Vulture - Find dead code

Vulture finds unused code in Python programs. This is useful for cleaning up and finding errors in large code bases. If you run Vulture on both your library and test suite you can find untested code.

Due to Python's dynamic nature, static code analyzers like Vulture are likely to miss some dead code. Also, code that is only called implicitly may be reported as unused. Nonetheless, Vulture can be a very helpful tool for higher code quality.

Features

fast: uses static code analysis
tested: tests itself and has complete test coverage
complements pyflakes and has the same output syntax
sorts unused classes and functions by size with --sort-by-size

Installation

$ pip install vulture

Usage

$ vulture myscript.py  # or
$ python3 -m vulture myscript.py
$ vulture myscript.py mypackage/
$ vulture myscript.py --min-confidence 100  # Only report 100% dead code.

The provided arguments may be Python files or directories. For each directory Vulture analyzes all contained *.py files.

After you have found and deleted dead code, run Vulture again, because it may discover more dead code.

Types of unused code

In addition to finding unused functions, classes, etc., Vulture can detect unreachable code. Each chunk of dead code is assigned a confidence value between 60% and 100%, where a value of 100% signals that it is certain that the code won't be executed. Values below 100% are very rough estimates (based on the type of code chunk) for how likely it is that the code is unused.

Code type	Confidence value
function/method/class argument, unreachable code	100%
import	90%
attribute, class, function, method, property, variable	60%

You can use the --min-confidence flag to set the minimum confidence for code to be reported as unused. Use --min-confidence 100 to only report code that is guaranteed to be unused within the analyzed files.

Handling false positives

When Vulture incorrectly reports chunks of code as unused, you have several options for suppressing the false positives. If fixing your false positives could benefit other users as well, please file an issue report.

Whitelists

The recommended option is to add used code that is reported as unused to a Python module and add it to the list of scanned paths. To obtain such a whitelist automatically, pass --make-whitelist to Vulture:

$ vulture mydir --make-whitelist > whitelist.py
$ vulture mydir whitelist.py

Note that the resulting whitelist.py file will contain valid Python syntax, but for Python to be able to run it, you will usually have to make some modifications.

We collect whitelists for common Python modules and packages in vulture/whitelists/ (pull requests are welcome).

Ignoring files

If you want to ignore a whole file or directory, use the --exclude parameter (e.g., --exclude "*settings.py,*/docs/*.py,*/test_*.py,*/.venv/*.py"). The exclude patterns are matched against absolute paths.

Flake8 noqa comments

For compatibility with flake8, Vulture supports the F401 and F841 error codes for ignoring unused imports (# noqa: F401) and unused local variables (# noqa: F841). However, we recommend using whitelists instead of noqa comments, since noqa comments add visual noise to the code and make it harder to read.

Ignoring names

You can use --ignore-names foo*,ba[rz] to let Vulture ignore all names starting with foo and the names bar and baz. Additionally, the --ignore-decorators option can be used to ignore the names of functions decorated with the given decorator (but not their arguments or function body). This is helpful for example in Flask projects, where you can use --ignore-decorators "@app.route" to ignore all function names with the @app.route decorator. Note that Vulture simplifies decorators it cannot parse: @foo.bar(x, y) becomes "@foo.bar" and @foo.bar(x, y).baz becomes "@" internally.

We recommend using whitelists instead of --ignore-names or --ignore-decorators whenever possible, since whitelists are automatically checked for syntactic correctness when passed to Vulture and often you can even pass them to your Python interpreter and let it check that all whitelisted code actually still exists in your project.

Marking unused variables

There are situations where you can't just remove unused variables, e.g., in function signatures. The recommended solution is to use the del keyword as described in the PyLint manual and on StackOverflow:

def foo(x, y):
    del y
    return x + 3

Vulture will also ignore all variables that start with an underscore, so you can use _x, y = get_pos() to mark unused tuple assignments or function arguments, e.g., def foo(x, _y).

Minimum confidence

Raise the minimum confidence value with the --min-confidence flag.

Unreachable code

If Vulture complains about code like if False:, you can use a Boolean flag debug = False and write if debug: instead. This makes the code more readable and silences Vulture.

Forward references for type annotations

See #216. For example, instead of def foo(arg: "Sequence"): ..., we recommend using

from __future__ import annotations

def foo(arg: Sequence):
    ...

Configuration

You can also store command line arguments in pyproject.toml under the tool.vulture section. Simply remove leading dashes and replace all remaining dashes with underscores.

Options given on the command line have precedence over options in pyproject.toml.

Example Config:

[tool.vulture]
exclude = ["*file*.py", "dir/"]
ignore_decorators = ["@app.route", "@require_*"]
ignore_names = ["visit_*", "do_*"]
make_whitelist = true
min_confidence = 80
paths = ["myscript.py", "mydir", "whitelist.py"]
sort_by_size = true
verbose = true

Vulture will automatically look for a pyproject.toml in the current working directory.

