Restrict visibility or mutability of Python object attributes

pyprotect is a python module that provides API to restrict visibility or mutability of selected Python object attributes in a robust manner.

The key functions in the pyprotect module API - private() and protect() wrap the python object (like a Proxy) to restrict visibility or mutability of selected attributes of the wrapped object, while allowing the wrapping object to behave virtually identical to the wrapped object.

Features

Can wrap virtually any Python object - instances, classes (types), modules, methods, classmethods, instancemethods, staticmethods, partials, lambdas.
Tested on Python 2.7 and Python 3.5, 3.6, 3.7, 3.8, 3.9, 3.10, 3.11
Tested on the following distributions with the latest versions of python2, python3, pypy3, pypy shipped by the respective distributions:
- Ubuntu Jammy 22.04
- Arch linux (20220101)
- Fedora 37
- Alpine Linux 3.15 (Alpine 3.16 does not have python2-dev)
Has extensive unit (functional) tests - in tests directory.

Quick start
Classes
Features of key classes
API
Calling wrap operations multiple times
Python rules for attributes of type ‘property’:
What kind of python objects can be wrapped?
Work in progress
Changelog

Quick start


freeze(o: object) -> Frozen:

If o is immutable (e.g. int , string), returns o UNCHANGED
If o is Wrapped, returns o UNCHANGED if object WRAPPPED INSIDE o is immutable, returns Frozen otherwise
If o is Frozen, returns o UNCHANGED
If o is FrozenPrivate, FrozenProtected or FrozenPrivacyDict, returns o UNCHANGED
If o is Private, returns FrozenPrivate
If o is Protected, returns FrozenProtected
Otherwise, returns Frozen

Object returned prevents modification of ANY attribute

private(o: object, frozen: bool = False) -> object:

If frozen is False:
- If o is an instance of Private, returns o UNCHANGED
  - If ‘o’ is an instance of Protected, returns o UNCHANGED
  - If frozen is True:
  - If o is an instance of Private, returns freeze(o) (FrozenPrivate)
  - If o is an instance of Protected, returns freeze(o) (FrozenProtected)
  - Otherwise:
    If frozen is True, returns FrozenPrivate; returns Private otherwise

protect(
    o: object frozen: bool = False,
    dynamic: bool = True,
    hide_private: bool = False,
    ro_data: bool = False,
    ro_method: bool = True,
    ro: List[str] = [],
    rw: List[str] = [],
    hide: List[str] = []
) -> object:
# o-->object to be wrapped

Returns—>Instance of FrozenProtected if frozen; Instance of Protected otherwise

If protect() is called on an object ‘o’ that is an instance of Protected, protect() will merge the protect() rules, enforcing the most restrictive combination among the two sets of protect() options:

hide and hide_private are OR-ed
ro_method, ro_data and ro are OR-ed
rw is AND-ed, but rw of second protect overrides ro* of second protect but not the first protect.

In short, by calling protect() a second time (or multiple times):

Additoinal attributes can be hidden
Additional attributes can be made read-only
No previously hidden attribute will become visible
No previously read-only attribute will become mutable

Visibility and mutability of attributes with protect() method

Option	Attribute Type	Restricts Visibility	Restricts Mutability
frozen	Any	NO	YES
hide_private	Private attributes	YES	YES (Indirect)
ro_data	Data attributes	NO	YES
ro_method	Method attributes	NO	YES
ro	ANY	NO	YES
rw	ANY	NO	YES
hide	ANY	YES	YES (Indirect)

