Summary -- Release highlights#

New checkers#

  • single-string-used-for-slots check was added, which is used whenever a class is using a single string as a slot value. While this is technically not a problem per se, it might trip users when manipulating the slots value as an iterable, which would in turn iterate over characters of the slot value. In order to be more straight-forward, always try to use a container such as a list or a tuple for defining slot values.

  • We added a new check, literal-comparison, which is used whenever pylint can detect a comparison to a literal. This is usually not what we want and, potentially, error prone. For instance, in the given example, the first string comparison returns true, since smaller strings are interned by the interpreter, while for larger ones, it will return False:

    mystring = "ok"
    if mystring is "ok": # Returns true
       ...
    
    mystring = "a" * 1000
    if mystring is ("a" * 1000): # This will return False
       ...
    

    Instead of using the is operator, you should use the == operator for this use case.

  • We added a new refactoring message, consider-merging-isinstance, which is emitted whenever we can detect that consecutive isinstance calls can be merged together. For instance, in this example, we can merge the first two isinstance calls:

    # $ cat a.py
    if isinstance(x, int) or isinstance(x, float):
        pass
    if isinstance(x, (int, float)) or isinstance(x, str):
        pass
    # $ pylint a.py
    # R:  1, 0: Consider merging these isinstance calls to isinstance(x, (float, int)) (consider-merging-isinstance)
    # R:  3, 0: Consider merging these isinstance calls to isinstance(x, (int, float, str)) (consider-merging-isinstance)
    
  • A new error check was added, invalid-metaclass, which is used whenever pylint can detect that a given class is using a metaclass which is invalid for the purpose of the class. This usually might indicate a problem in the code, rather than something done on purpose.

    # Needs to inherit from *type* in order to be valid
    class SomeClass(object):
        ...
    
    class MyClass(metaclass=SomeClass):
        pass
    
  • A new warning was added, useless-super-delegation, which is used whenever we can detect that an overridden method is useless, relying on super() delegation to do the same thing as another method from the MRO.

    For instance, in this example, the first two methods are useless, since they do the exact same thing as the methods from the base classes, while the next two methods are not, since they do some extra operations with the passed arguments.

    class Impl(Base):
    
        def __init__(self, param1, param2):
            super(Impl, self).__init__(param1, param2)
    
        def useless(self, first, second):
            return super(Impl, self).useless(first, second)
    
        def not_useless(self, first, **kwargs):
            debug = kwargs.pop('debug', False)
            if debug:
                ...
            return super(Impl, self).not_useless(first, **kwargs)
    
        def not_useless_1(self, first, *args):
            return super(Impl, self).not_useless_1(first + some_value, *args)
    
  • A new warning was added, len-as-condition, which is used whenever we detect that a condition uses len(SEQUENCE) incorrectly. Instead one could use if SEQUENCE or if not SEQUENCE.

    For instance, all of the examples below:

    if len(S):
      pass
    
    if not len(S):
      pass
    
    if len(S) > 0:
      pass
    
    if len(S) != 0:
      pass
    
    if len(S) == 0:
      pass
    

    can be written in a more natural way:

    if S:
      pass
    
    if not S:
      pass
    

    See https://peps.python.org/pep-0008/#programming-recommendations for more information.

  • A new extension was added, emptystring.py which detects whenever we detect comparisons to empty string constants. This extension is disabled by default. For instance, the examples below:

    if S != "":
      pass
    
    if S == '':
      pass
    

    can be written in a more natural way:

    if S:
      pass
    
    if not S:
      pass
    

    An exception to this is when empty string is an allowed value whose meaning is treated differently than None. For example the meaning could be user selected no additional options vs. user has not made their selection yet!

    You can activate this checker by adding the line:

    load-plugins=pylint.extensions.emptystring
    

    to the MAIN section of your .pylintrc or using the command:

    $ pylint a.py --load-plugins=pylint.extensions.emptystring
    
  • A new extension was added, comparetozero.py which detects whenever we compare integers to zero. This extension is disabled by default. For instance, the examples below:

    if X != 0:
      pass
    
    if X == 0:
      pass
    

    can be written in a more natural way:

    if X:
      pass
    
    if not X:
      pass
    

    An exception to this is when zero is an allowed value whose meaning is treated differently than None. For example the meaning could be None means no limit, while 0 means the limit it zero!