To use a pyproject.toml in another directory, you can use the --config path/to/pyproject.toml flag.

Integrations

You can use a pre-commit hook to run Vulture before each commit. For this, install pre-commit and add the following to the .pre-commit-config.yaml file in your repository:

repos:
  - repo: https://github.com/jendrikseipp/vulture
    rev: 'v2.3'  # or any later Vulture version
    hooks:
      - id: vulture

Then run pre-commit install. Finally, create a pyproject.toml file in your repository and specify all files that Vulture should check under [tool.vulture] --> paths (see above).

There's also a GitHub Action for Vulture and you can use Vulture programatically. For example:

import vulture

v = vulture.Vulture()
v.scavenge(['.'])
unused_code = v.get_unused_code()  # returns a list of `Item` objects

How does it work?

Vulture uses the ast module to build abstract syntax trees for all given files. While traversing all syntax trees it records the names of defined and used objects. Afterwards, it reports the objects which have been defined, but not used. This analysis ignores scopes and only takes object names into account.

Vulture also detects unreachable code by looking for code after return, break, continue and raise statements, and by searching for unsatisfiable if- and while-conditions.

Sort by size

When using the --sort-by-size option, Vulture sorts unused code by its number of lines. This helps developers prioritize where to look for dead code first.

Examples

Consider the following Python script (dead_code.py):

import os

class Greeter:
    def greet(self):
        print("Hi")

def hello_world():
    message = "Hello, world!"
    greeter = Greeter()
    func_name = "greet"
    greet_func = getattr(greeter, func_name)
    greet_func()

if __name__ == "__main__":
    hello_world()

Calling :

$ vulture dead_code.py

results in the following output:

dead_code.py:1: unused import 'os' (90% confidence)
dead_code.py:4: unused function 'greet' (60% confidence)
dead_code.py:8: unused variable 'message' (60% confidence)

Vulture correctly reports os and message as unused but it fails to detect that greet is actually used. The recommended method to deal with false positives like this is to create a whitelist Python file.

Preparing whitelists

In a whitelist we simulate the usage of variables, attributes, etc. For the program above, a whitelist could look as follows:

# whitelist_dead_code.py
from dead_code import Greeter
Greeter.greet

Alternatively, you can pass --make-whitelist to Vulture and obtain an automatically generated whitelist.

Passing both the original program and the whitelist to Vulture

$ vulture dead_code.py whitelist_dead_code.py

makes Vulture ignore the greet method:

dead_code.py:1: unused import 'os' (90% confidence)
dead_code.py:8: unused variable 'message' (60% confidence)

Exit codes

Exit code	Description
0	No dead code found
1	Invalid input (file missing, syntax error, wrong encoding)
2	Invalid command line arguments
3	Dead code found

Similar programs

pyflakes finds unused imports and unused local variables (in addition to many other programmatic errors).
coverage finds unused code more reliably than Vulture, but requires all branches of the code to actually be run.
uncalled finds dead code by using the abstract syntax tree (like Vulture), regular expressions, or both.
dead finds dead code by using the abstract syntax tree (like Vulture).

Participate

Please visit https://github.com/jendrikseipp/vulture to report any issues or to make pull requests.

Contributing guide: CONTRIBUTING.md
Release notes: CHANGELOG.md
Roadmap: TODO.md

2.14 (2024-12-08)

Improve reachability analysis (kreathon, #270, #302).
Add type hints for get_unused_code and the fields of the Item class (John Doknjas, #361).

2.13 (2024-10-02)

Add support for Python 3.13 (Jendrik Seipp, #369).
Add PyPI and conda-forge badges to README file (Trevor James Smith, #356).
Include tests/**/*.toml in sdist (Colin Watson).

2.12 (2024-09-17)

Use ruff for linting and formatting (Anh Trinh, #347, #349).
Replace tox by pre-commit for linting and formatting (Anh Trinh, #349).
Add --config flag to specify path to pyproject.toml configuration file (Glen Robertson, #352).

2.11 (2024-01-06)

Switch to tomllib/tomli to support heterogeneous arrays (Sebastian Csar, #340).
Bump flake8, flake8-comprehensions and flake8-bugbear (Sebastian Csar, #341).
Provide whitelist parity for MagicMock and Mock (maxrake, #342).

2.10 (2023-10-06)

Drop support for Python 3.7 (Jendrik Seipp, #323).
Add support for Python 3.12 (Jendrik Seipp, #332).
Use end_lineno AST attribute to obtain more accurate line counts (Jendrik Seipp).

2.9.1 (2023-08-21)

Use exit code 0 for --help and --version again (Jendrik Seipp, #321).

