Skip to main content

Python mutation testing.

Project description

mutatest: Python mutation testing

https://img.shields.io/badge/Python_version-3.7-green.svg https://travis-ci.org/EvanKepner/mutatest.svg?branch=master https://codecov.io/gh/EvanKepner/mutatest/branch/master/graph/badge.svg https://img.shields.io/badge/code%20style-black-000000.svg https://pepy.tech/badge/mutatest

Are you confident in your tests? Try out mutatest and see if your tests will detect small modifications (mutations) in the code. Surviving mutations represent subtle changes that might slip past your continuous integration checks and are undetectable by your tests.

Features:
  • Simple command line tool.
  • Integrated with coverage to only create meaningful mutants.
  • Built on Python’s Abstract Syntax Tree (AST) grammar to ensure mutants are valid.
  • Built for efficiency with multiple running modes and random sampling of mutation targets.
  • Does not modify your source code, only the __pycache__.
  • Flexible enough to run on a whole package or a single file.

Installation

mutatest requires Python 3.7. You can install with pip:

$ pip install mutatest

Alternatively, clone this repo and install locally:

$ cd mutatest
$ pip install .

mutatest is designed to work when your test files are separated from your source directory and are prefixed with test_. See Pytest Test Layout [2] for more details.

Using mutatest

mutatest is designed to be a diagnostic command line tool for your test coverage assessment.

The mutation trial process follows these steps when mutatest is run:

  1. Scan for your existing Python package, or use the input source location.
  2. Create an abstract syntax tree (AST) from the source files.
  3. Identify locations in the code that may be mutated (line and column). If you are running with coverage the sample is restricted only to lines that are marked as covered in the .coverage file.
  4. Take a random sample of the identified locations.
  5. Apply a mutation at the location by modifying a copy of the AST and writing a new cache file to the appropriate __pycache__ location with the source file statistics.
  6. Run the test suite. This will use the mutated __pycache__ file since the source statistics are the same for modification time.
  7. See if the test suite detected the mutation by a failed test.
  8. Remove the modified __pycache__ file.
  9. Repeat steps 5-9 for the remaining selected locations to mutate.
  10. Write an output report of the various mutation results.

A “clean trial” of your tests is run before any mutations are applied. This same “clean trial” is run at the end of the mutation testing. This ensures that your original test suite passes before attempting to detect surviving mutations and that the __pycache__ has been appropriately reset when the mutation trials are finished.

Specifying source files and test commands

If you have a Python package in a directory with an associated tests/ folder (or internal test_ prefixed files, see the examples below) that are auto-detected with pytest, then you can run mutatest without any arguments.

$ mutatest

It will detect the package, and run pytest by default. If you want to run with special arguments, such as to exclude a custom marker, you can pass in the --testcmds argument with the desired string.

Here is the command to run pytest and exclude tests marked with pytest.mark.slow.

$ mutatest --testcmds "pytest -m 'not slow'"

# using shorthand arguments
$ mutatest -t "pytest -m 'not slow'"

You can use this syntax if you want to specify a single module in your package to run and test.

$ mutatest --src mypackage/run.py --testcmds "pytest tests/test_run.py"

# using shorthand arguments
$ mutatest -s mypackage/run.py -t "pytest tests/test_run.py"

There is an option to exclude files from the source set. By default, __init__.py is excluded. Exclude files using the --exclude argument with a space delimited list of files in a string. Only list the file name, not paths.

$ mutatest --exclude "__init__.py _devtools.py"

# using shorthand arguments
$ mutatest -e "__init__.py _devtools.py"

Coverage optimization

Any command combination that generates a .coverage file will use that as a restriction mechanism for the sample space to only select mutation locations that are covered. For example, running:

$ mutatest --testcmds "pytest --cov=mypackage tests/test_run.py"

# using shorthand arguments
$ mutatest -t "pytest --cov=mypackage tests/test_run.py"

would generate the .coverage file based on tests/test_run.py. Therefore, even though the entire package is seen only the lines covered by tests/test_run.py will be mutated during the trials. You can override this behavior with the --nocov flag on the command line.