    You can activate this checker by adding the line:

    load-plugins=pylint.extensions.comparetozero
    

    to the MAIN section of your .pylintrc or using the command:

    $ pylint a.py --load-plugins=pylint.extensions.comparetozero
    
  • We've added new error conditions for bad-super-call which now detect the usage of super(type(self), self) and super(self.__class__, self) patterns. These can lead to recursion loop in derived classes. The problem is visible only if you override a class that uses these incorrect invocations of super().

    For instance, Derived.__init__() will correctly call Base.__init__. At this point type(self) will be equal to Derived and the call again goes to Base.__init__ and we enter a recursion loop.

    class Base(object):
        def __init__(self, param1, param2):
            super(type(self), self).__init__(param1, param2)
    
    class Derived(Base):
        def __init__(self, param1, param2):
            super(Derived, self).__init__(param1, param2)
    
  • The warnings missing-returns-doc and missing-yields-doc have each been replaced with two new warnings - missing-[return|yield]-doc and missing-[return|yield]-type-doc. Having these as separate warnings allows the user to choose whether their documentation style requires text descriptions of function return/yield, specification of return/yield types, or both.

    # This will raise missing-return-type-doc but not missing-return-doc
    def my_sphinx_style_func(self):
        """This is a Sphinx-style docstring.
    
        :returns: Always False
        """
        return False
    
    # This will raise missing-return-doc but not missing-return-type-doc
    def my_google_style_func(self):
        """This is a Google-style docstring.
    
        Returns:
            bool:
        """
        return False
    
  • A new refactoring check was added, redefined-argument-from-local, which is emitted when pylint can detect than a function argument is redefined locally in some potential error prone cases. For instance, in the following piece of code, we have a bug, since the check will never return True, given the fact that we are comparing the same object to its attributes.

    def test(resource):
        for resource in resources:
            # The ``for`` is reusing ``resource``, which means that the following
            # ``resource`` is not what we wanted to check against.
            if resource.resource_type == resource:
               call_resource(resource)
    

    Other places where this check looks are with statement name bindings and except handler's name binding.

  • A new refactoring check was added, no-else-return, which is emitted when pylint encounters an else following a chain of ifs, all of them containing a return statement.

    def foo1(x, y, z):
        if x:
            return y
        else:  # This is unnecessary here.
            return z
    

    We could fix it deleting the else statement.

    def foo1(x, y, z):
        if x:
            return y
        return z
    
  • A new Python 3 check was added, eq-without-hash, which enforces classes that implement __eq__ also implement __hash__. The behavior around classes which implement __eq__ but not __hash__ changed in Python 3; in Python 2 such classes would get object.__hash__ as their default implementation. In Python 3, aforementioned classes get None as their implementation thus making them unhashable.

    class JustEq(object):
       def __init__(self, x):
         self.x = x
    
       def __eq__(self, other):
         return self.x == other.x
    
    class Neither(object):
      def __init__(self, x):
        self.x = x
    
    class HashAndEq(object):
       def __init__(self, x):
         self.x = x
    
       def __eq__(self, other):
         return self.x == other.x
    
       def __hash__(self):
         return hash(self.x)
    
    {Neither(1), Neither(2)}  # OK in Python 2 and Python 3
    {HashAndEq(1), HashAndEq(2)}  # OK in Python 2 and Python 3
    {JustEq(1), JustEq(2)}  # Works in Python 2, throws in Python 3
    

    In general, this is a poor practice which motivated the behavior change.

    as_set = {JustEq(1), JustEq(2)}
    print(JustEq(1) in as_set)  # prints False
    print(JustEq(1) in list(as_set))  # prints True
    

    In order to fix this error and avoid behavior differences between Python 2 and Python 3, classes should either explicitly set __hash__ to None or implement a hashing function.

    class JustEq(object):
       def __init__(self, x):
         self.x = x
    
       def __eq__(self, other):
         return self.x == other.x
    
       __hash__ = None
    
    {JustEq(1), JustEq(2)}  # Now throws an exception in both Python 2 and Python 3.
    