2.9 (2023-08-20)

Use exit code 3 when dead code is found (whosayn, #319).
Treat non-supported decorator names as "@" instead of crashing (Llandy3d and Jendrik Seipp, #284).
Drop support for Python 3.6 (Jendrik Seipp).

2.8 (2023-08-10)

Add UnicodeEncodeError exception handling to core.py (milanbalazs, #299).
Add whitelist for Enum attributes _name_ and _value_ (Eugene Toder, #305).
Run tests and add PyPI trove for Python 3.11 (Jendrik Seipp).

2.7 (2023-01-08)

Ignore setup_module(), teardown_module(), etc. in pytest test_*.py files (Jendrik Seipp).
Add whitelist for socketserver.TCPServer.allow_reuse_address (Ben Elliston).
Clarify that --exclude patterns are matched against absolute paths (Jendrik Seipp, #260).
Fix example in README file (Jendrik Seipp, #272).

2.6 (2022-09-19)

Add basic match statement support (kreathon, #276, #291).

2.5 (2022-07-03)

Mark imports in __all__ as used (kreathon, #172, #282).
Add whitelist for pint.UnitRegistry.default_formatter (Ben Elliston, #258).

2.4 (2022-05-19)

Print absolute filepaths as relative again (as in version 2.1 and before) if they are below the current directory (The-Compiler, #246).
Run tests and add PyPI trove for Python 3.10 (chayim, #266).
Allow using the del keyword to mark unused variables (sshishov, #279).

2.3 (2021-01-16)

Add pre-commit hook (Clément Robert, #244).

2.2 (2021-01-15)

Only parse format strings when being used with locals() (jingw, #225).
Don't override paths in pyproject.toml with empty CLI paths (bcbnz, #228).
Run continuous integration tests for Python 3.9 (ju-sh, #232).
Use pathlib internally (ju-sh, #226).

2.1 (2020-08-19)

Treat getattr/hasattr(obj, "constant_string", ...) as a reference to obj.constant_string (jingw, #219).
Fix false positives when assigning to x.some_name but reading via some_name, at the cost of potential false negatives (jingw, #221).
Allow reading options from pyproject.toml (Michel Albert, #164, #215).

2.0 (2020-08-11)

Parse # type: ... comments if on Python 3.8+ (jingw, #220).
Bump minimum Python version to 3.6 (Jendrik Seipp, #218). The last Vulture release that supports Python 2.7 and Python 3.5 is version 1.6.
Consider all files under test or tests directories test files (Jendrik Seipp).
Ignore logging.Logger.propagate attribute (Jendrik Seipp).

1.6 (2020-07-28)

Differentiate between functions and methods (Jendrik Seipp, #112, #209).
Move from Travis to GitHub actions (RJ722, #211).

1.5 (2020-05-24)

Support flake8 "noqa" error codes F401 (unused import) and F841 (unused local variable) (RJ722, #195).
Detect unreachable code in conditional expressions (Agathiyan Bragadeesh, #178).

1.4 (2020-03-30)

Ignore unused import statements in __init__.py (RJ722, #192).
Report first decorator's line number for unused decorated objects on Python 3.8+ (RJ722, #200).
Check code with black and pyupgrade.

1.3 (2020-02-03)

Detect redundant 'if' conditions without 'else' blocks.
Add whitelist for string.Formatter (Joseph Bylund, #183).

1.2 (2019-11-22)

Fix tests for Python 3.8 (#166).
Use new Constant AST node under Python 3.8+ (#175).
Add test for f-strings (#177).
Add whitelist for logging module.

1.1 (2019-09-23)

Add sys.excepthook to sys whitelist.
Add whitelist for ctypes module.
Check that type annotations are parsed and type comments are ignored (thanks @kx-chen).
Support checking files with BOM under Python 2.7 (#170).

1.0 (2018-10-23)

Add --ignore-decorators flag (thanks @RJ722).
Add whitelist for threading module (thanks @andrewhalle).

0.29 (2018-07-31)

Add --ignore-names flag for ignoring names matching the given glob patterns (thanks @RJ722).

0.28 (2018-07-05)

Add --make-whitelist flag for reporting output in whitelist format (thanks @RJ722).
Ignore case of --exclude arguments on Windows.
Add *-test.py to recognized test file patterns.
Add failureException, longMessage and maxDiff to unittest whitelist.
Refer to actual objects rather than their mocks in default whitelists (thanks @RJ722).
Don't import any Vulture modules in setup.py (thanks @RJ722).

0.27 (2018-06-05)

Report while (True): ... else: ... as unreachable (thanks @RJ722).
Use argparse instead of optparse.
Whitelist Mock.return_value and Mock.side_effect in unittest.mock module.
Drop support for Python 2.6 and 3.3.
Improve documentation and test coverage (thanks @RJ722).

0.26 (2017-08-28)

Detect async function definitions (thanks @RJ722).
Add Item.get_report() method (thanks @RJ722).
Move method for finding Python modules out of Vulture class.