Classes

class diagram

Features of key classes

Wrapped

Visibility: No additional restrictions
Mutability: No additional restrictions
Frozen
Visibility: Does not additionally restrict visibility of any attributes in wrapped object accessed through wrapping object
Mutability: Prevents modification of ANY attribute
Private
Visibility:
- Cannot access traditionally ‘private’ mangled python attributes
- Cannot modify traditionally private attributes (form ‘var‘)
- Attributes not part of dir(wrapped_object) are not visible
Mutability:
- The following attributes of wrapped object are NEVER writeable:
  __class__, __dict__, __delattr__, __setattr__, __slots__, __getattribute__
- Traditional (mangled) Python private vars are ALWAYS hidden
- Private vars (form _var) will be read-only
- Attributes cannot be added or removed
- Attributes that are properties are ALWAYS visible AND WRITABLE (except if ‘frozen‘ is used)
  - Properties indicate an intention of class author to expose them
  - Whether they are actually writable depends on whether class author implemented property.setter
    FrozenPrivate
Visibility: Same as Private
Mutability: Prevents modification of ANY attribute
Protected
Features of Private PLUS allows further restriction of:
- Which attributes are VISIBLE
- Which attributes are WRITEABLE
Default settings:
- Features of Private - see above
- dynamic == True
  Attribute additions, deletions, type changes automatically visible
- ro_method == True: Method attributes will be read-only
- All other non-private data attributes are read-write
  FrozenProtected
Features of Protected PLUS prevents modification of ANY attribute

API

Wrapping API

freeze


freeze(o: object) -> Frozen:

If o is immutable (e.g. int , string), returns o UNCHANGED
If o is Wrapped, returns o UNCHANGED if object WRAPPPED INSIDE o is immutable, returns Frozen otherwise
If o is Frozen, returns o UNCHANGED
If o is FrozenPrivate, FrozenProtected or FrozenPrivacyDict, returns o UNCHANGED
If o is Private, returns FrozenPrivate
If o is Protected, returns FrozenProtected
Otherwise, returns Frozen

Object returned prevents modification of ANY attribute

private

private(o: object, frozen: bool = False) -> object:

If frozen is False:
- If o is an instance of Private, returns o UNCHANGED
  - If ‘o’ is an instance of Protected, returns o UNCHANGED
  - If frozen is True:
  - If o is an instance of Private, returns freeze(o) (FrozenPrivate)
  - If o is an instance of Protected, returns freeze(o) (FrozenProtected)
  - Otherwise:
    If frozen is True, returns FrozenPrivate; returns Private otherwise

protect

protect(
    o: object frozen: bool = False,
    dynamic: bool = True,
    hide_private: bool = False,
    ro_data: bool = False,
    ro_method: bool = True,
    ro: List[str] = [],
    rw: List[str] = [],
    hide: List[str] = []
) -> object:
# o-->object to be wrapped

Returns—>Instance of FrozenProtected if frozen; Instance of Protected otherwise

hide and hide_private are OR-ed
ro_method, ro_data and ro are OR-ed
rw is AND-ed, but rw of second protect overrides ro* of second protect but not the first protect.

In short, by calling protect() a second time (or multiple times):

Additoinal attributes can be hidden
Additional attributes can be made read-only
No previously hidden attribute will become visible
No previously read-only attribute will become mutable

Visibility and mutability of attributes with protect() method

Option	Attribute Type	Restricts Visibility	Restricts Mutability
frozen	Any	NO	YES
hide_private	Private attributes	YES	YES (Indirect)
ro_data	Data attributes	NO	YES
ro_method	Method attributes	NO	YES
ro	ANY	NO	YES
rw	ANY	NO	YES
hide	ANY	YES	YES (Indirect)

wrap

wrap(o: object) -> Wrapped:

Should behave just like the wrapped object, except following attributes cannot be modified:
__getattribute__, __delattr__, __setattr__, __slots__
Explicitly does NOT support pickling, and will raise pickle.PicklingError
Does NOT protect CLASS (or __class__) of wrapped object from modification
Does NOT protect __dict__ or __slots__

Useful for testing if wrapping is failing for a particular type of object

Checking types of wrapped objects

isfrozen

isfrozen(x: object) -> bool

x was created using freeze() or private(o, frozen=True) or protect(o, frozen=True)

isimmutable

isimmutable(x: object) -> bool

x is known to be immutable

isprivate

isprivate(x: object) -> bool

x was created using private()

isprotected

isprotected(x: object) -> bool

x was created using protect()