  • 3 new Python 3 checkers were added, div-method, idiv-method and rdiv-method. The magic methods __div__ and __idiv__ have been phased out in Python 3 in favor of __truediv__. Classes implementing __div__ that still need to be used from Python 2 code not using from __future__ import division should implement __truediv__ and alias __div__ to that implementation.

    from __future__ import division
    
    class DivisibleThing(object):
       def __init__(self, x):
         self.x = x
    
       def __truediv__(self, other):
         return DivisibleThing(self.x / other.x)
    
       __div__ = __truediv__
    
  • A new Python 3 checker was added to warn about accessing the message attribute on Exceptions. The message attribute was deprecated in Python 2.7 and was removed in Python 3. See https://peps.python.org/pep-0352/#retracted-ideas for more information.

    try:
      raise Exception("Oh No!!")
    except Exception as e:
      print(e.message)
    

    Instead of relying on the message attribute, you should explicitly cast the exception to a string:

    try:
      raise Exception("Oh No!!")
    except Exception as e:
      print(str(e))
    
  • A new Python 3 checker was added to warn about using encode or decode on strings with non-text codecs. This check also checks calls to open with the keyword argument encoding. See https://docs.python.org/3/whatsnew/3.4.html#improvements-to-codec-handling for more information.

    'hello world'.encode('hex')
    

    Instead of using the encode method for non-text codecs use the codecs module.

    import codecs
    codecs.encode('hello world', 'hex')
    
  • A new warning was added, overlapping-except, which is emitted when an except handler treats two exceptions which are overlapping. This means that one exception is an ancestor of the other one or it is just an alias.

    For example, in Python 3.3+, IOError is an alias for OSError. In addition, socket.error is an alias for OSError. The intention is to find cases like the following:

    import socket
    try:
        pass
    except (ConnectionError, IOError, OSError, socket.error):
        pass
    
  • A new Python 3 checker was added to warn about accessing sys.maxint. This attribute was removed in Python 3 in favor of sys.maxsize.

    import sys
    print(sys.maxint)
    

    Instead of using sys.maxint, use sys.maxsize

    import sys
    print(sys.maxsize)
    
  • A new Python 3 checker was added to warn about importing modules that have either moved or been removed from the standard library.

    One of the major undertakings with Python 3 was a reorganization of the standard library to remove old or supplanted modules and reorganize some of the existing modules. As a result, roughly 100 modules that exist in Python 2 no longer exist in Python 3. See https://peps.python.org/pep-3108/ and https://peps.python.org/pep-0004/ for more information. There were suggestions on how to handle this, at pythonhosted.org/six/#module-six.moves (dead link) or python3porting.com/stdlib.html (dead link).

    from cStringIO import StringIO
    

    Instead of directly importing the deprecated module, either use six.moves or a conditional import.

    from six.moves import cStringIO as StringIO
    
    if sys.version_info[0] >= 3:
        from io import StringIO
    else:
        from cStringIO import StringIO
    

    This checker will assume any imports that happen within a conditional or a try/except block are valid.

  • A new Python 3 checker was added to warn about accessing deprecated functions on the string module. Python 3 removed functions that were duplicated from the builtin str class. See https://docs.python.org/2/library/string.html#deprecated-string-functions for more information.

    import string
    print(string.upper('hello world!'))
    

    Instead of using string.upper, call the upper method directly on the string object.

    "hello world!".upper()
    
  • A new Python 3 checker was added to warn about calling str.translate with the removed deletechars parameter. str.translate is frequently used as a way to remove characters from a string.

    'hello world'.translate(None, 'low')
    

    Unfortunately, there is not an idiomatic way of writing this call in a 2and3 compatible way. If this code is not in the critical path for your application and the use of translate was a premature optimization, consider using re.sub instead:

    import re
    chars_to_remove = re.compile('[low]')
    chars_to_remove.sub('', 'hello world')
    

    If this code is in your critical path and must be as fast as possible, consider declaring a helper method that varies based upon Python version.

    if six.PY3:
        def _remove_characters(text, deletechars):
            return text.translate({ord(x): None for x in deletechars})
    else:
        def _remove_characters(text, deletechars):
            return text.translate(None, deletechars)
    
  • A new refactoring check was added, consider-using-ternary, which is emitted when pylint encounters constructs which were used to emulate ternary statement before it was introduced in Python 2.5.

    value = condition and truth_value or false_value
    

    Warning can be fixed by using standard ternary construct:

    value = truth_value if condition else false_value
    
  • A new refactoring check was added, trailing-comma-tuple, which is emitted when pylint finds an one-element tuple, created by a stray comma. This can suggest a potential problem in the code and it is recommended to use parentheses in order to emphasise the creation of a tuple, rather than relying on the comma itself.

