lockup 2.0rc1

Immutable and concealed attributes for classes, modules, and namespaces.

API Documentation (stable) | API Documentation (current) | Code of Conduct | Contribution Guide

Overview

Enables the creation of classes, modules, and namespaces on which the following properties are true:

• All attributes are immutable. Immutability increases code safety by discouraging monkey-patching and preventing changes to state, accidental or otherwise.

  >>> import getpass
  >>> def steal_password( prompt = 'Password: ', stream = None ):
  ...     pwned = getpass.getpass( prompt = prompt, stream = stream )
  ...     # Send host address, username, and password to Dark Web collector.
  ...     return pwned
  ...
  >>> import lockup
  >>> lockup.reclassify_module( getpass )
  >>> getpass.getpass = steal_password
  Traceback (most recent call last):
  ...
  lockup.exceptions.ImpermissibleAttributeOperation: Attempt to assign immutable attribute 'getpass' on module 'getpass'.

  >>> import lockup
  >>> ns = lockup.create_namespace( some_constant = 6 )
  >>> ns.some_constant = 13
  Traceback (most recent call last):
  ...
  lockup.exceptions.ImpermissibleAttributeOperation: Attempt to assign immutable attribute 'some_constant' on class 'lockup.Namespace'.

• Non-public attributes are concealed. Concealment means that the dir function will report a subset of attributes that are intended for programmers to use… without exposing internals.

  >>> import lockup
  >>> class Demo( metaclass = lockup.Class ):
  ...     _foo = 'Semi-private class variable.'
  ...     hello = 'Public class variable.'
  ...     def __len__( self ): return 1
  ...
  >>> dir( Demo )
  ['hello']

In addition to the above, the package also provides the ability to apprehend “fugitive” exceptions attempting to cross API boundaries. Various auxiliary functionalities are provided as well; these are used internally within the package but are deemed useful enough for public consumption. Please see the documentation for more details.

Quick Tour

Module

Let us consider the mutable os module from the Python standard library and how we can alter “constants” that may be used in many places:

>>> import os
>>> type( os )
<class 'module'>
>>> os.O_RDONLY
0
>>> os.O_RDONLY = os.O_RDWR
>>> os.O_RDONLY
2
>>> os.O_RDONLY = 0

Now, let us see what protection it gains from becoming immutable:

>>> import os
>>> import lockup
>>> lockup.reclassify_module( os )
>>> type( os )
<class 'lockup.module.Module'>
>>> # How? https://docs.python.org/3/reference/datamodel.html#customizing-module-attribute-access
>>> os.O_RDONLY = os.O_RDWR
Traceback (most recent call last):
...
lockup.exceptions.ImpermissibleAttributeOperation: Attempt to assign immutable attribute 'O_RDONLY' on module 'os'.
>>> del os.O_RDONLY
Traceback (most recent call last):
...
lockup.exceptions.ImpermissibleAttributeOperation: Attempt to delete indelible attribute 'O_RDONLY' on module 'os'.

Class Factory

Let us monkey-patch a mutable class:

>>> class A:
...     def expected_functionality( self ): return 42
...
>>> a = A( )
>>> a.expected_functionality( )
42
>>> def monkey_patch( self ):
...     return 'I selfishly change behavior upon which other consumers depend.'
...
>>> A.expected_functionality = monkey_patch
>>> a = A( )
>>> a.expected_functionality( )
'I selfishly change behavior upon which other consumers depend.'

Now, let us try to monkey-patch an immutable class:

>>> import lockup
>>> class B( metaclass = lockup.Class ):
...     def expected_functionality( self ): return 42
...
>>> b = B( )
>>> b.expected_functionality( )
42
>>> def monkey_patch( self ):
...     return 'I selfishly change behavior upon which other consumers depend.'
...
>>> B.expected_functionality = monkey_patch
Traceback (most recent call last):
...
lockup.exceptions.ImpermissibleAttributeOperation: Attempt to assign immutable attribute 'expected_functionality' on class ...
>>> del B.expected_functionality
Traceback (most recent call last):
...
lockup.exceptions.ImpermissibleAttributeOperation: Attempt to delete indelible attribute 'expected_functionality' on class ...

Note

Only class attributes are immutable. Instances of immutable classes will have mutable attributes without additional intervention beyond the scope of this package.