If you have a pytest.ini file that includes the --cov command the default behavior of mutatest will generate the coverage file. You will see this in the CLI output at the beginning of the trials:

$ mutatest -n 4 -t "pytest --cov=mypackage"

... prior output...

... Get mutatest targets from AST.
... Full sample space size: 115
... Coverage optimized sample space size: 75
... Selecting 4 locations from 75 potentials.

... continued output...

Auto-detected package structures

The following package structures would be auto-detected if you ran mutatest from the same directory holding examplepkg/. You can always point to a specific directory using the --source argument. These are outlined in the Pytest Test Layout [2] documentation.

Example with internal tests

.
└── examplepkg
    ├── __init__.py
    ├── run.py
    └── test_run.py

Example with external tests

.
├── examplepkg
│   ├── __init__.py
│   └── run.py
└── tests
    └── test_run.py

Selecting a running mode

mutatest has different running modes to make trials faster. The running modes determine what will happen after a mutation trial. For example, you can choose to stop further mutations at a location as soon as a survivor is detected. The different running mode choices are:

Run modes:
  • f: full mode, run all possible combinations (slowest but most thorough).
  • s: break on first SURVIVOR per mutated location e.g. if there is a single surviving mutation at a location move to the next location without further testing. This is the default mode.
  • d: break on the first DETECTION per mutated location e.g. if there is a detected mutation on at a location move to the next one.
  • sd: break on the first SURVIVOR or DETECTION (fastest, and least thorough).

The API for mutatest.controller.run_mutation_trials offers finer control over the run method beyond the CLI.

A good practice when first starting is to set the mode to sd which will stop if a mutation survives or is detected, effectively running a single mutation per candidate location. This is the fastest running mode and can give you a sense of investigation areas quickly.

$ mutatest --mode sd

# using shorthand arguments
$ mutatest -m sd

Controlling randomization behavior and trial number

mutatest uses random sampling of all source candidate locations and of potential mutations to substitute at a location. You can set a random seed for repeatable trials using the --rseed argument. The --nlocations argument controls the size of the sample of locations to mutate. If it exceeds the number of candidate locations then the full set of candidate locations is used.

$ mutatest --nlocations 5 --rseed 314

# using shorthand arguments
$ mutatest -n 5 -r 314

Setting the output location

By default, mutatest will write a mutation_report.rst to the current working directory. You can set this file name and path location using the --output argument.

$ mutatest --output path/to/my_custom_file.rst

# using shorthand arguments
$ mutatest -o path/to/my_custom_file.rst

The output report will include the arguments used to generate it along with the total runtimes. The SURVIVORS section of the output report is the one you should pay attention to. These are the mutations that were undetected by your test suite. The report includes file names, line numbers, column numbers, original operation, and mutation for ease of diagnostic investigation.

Putting it all together

If you want to run 5 trials, in fast sd mode, with a random seed of 345 and an output file name of mutation_345.rst, you would do the following if your directory structure has a Python package folder and tests that are auto-discoverable and run by pytest.

$ mutatest -n 5 -m sd -r 345 -o mutation_345.rst

With coverage optimization if your pytest.ini file does not already specify it:

$ mutatest -n 5 -m sd -r 345 -o mutation_345.rst -t "pytest --cov=mypackage"

Getting help

Run mutatest --help to see command line arguments and supported operations:

$ mutatest --help

usage: Mutatest [-h] [-e STR_LIST] [-m {f,s,d,sd}] [-n INT] [-o PATH] [-r INT]
                [-s PATH] [-t STR_CMDS] [--debug]

Python mutation testing. Mutatest will manipulate local __pycache__ files.