    The warning is emitted for such a construct:

    a = 1,
    

    The warning can be fixed by adding parentheses:

    a = (1, )
    
  • Two new check were added for detecting an unsupported operation over an instance, unsupported-assignment-operation and unsupported-delete-operation. The first one is emitted whenever an object does not support item assignment, while the second is emitted when an object does not support item deletion:

    class A:
        pass
    instance = A()
    instance[4] = 4 # unsupported-assignment-operation
    del instance[4] # unsupported-delete-operation
    
  • A new check was added, relative-beyond-top-level, which is emitted when a relative import tries to access too many levels in the current package.

  • A new check was added, trailing-newlines, which is emitted when a file has trailing new lines.

  • invalid-length-returned check was added, which is emitted when a __len__ implementation does not return a non-negative integer.

  • There is a new extension, pylint.extensions.mccabe, which can be used for computing the McCabe complexity of classes and functions.

    You can enable this extension through --load-plugins=pylint.extensions.mccabe

  • A new check was added, used-prior-global-declaration. This is emitted when a name is used prior a global declaration, resulting in a SyntaxError in Python 3.6.

  • A new message was added, assign-to-new-keyword. This is emitted when used name is known to become a keyword in future Python release. Assignments to keywords would result in SyntaxError after switching to newer interpreter version.

    # While it's correct in Python 2.x, it raises a SyntaxError in Python 3.x
    True = 1
    False = 0
    
    # Same as above, but it'll be a SyntaxError starting from Python 3.7
    async = "async"
    await = "await"
    

Other Changes#

  • We don't emit by default no-member if we have opaque inference objects in the inference results

    This is controlled through the new flag --ignore-on-opaque-inference, which is by default True. The inference can return multiple potential results while evaluating a Python object, but some branches might not be evaluated, which results in partial inference. In that case, it might be useful to still emit no-member and other checks for the rest of the inferred objects.

  • Namespace packages are now supported by pylint. This includes both explicit namespace packages and implicit namespace packages, supported in Python 3 through PEP 420.

  • A new option was added, --analyse-fallback-block.

    This can be used to support both Python 2 and 3 compatible import block code, which means that the import block might have code that exists only in one or another interpreter, leading to false positives when analysed. By default, this is false, you can enable the analysis for both branches using this flag.

  • ignored-argument-names option is now used for ignoring arguments for unused-variable check.

    This option was used for ignoring arguments when computing the correct number of arguments a function should have, but for handling the arguments with regard to unused-variable check, dummy-variables-rgx was used instead. Now, ignored-argument-names is used for its original purpose and also for ignoring the matched arguments for the unused-variable check. This offers a better control of what should be ignored and how. Also, the same option was moved from the design checker to the variables checker, which means that the option now appears under the [VARIABLES] section inside the configuration file.

  • A new option was added, redefining-builtins-modules, for controlling the modules which can redefine builtins, such as six.moves and future.builtins.

  • A new option was added, ignore-patterns, which is used for building a ignore list of directories and files matching the regex patterns, similar to the ignore option.

  • The reports are now disabled by default, as well as the information category warnings.

  • arguments-differ check was rewritten to take in consideration keyword only parameters and variadics.

    Now it also complains about losing or adding capabilities to a method, by introducing positional or keyword variadics. For instance, pylint now complains about these cases:

    class Parent(object):
    
        def foo(self, first, second):
            ...
    
        def bar(self, **kwargs):
            ...
    
        def baz(self, *, first):
            ...
    
    class Child(Parent):
    
        # Why subclassing in the first place?
        def foo(self, *args, **kwargs):
            # mutate args or kwargs.
            super(Child, self).foo(*args, **kwargs)
    
        def bar(self, first=None, second=None, **kwargs):
            ...
            # The overridden method adds two new parameters,
            # which can also be passed as positional arguments,
            # breaking the contract of the parent's method.
    
        def baz(self, first):
            ...
            # Not keyword-only
    
  • redefined-outer-name is now also emitted when a nested loop's target variable is the same as an outer loop.

    for i, j in [(1, 2), (3, 4)]:
        for j in range(i):
            print(j)
    
  • relax character limit for method and function names that starts with _. This will let people to use longer descriptive names for methods and functions with a shorter scope (considered as private). The same idea applies to variable names, only with an inverse rule: you want long descriptive names for variables with bigger scope, like globals.

