SlipCover: Near Zero-Overhead Python Code Coverage
by Juan Altmayer Pizzorno and Emery Berger
at UMass Amherst's PLASMA lab.
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.
Configuration via pyproject.toml
Instead of passing options on every command invocation, you can store them
in your project's pyproject.toml under the [tool.slipcover] section.
SlipCover automatically discovers the nearest pyproject.toml by walking up
from the current working directory.
[tool.slipcover]
branch = true
source = "src" # or ["src", "lib"]
omit = "tests/*" # or ["tests/*", "*.pyc"]
fail-under = 80.0
json = true
xml = false
pretty-print = true
skip-covered = true
immediate = false
out = "coverage.json"
threshold = 75
missing-width = 120
xml-package-depth = 3
Most command-line flags have a matching key (use hyphens, as shown above);
source and omit also accept a TOML array instead of a single
comma-separated string. --merge, -m/module, the script argument, --version,
and --help are per-invocation choices rather than settings, so they aren't
configurable this way. Command-line arguments always take precedence over
values in pyproject.toml, so you can override any setting on a per-run basis.
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.