0.25 (2017-08-15)

Detect unsatisfiable statements containing and, or and not.
Use filenames and line numbers as tie-breakers when sorting by size.
Store first and last line numbers in Item objects.
Pass relevant options directly to scavenge() and report().

0.24 (2017-08-14)

Detect unsatisfiable while-conditions (thanks @RJ722).
Detect unsatisfiable if- and else-conditions (thanks @RJ722).
Handle null bytes in source code.

0.23 (2017-08-10)

Add --min-confidence flag (thanks @RJ722).

0.22 (2017-08-04)

Detect unreachable code after return, break, continue and raise (thanks @RJ722).
Parse all variable and attribute names in new format strings.
Extend ast whitelist.

0.21 (2017-07-26)

If an unused item is defined multiple times, report it multiple times.
Make size estimates for function calls more accurate.
Create wheel files for Vulture (thanks @RJ722).

0.20 (2017-07-26)

Report unused tuple assignments as dead code.
Report attribute names that have the same names as variables as dead code.
Let Item class inherit from object (thanks @RJ722).
Handle names imported as aliases like all other used variable names.
Rename Vulture.used_vars to Vulture.used_names.
Use function for determining which imports to ignore.
Only try to import each whitelist file once.
Store used names and used attributes in sets instead of lists.
Fix estimating the size of code containing ellipses (...).
Refactor and simplify code.

0.19 (2017-07-20)

Don't ignore __foo variable names.
Use separate methods for determining whether to ignore classes and functions.
Only try to find a whitelist for each defined import once (thanks @roivanov).
Fix finding the last child for many types of AST nodes.

0.18 (2017-07-17)

Make --sort-by-size faster and more accurate (thanks @RJ722).

0.17 (2017-07-17)

Add get_unused_code() method.
Return with exit code 1 when syntax errors are found or files can't be read.

0.16 (2017-07-12)

Differentiate between unused classes and functions (thanks @RJ722).
Add --sort-by-size option (thanks @jackric and @RJ722).
Count imports as used if they are accessed as module attributes.

0.15 (2017-07-04)

Automatically include whitelists based on imported modules (thanks @RJ722).
Add --version parameter (thanks @RJ722).
Add appveyor tests for testing on Windows (thanks @RJ722).

0.14 (2017-04-06)

Add stub whitelist file for Python standard library (thanks @RJ722)
Ignore class names starting with "Test" in "test_" files (thanks @thisch).
Ignore "test_" functions only in "test_" files.

0.13 (2017-03-06)

Ignore star-imported names since we cannot detect whether they are used.
Move repository to GitHub.

0.12 (2017-01-05)

Detect unused imports.
Use tokenize.open() on Python >= 3.2 for reading input files, assume UTF-8 encoding on older Python versions.

0.11 (2016-11-27)

Use the system's default encoding when reading files.
Report syntax errors instead of aborting.

0.10 (2016-07-14)

Detect unused function and method arguments (issue #15).
Detect unused *args and **kwargs parameters.
Change license from GPL to MIT.

0.9 (2016-06-29)

Don't flag attributes as unused if they are used as global variables in another module (thanks Florian Bruhin).
Don't consider "True" and "False" variable names.
Abort with error message when invoked on .pyc files.

0.8.1 (2015-09-28)

Fix code for Python 3.

0.8 (2015-09-28)

Do not flag names imported with "import as" as dead code (thanks Tom Terrace).

0.7 (2015-09-26)

Exit with exitcode 1 if path on commandline can't be found.
Test vulture with vulture using a whitelist module for false positives.
Add tests that run vulture as a script.
Add "python setup.py test" command for running tests.
Add support for tox.
Raise test coverage to 100%.
Remove ez_setup.py.

0.6 (2014-09-06)

Ignore function names starting with "test_".
Parse variable names in new format strings (e.g. "This is {x}".format(x="nice")).
Only parse alphanumeric variable names in format strings and ignore types.
Abort with exit code 1 on syntax errors.
Support installation under Windows by using setuptools (thanks Reuben Fletcher-Costin).

0.5 (2014-05-09)

If dead code is found, exit with 1.

0.4.1 (2013-09-17)

Only warn if a path given on the command line cannot be found.

0.4 (2013-06-23)

Ignore unused variables starting with an underscore.
Show warning for syntax errors instead of aborting directly.
Print warning if a file cannot be found.

0.3 (2012-03-19)

Add support for python3
Report unused attributes
Find tuple assignments in comprehensions
Scan files given on the command line even if they don't end with .py

0.2 (2012-03-18)

Only format nodes in verbose mode (gives 4x speedup).

0.1 (2012-03-17)

First release.

vulture2.14

Package Downloads

Authors

Project URLs

Requires Python

Dependencies