  • Add InvalidMessageError exception class and replace assert in pylint.utils with raise InvalidMessageError.

  • UnknownMessageError (formerly UnknownMessage) and EmptyReportError (formerly EmptyReport) are now provided by the new pylint.exceptions submodule instead of pylint.utils as before.

  • We now support inline comments for comma separated values in the configurations

    For instance, you can now use the # sign for having comments inside comma separated values, as seen below:

    disable=no-member, # Don't care about it for now
            bad-indentation, # No need for this
            import-error
    

    Of course, interweaving comments with values is also working:

    disable=no-member,
            # Don't care about it for now
            bad-indentation # No need for this
    

    This works by setting the inline comment prefixes accordingly.

  • Added epytext docstring support to the docparams extension.

  • We added support for providing hints when not finding a missing member.

    For example, given the following code, it should be obvious that the programmer intended to use the mail attribute, rather than email.

    class Contribution:
        def __init__(self, name, email, date):
            self.name = name
            self.mail = mail
            self.date = date
    
    for c in contributions:
        print(c.email) # Oups
    

    pylint will now warn that there is a chance of having a typo, suggesting new names that could be used instead.

    $ pylint a.py
    E: 8,10: Instance of 'Contribution' has no 'email' member; maybe 'mail'?
    

    The behaviour is controlled through the --missing-member-hint option. Other options that come with this change are --missing-member-max-choices for choosing the total number of choices that should be picked in this situation and --missing-member-hint-distance, which specifies a metric for computing the distance between the names (this is based on Levenshtein distance, which means the lower the number, the more pickier the algorithm will be).

  • PyLinter.should_analyze_file has a new parameter, is_argument, which specifies if the given path is a pylint argument or not.

    should_analyze_file is called whenever pylint tries to determine if a file should be analyzed, defaulting to files with the .py extension, but this function gets called only in the case where the said file is not passed as a command line argument to pylint. This usually means that pylint will analyze a file, even if that file has a different extension, as long as the file was explicitly passed at command line. Since should_analyze_file cannot be overridden to handle all the cases, the check for the provenience of files was moved into should_analyze_file. This means we now can write something similar with this example, for ignoring every file respecting the desired property, disregarding the provenience of the file, being it a file passed as CLI argument or part of a package.

    from pylint.lint import Run, PyLinter
    
    class CustomPyLinter(PyLinter):
    
         def should_analyze_file(self, modname, path, is_argument=False):
             if respect_condition(path):
                 return False
             return super().should_analyze_file(modname, path, is_argument=is_argument)
    
    
    class CustomRun(Run):
         LinterClass = CustomPyLinter
    
    CustomRun(sys.argv[1:])
    
  • Imports aliased with underscore are skipped when checking for unused imports.

  • bad-builtin and redefined-variable-type are now extensions, being disabled by default. They can be enabled through: --load-plugins=pylint.extensions.redefined_variable_type,pylint.extensions.bad_builtin

    • Imports checker supports new switch allow-wildcard-with-all which disables warning on wildcard import when imported module defines __all__ variable.

  • differing-param-doc is now used for the differing part of the old missing-param-doc, and differing-type-doc for the differing part of the old missing-type-doc.

Bug fixes#

  • Fix a false positive of redundant-returns-doc, occurred when the documented function was using yield instead of return.

  • Fix a false positive of missing-param-doc and missing-type-doc, occurred when a class docstring uses the For the parameters, see magic string but the class __init__ docstring does not, or vice versa.

  • Added proper exception type inference for missing-raises-doc. Now:

    def my_func():
        """"My function."""
        ex = ValueError('foo')
        raise ex
    

    will properly be flagged for missing documentation of :raises ValueError: instead of :raises ex:, among other scenarios.