optional arguments:
  -h, --help            show this help message and exit
  -e STR_LIST, --exclude STR_LIST
                        Space delimited string list of .py file names to exclude. (default: '__init__.py')
  -m {f,s,d,sd}, --mode {f,s,d,sd}
                        Running modes, see the choice option descriptions below. (default: s)
  -n INT, --nlocations INT
                        Number of locations in code to randomly select for mutation from possible targets. (default: 10)
  -o PATH, --output PATH
                        Output file location for results. (default: mutation_report.rst)
  -r INT, --rseed INT   Random seed to use for sample selection.
  -s PATH, --src PATH   Source code (file or directory) for mutation testing. (default: auto-detection attempt).
  -t STR_CMDS, --testcmds STR_CMDS
                        Test command string to execute. (default: 'pytest')
  --debug               Turn on DEBUG level logging output.
  --nocov               Ignore coverage files for optimization.

Mutations

mutatest is early in development and supports the following mutation operations based on the Python AST grammar [1]:

Supported operations:
  • AugAssign mutations e.g. += -= *= /=.
  • BinOp mutations e.g. + - / *.
  • BinOp Bitwise Comparison mutations e.g. x&y x|y x^y.
  • BinOp Bitwise Shift mutations e.g. << >>.
  • BoolOp mutations e.g. and or.
  • Compare mutations e.g. == >= < <= !=.
  • Compare In mutations e.g. in, not in.
  • Compare Is mutations e.g. is, is not.
  • Index mutations e.g. i[0] becomes i[1] and i[-1].
  • NameConstant mutations e.g. True, False, and None.
  • Slice mutations e.g. changing x[:2] to x[2:]

These are the current operations that are mutated as compatible sets.

AugAssign

Augmented assignment e.g. += -= /= *=.

Members:
  • AugAssign_Add
  • AugAssign_Div
  • AugAssign_Mult
  • AugAssign_Sub

Example:

# source code
x += y

# mutations
x -= y  # AugAssign_Sub
x *= y  # AugAssign_Mult
x /= y  # AugAssign_Div

BinOp

Binary operations e.g. add, subtract, divide, etc.

Members:
  • ast.Add
  • ast.Div
  • ast.FloorDiv
  • ast.Mod
  • ast.Mult
  • ast.Pow
  • ast.Sub

Example:

# source code
x = a + b

# mutations
x = a / b  # ast.Div
x = a - b  # ast.Sub

BinOp Bit Comparison

Bitwise comparison operations e.g. x & y, x | y, x ^ y.

Members:
  • ast.BitAnd
  • ast.BitOr
  • ast.BitXor

Example:

# source code
x = a & y

# mutations
x = a | y  # ast.BitOr
x = a ^ y  # ast.BitXor

BinOp Bit Shifts

Bitwise shift operations e.g. << >>.

Members:
  • ast.LShift
  • ast.RShift

Example:

# source code
x >> y

# mutation
x << y

BoolOp

Boolean operations e.g. and or.

Members:
  • ast.And
  • ast.Or

Example:

# source code
if x and y:

# mutation
if x or y:

Compare

Comparison operations e.g. == >= <= > <.

Members:
  • ast.Eq
  • ast.Gt
  • ast.GtE
  • ast.Lt
  • ast.LtE
  • ast.NotEq

Example:

# source code
x >= y

# mutations
x < y  # ast.Lt
x > y  # ast.Gt
x != y  # ast.NotEq

Compare In

Compare membership e.g. in, not in.

Members:
  • ast.In
  • ast.NotIn

Example:

# source code
x in [1, 2, 3, 4]

# mutation
x not in [1, 2, 3, 4]

Compare Is

Comapre identity e.g. is, is not.