Namespace Factory

An alternative to types.SimpleNamespace is provided. First, let us observe the behaviors on a standard namespace:

>>> import types
>>> sn = types.SimpleNamespace( run = lambda: 42 )
>>> sn
namespace(run=<function <lambda> at ...>)
>>> sn.run( )
42
>>> type( sn )
<class 'types.SimpleNamespace'>
>>> sn.__dict__
{'run': <function <lambda> at ...>}
>>> type( sn.run )
<class 'function'>
>>> sn.run = lambda: 666
>>> sn.run( )
666
>>> sn( )  # doctest: +SKIP
Traceback (most recent call last):
...
TypeError: 'types.SimpleNamespace' object is not callable

Now, let us compare those behaviors to an immutable namespace:

>>> import lockup
>>> ns = lockup.create_namespace( run = lambda: 42 )
>>> ns
NamespaceClass( 'Namespace', ('object',), { ... } )
>>> ns.run( )
42
>>> type( ns )
<class 'lockup.factories.NamespaceClass'>
>>> ns.__dict__
mappingproxy({...})
>>> type( ns.run )
<class 'function'>
>>> ns.run = lambda: 666
Traceback (most recent call last):
...
lockup.exceptions.ImpermissibleAttributeOperation: Attempt to assign immutable attribute 'run' on class 'lockup.Namespace'.
>>> ns.__dict__[ 'run' ] = lambda: 666
Traceback (most recent call last):
...
TypeError: 'mappingproxy' object does not support item assignment
>>> ns( )
Traceback (most recent call last):
...
lockup.exceptions.ImpermissibleOperation: Impermissible instantiation of class 'lockup.Namespace'.

Also of note is that we can define namespace classes directly, allowing us to capture imports for internal use in a module without publicly exposing them as part of the module API, for example:

>>> import lockup
>>> class __( metaclass = lockup.NamespaceClass ):
...     from os import O_RDONLY, O_RDWR
...
>>> __.O_RDONLY
0

The above technique is used internally within this package itself.

Interception

If a particular exceptional condition is not anticipated in Python code, a “fugitive” exception can escape across the boundary of a published API. If you have told the consumers of the API that it will only emit certain classes of exceptions, then consumers might not handle exceptions outside of the expected classes, i.e., fugitive exceptions. If you apprehend all fugitives at the API boundary, then you can guarantee to your consumers that they will only need to anticipate certain classes of exceptions.

Here is an example with an interceptor, which includes fugitive exception apprehension, that this package uses internally:

>>> from lockup.exceptions import InvalidState
>>> from lockup.interception import our_interceptor
>>> @our_interceptor
... def divide_by_zero( number ): return number / 0
...
>>> try: divide_by_zero( 42 )
... except InvalidState as exc:
...     type( exc ), type( exc.__cause__ ), str( exc )
...
(<class 'lockup.exceptions.InvalidState'>, <class 'ZeroDivisionError'>, "Apprehension of fugitive exception of class 'builtins.ZeroDivisionError' at boundary of function 'divide_by_zero' on module '__main__'.")

As can be seen, the ZeroDivisionError is in the custody of an exception that is of an expected class.

You can create your own interceptors with custom fugitive apprehension behaviors using the create_interception_decorator function.

Compatibility

This package has been verified to work on the following Python implementations:

• CPython

• PyPy

• Pyston

  Warning

  Support for Pyston may disappear in the future as the maintainers have decided to invest in a JIT module for CPython rather than a separate implementation.

It likely works on others as well, but please report if it does not.

More Flair

…than the required minimum

Changelog

v2.0.0 (not released yet)

API

• No more separate API for package-internal development. Everything is now exposed as part of an auxiliary public API as opposed to the primary public API.

  Note

  Some parts of the auxiliary public API may be refactored into separate packages at a later point.

• Provide create_interception_decorator function which creates function decorators that can apprehend “fugitive” exceptions before they escape across the boundary of a public API. Fugitive exceptions are exceptions which are unexpected and which should have been caught internally.

• Provide reassign_class_factory function which allows for a class to be assigned a new factory class (“metaclass”). This can even be used on a class factory class itself, resulting in a factory class similar to how type behaves. This package uses it internally, when possible, to allow class factory classes to enforce attribute concealment and immutability on themselves and not just their instances. But, it can be put to other purposes too.

• Provide exception management utilities, including factories which can inject labels into instances of a single omniexception class as an alternative to working with a class hierarchy. This package internally uses the utilities to create exceptions, which have good descriptions and helpful labels.

• Provide nomenclatural utilities which determine the classification of objects that are provided to them. These are useful for the creation of more helpful exception messages or log entries. This package internally uses the utilities to create descritpive exception messages. A suite of exception factories, which use these utilities, is also exposed.

• Provide validation utilities which return back the object they validation if the validation is successful. Otherwise, they raise a validation error. This allows for multiple validators to be fluently applied in succession. This package internally uses the validators on arguments to functions that are part of its public API.

• Provide visibility utilities which determine if an attribute is considered public or non-public and what attributes should be concealed on an object. This package uses the utilities internally to conceal non-public attributes on classes, modules, and namespaces. But, they can be put to other purposes as well.

Python Support

• Remove CPython 3.6 because it is past end-of-life.

• Deprecate Pyston because of its new development direction.

• Add PyPy 3.9.

v1.1.0

Documentation

• Officially verify and mention PyPy and Pyston support.