  • Fix false positives of missing-[raises|params|type]-doc due to not recognizing valid keyword synonyms supported by Sphinx.

  • More thorough validation in MessagesStore.register_messages() to detect conflicts between a new message and any existing message id, symbol, or old_names.

  • We now support having plugins that shares the same name and with each one providing options.

    A plugin can be logically split into multiple classes, each class providing certain capabilities, all of them being tied under the same name. But when two or more such classes are also adding options, then pylint crashed, since it already added the first encountered section. Now, these should work as expected.

    from pylint.checkers import BaseChecker
    
    
    class DummyPlugin1(BaseChecker):
        name = 'dummy_plugin'
        msgs = {'I9061': ('Dummy short desc 01', 'dummy-message-01', 'Dummy long desc')}
        options = (
            ('dummy_option_1', {
                'type': 'string',
                'metavar': '<string>',
                'help': 'Dummy option 1',
            }),
        )
    
    
    class DummyPlugin2(BaseChecker):
        name = 'dummy_plugin'
        msgs = {'I9060': ('Dummy short desc 02', 'dummy-message-02', 'Dummy long desc')}
        options = (
            ('dummy_option_2', {
                'type': 'string',
                'metavar': '<string>',
                'help': 'Dummy option 2',
            }),
        )
    
    
    def register(linter):
        linter.register_checker(DummyPlugin1(linter))
        linter.register_checker(DummyPlugin2(linter))
    
  • We do not yield unused-argument for singledispatch implementations and do not warn about function-redefined for multiple implementations with same name.

    from functools import singledispatch
    
    @singledispatch
    def f(x):
        return 2*x
    
    @f.register(str)
    def _(x):
        return -1
    
    @f.register(int)
    @f.register(float)
    def _(x):
        return -x
    
  • unused-variable checker has new functionality of warning about unused variables in global module namespace. Since globals in module namespace may be a part of exposed API, this check is disabled by default. For enabling it, set allow-global-unused-variables option to false.

  • Fix a false-positive logging-format-interpolation message, when format specifications are used in formatted string. In general, these operations are not always convertible to old-style formatting used by logging module.

  • Added a new switch single-line-class-stmt to allow single-line declaration of empty class bodies (as seen in the example below). Pylint won't emit a multiple-statements message when this option is enabled.

    class MyError(Exception): pass
    
    • too-many-format-args and too-few-format-args are emitted correctly (or not emitted at all, when exact count of elements in RHS cannot be inferred) when starred expressions are used in RHS tuple. For example, code block as shown below detects correctly that the used tuple has in fact three elements, not two.

    meat = ['spam', 'ham']
    print('%s%s%s' % ('eggs', *meat))
    
  • cyclic-import checker supports local disable clauses. When one of cycle imports was done in scope where disable clause was active, cycle is not reported as violation.

Removed Changes#

  • pylint-gui was removed, because it was deemed unfit for being included in pylint. It had a couple of bugs and misfeatures, its usability was subpar and since its development was neglected, we decided it is best to move on without it.

  • The HTML reporter was removed, including the --output-format=html option. It was lately a second class citizen in Pylint, being mostly neglected. Since we now have the JSON reporter, it can be used as a basis for building more prettier HTML reports than what Pylint can currently generate. This is part of the effort of removing cruft from Pylint, by removing less used features.

  • The --files-output option was removed. While the same functionality cannot be easily replicated, the JSON reporter, for instance, can be used as a basis for generating the messages per each file.

  • --required-attributes option was removed.

  • --ignore-iface-methods option was removed.

  • The --optimize-ast flag was removed.

    The option was initially added for handling pathological cases, such as joining too many strings using the addition operator, which was leading pylint to have a recursion error when trying to figure out what the string was. Unfortunately, we decided to ignore the issue, since the pathological case would have happen when the code was parsed by Python as well, without actually reaching the runtime step and as such, we decided to remove the error altogether.

  • epylint.py_run's script parameter was removed.

    Now epylint.py_run is always using the underlying epylint.lint method from the current interpreter. This avoids some issues when multiple instances of pylint are installed, which means that epylint.py_run might have ran a different epylint script than what was intended.