Members:
  • ast.Is
  • ast.IsNot

Example:

# source code
x is None

# mutation
x is not None

Index

Index values for iterables e.g. i[-1], i[0], i[0][1]. It is worth noting that this is a unique mutation form in that any index value that is positive will be marked as Index_NumPos and the same relative behavior will happen for negative index values to Index_NumNeg. During the mutation process there are three possible outcomes: the index is set to 0, -1 or 1. The alternate values are chosen as potential mutations e.g. if the original operation is classified as Index_NumPos such as x[10] then valid mutations are to x[0] or x[-1].

Members:
  • Index_NumNeg
  • Index_NumPos
  • Index_NumZero

Example:

# source code
x = [a[10], a[-4], a[0]]

# mutations
x = [a[-1], a[-4], a[0]]  # a[10] mutated to Index_NumNeg
x = [a[10], a[0], a[0]]  # a[-4] mutated to Index_NumZero
x = [a[10], a[1], a[0]]  # a[-4] mutated to Index_NumPos
x = [a[10], a[-4], a[1]]  # a[0] mutated to Index_NumPos

NameConstant

Named constant mutations e.g. True, False, None.

Members:
  • False
  • None
  • True

Example:

# source code
x = True

# mutations
x = False
X = None

Slices

Slice mutations to swap lower/upper values, or change range e.g. x[2:] to x[:2] or x[1:5] to x[1:4]. This is a unique mutation. If the upper or lower bound is set to None then the bound values are swapped. This is represented by the operations of Slice_UnboundedUpper for swap None to the “upper” value from “lower”.

The “ToZero” operations change the list by moving the upper bound by one unit towards zero from the absolute value and then applying the original sign e.g. x[0:2] would become x[0:1], and x[-4:-1] would become x[-4:0]. In the positive case, which is assumed to be the more common pattern, this results in shrinking the index slice by 1. Note that these “ToZero” operations appear self-referential in the report output. This is because an operation identified as a Slice_UPosToZero remains as a Slice_UPosToZero but with updated values.

Members:
  • Slice_Unbounded
  • Slice_UnboundedLower
  • Slice_UnboundedUpper
  • Slice_UNegToZero
  • Slice_UPosToZero

Example:

# source code
w = a[:2]
x = a[4:]
y = a[1:5]
z = a[-5:-1]

# mutation
w = a[2:]  # Slice_UnboundedUpper, upper is now unbounded and lower has a value
x = a[4:]
y = a[1:5]
z = a[-5:-1]

# mutation
w = a[:2]
x = a[:4]  # Slice_UnboundedLower, lower is now unbounded and upper has a value
y = a[1:5]
z = a[-5:-1]

# mutation
w = a[:2]
x = a[:]  # Slice_Unbounded, both upper and lower are unbounded
y = a[1:5]
z = a[-5:-1]


# mutation
w = a[:2]
x = a[4:]
y = a[1:4]  # Slice_UPosToZero, upper bound moves towards zero bound by 1 when positive
z = a[-5:-1]

# mutation
w = a[:2]
x = a[4:]
y = a[1:5]
z = a[-5:0]  # Slice_UNegToZero, upper bound moves by 1 from absolute value when negative

Known limitations

Since mutatest operates on the local __pycache__ it is a serial execution process. This means it can be slow, and will take as long as running your test suite in series for the number of operations. It’s designed as a diagnostic tool, not something you would run in your CICD pipeline.

If you kill the mutatest process before the trials complete you may end up with partially mutated __pycache__ files. If this happens the best fix is to remove the __pycache__ directories and let them rebuild automatically the next time your package is imported (for instance, by re-running your test suite).

The mutation status is based on the return code of the test suite e.g. 0 for success, 1 for failure. mutatest can theoretically be run with any test suite that you pass with the --testcmds argument; however, only pytest has been tested to date. The mutatest.maker.MutantTrialResult namedtuple contains the definitions for translating return codes into mutation trial statuses.

[1]https://docs.python.org/3/library/ast.html#abstract-grammar
[2](1, 2) https://docs.pytest.org/en/latest/goodpractices.html#choosing-a-test-layout-import-rules

Changelog

mutatest is alpha software, and backwards compatibility between releases is not guaranteed while under development.

