCover's results to be accurate, in fact, in certain cases more accurate.

Getting started

SlipCover is available from PyPI. You can install it like any other Python module with

pip3 install slipcover

You could then run your Python script with:

python3 -m slipcover myscript.py

Using it with a test harness

SlipCover can also execute a Python module, as in:

python3 -m slipcover -m pytest -x -v

which starts pytest, passing it any options (-x -v in this example) after the module name. No plug-in is required for pytest.

Usage example

$ python3 -m slipcover -m pytest
================================================================ test session starts ================================================================
platform darwin -- Python 3.9.12, pytest-7.1.2, pluggy-1.0.0
rootdir: /Users/juan/project/wally/d2k-5, configfile: pytest.ini
plugins: hypothesis-6.39.3, mock-3.7.0, repeat-0.9.1, doctestplus-0.12.0, arraydiff-0.5.0
collected 439 items                                                                                                                                 

tests/box_test.py .........................                                                                                                   [  5%]
tests/image_test.py ...............                                                                                                           [  9%]
tests/network_equivalence_test.py .........................................s................................................................. [ 33%]
..............................................................................                                                                [ 51%]
tests/network_test.py ....................................................................................................................... [ 78%]
...............................................................................................                                               [100%]

=================================================== 438 passed, 1 skipped, 62 warnings in 48.43s ====================================================

File                                 #lines    #miss    Cover%  Lines missing
---------------------------------  --------  -------  --------  ------------------------
d2k/__init__.py                           3        0       100
d2k/box.py                              105       27        74  73, 142-181
d2k/image.py                             38        4        89  70-73
d2k/network.py                          359        1        99  236
tests/box_test.py                       178        0       100
tests/darknet.py                        132       11        91  146, 179-191
tests/image_test.py                      45        0       100
tests/network_equivalence_test.py       304       30        90  63, 68, 191-215, 455-465
tests/network_test.py                   453        0       100
$ 

As can be seen in the coverage report, d2k lacks some coverage, especially in its box.py and image.py components.

Platforms

Our GitHub workflows run the automated test suite on Linux, MacOS and Windows, but really it should work anywhere where CPython/PyPy does.

Contributing

SlipCover is under active development; contributions are welcome! Please also feel free to create a new issue with any suggestions or issues you may encounter.

Technical Information

For more details about how SlipCover works please see the following paper, published at ISSTA'23: SlipCover: Near Zero-Overhead Code Coverage for Python.

Acknowledgements

Logo design by Sophia Berger.

This material is based upon work supported by the National Science Foundation under Grant No. 1955610. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.