0.7.1

  • Internal changes to Slice mutations for clearer categorization and report output.
  • Includes clearing names to Slice_Swap and Slice_RangeChange for categories.
  • Updates operation names to Slice_Unbounded... with “lower” or “upper”.

0.7.0

  • Added new slice mutations:
    1. Slice_SwapNoneUL and Slice_SwapNoneLU for swapping the upper and lower bound values when only one is specified e.g. x[1:] to x[:1].
    2. Slice_UPosToZero and Slice_UNegToZero for moving the upper bound of a slice by 1 unit e.g. x[1:5] becomes x[1:4].

0.6.1

  • Added explicit tests for argparse cli options.
  • Added mechanism to sort reporting mutations by source file, then line number, then column number.

0.6.0

  • Including pytest in the installation requirements. Technically, any test runner can be used but with all base package assumptions being built around pytest this feels like the right assumption to call out as an install dependency. It is the default behavior.
  • Updated controller for test file exclusion to explicitly match prefix or suffix cases for "test_" and "_test" per pytest conventions.
  • Changed error and unknown status results to console color as yellow instead of red.
  • Added multiple invariant property tests, primarily to controller and cache.
  • Added hypothesis to the test components of extras_require.
  • Moved to @property decorators for internal class properties that should only be set at initialization, may add custom setters at a later time.
  • Fixed a zero-division bug in the cli when reporting coverage percentage.

0.5.0

  • Addition of optimizers, including the new class CoverageOptimizer.
  • This optimizer restricts the full sample space only to source locations that are marked as covered in the .coverage file. If you have a pytest.ini that includes the --cov= command it will automatically generate during the clean-trial run.

0.4.2

  • More behind the scenes maintenance: updated debug level logging to include source file names and line numbers for all visit operations and separated colorized output to a new function.

0.4.1

  • Updated the reporting functions to return colorized display results to CLI.

0.4.0

  • Added new mutation support for:
    1. AugAssign in AST e.g. += -= *= /=.
    2. Index substitution in lists e.g. take a positive number like i[1] and mutate to zero and a negative number e.g. i[-1] i[0].
  • Added a desc attribute to transformers.MutationOpSet that is used in the cli help display.
  • Updated the cli help display to show the description and valid members.

0.3.0

  • Added new mutation support for NameConstant in AST.
  • This includes substitutions for singleton assignments such as: True, False, and None.
  • This is the first non-type mutation and required adding a readonly parameter to the transformers.MutateAST class. Additionally, the type-hints for the LocIndex and MutationOpSet were updated to Any to support the mixed types. This was more flexible than a series of overload signatures.

0.2.0

  • Added new compare mutation support for:
    1. Compare Is mutations e.g. is, is not.
    2. Compare In mutations e.g. in, not in.

0.1.0

  • Initial release!

  • Requires Python 3.7 due to the importlib internal references for manipulating cache.

  • Run mutation tests using the mutatest command line interface.

  • Supported operations:

    1. BinOp mutations e.g. + - / * including bit-operations.
    2. Compare mutations e.g. == >= < <= !=.
    3. BoolOp mutations e.g. and or.

Authors

  • Evan Kepner

Project details


Download files

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

Filename, size & hash SHA256 hash help File type Python version Upload date
mutatest-0.7.1-py37-none-any.whl (31.3 kB) Copy SHA256 hash SHA256 Wheel py37
mutatest-0.7.1.tar.gz (53.1 kB) Copy SHA256 hash SHA256 Source None

Supported by

Elastic Elastic Search Pingdom Pingdom Monitoring Google Google BigQuery Sentry Sentry Error logging AWS AWS Cloud computing DataDog DataDog Monitoring Fastly Fastly CDN SignalFx SignalFx Supporter DigiCert DigiCert EV certificate StatusPage StatusPage Status page