Checking properties of objects inside wrapped objects

contains

contains(w: object, o: object) -> bool

If w is a wrapped object (iswrapped(w) is True), returns whether w wraps o
Otherwise unconditionally returns False

help_protected

help_protected(x: object) -> None

If x wraps o, executes help(o)
Otherwise executes help(x)

id_protected

id_protected(x: object) -> int

if x is a wrapped object (iswrapped(x) is True) and x wraps o, returns id(o)
Otherwise returns id(x)

isinstance_protected

isinstance_protected(x: object, t: type) -> bool

If x is a wrapped object (iswrapped(x) is True) and x wraps o, returns isinstance(o, t)
Otherwise returns isinstance(x, t)

isreadonly

isreadonly(x: object, a: str) -> bool

If x is a wrapped object - with iswrapped(x) == True - and x wraps o, isreadonly(x, a) returns whether rules of wrapper make attribute a read-only when accessed through x
This represents rule of wrapped object - does not guarantee thato has attributea or that setting attribute a in object o will not raise any exception
If x is not a wrapped object (iswrapped(x) is False) , unconditionally returns False

instance_of_protected

instance_of_protected(x: object, o: object) -> bool

If x is a wrapped object - with iswrapped(x) == True - and x wraps o, instance_of_protected(x, o) returns True if and only if isinstance(x, type(o))
If x is not a wrapped object - iswrapped(x) == False - instance_of_protected(x, o) returns isinstance(x, o)

isvisible

isvisible(x: object, a: str) -> bool

Returns False if and only if iswrapped(x) is True AND x makes attribute a invisible if present in wrapped object

This represents rule of wrapped object - does not guarantee that wrapped object has attribute a or that accessing attribute a in object x will not raise any exception

If x is not a wrapped object, unconditionally returns False

same_class_protected

same_class_protected(c: type, w: object) -> bool

If iswrapped(w) and w wraps o: Returns (c is type(o))

Otherwise: returns (c is type(w))

subclass_of_protected

subclass_of_protected(x: object, w: object) -> bool

If iswrapped(w) and w wraps o: Returns issubclass(x, type(o))

Otherwise: returns issubclass(x, w)

pyprotect module metadata

immutable_builtin_attributes

immutable_builtin_attributes() -> Set[str]

Returns—>set of str: attributes in builtins that are immutable
Used in unit tests

always_delegated_attributes

always_delegated_attributes() -> set(str)

Attributes that are always delegated to wrapped object

attribute_protected

attribute_protected() -> str

Name of special attribute in Wrapped objects

hidden_pickle_attributes

hidden_pickle_attributes() -> set(str)

Attributes that are never visible in object ‘o’ if iswrapped(o) - to disallow pickling

never_writeable

never_writeable() -> set(str)

Attributes that are never writeable in object o if iswrapped(o)

never_writeable_private

never_writeable_private() -> set(str)

Attributes that are never writeable in object o if isprivate(o)

Calling wrap operations multiple times

In the table below:

The left-most column shows starting state.
The top row shows operation applied to the starting state.
The intersecting cell shows the result.
UNCH represents operation returning the starting state unchanged

Operation 🡆 On type 🡇	wrap	freeze	private	private + frozen	protect	protect + frozen
Ordinary object iswrapped(x) is False	Wrapped	Frozen (2)	Private	FrozenPrivate	Protected	FrozenProtected
Wrapped	UNCH	Frozen (2)	Private	FrozenPrivate	Protected	FrozenProtected
Frozen	Wrapped (2)	UNCH (2)	FrozenPrivate	FrozenPrivate	FrozenProtected	FrozenProtected
Private	UNCH	FrozenPrivate	UNCH	FrozenPrivate	Protected	FrozenProtected
FrozenPrivate	UNCH	UNCH	UNCH	UNCH	FrozenProtected	FrozenProtected
Protected	UNCH	FrozenProtected	UNCH	FrozenProtected	Protected (1)	FrozenProtected (1)
FrozenProtected	UNCH	UNCH	UNCH	UNCH	FrozenProtected (1)	FrozenProtected (1)

protect() applied twice will merge the protect() rules, enforcing the most restrictive combination among the two sets of protect() options:

hide and hide_private are OR-ed
ro_method, ro_data and ro are OR-ed
rw is AND-ed, but rw of second protect overrides ro* of second protect but not the first protect.

In short, by calling protect() a second time (or multiple times):

- Additoinal attributes can be hidden
- Additional attributes can be made read-only

but:

- No previously hidden attribute will become visible
- No previously read-only attribute will become mutable

If x is an immutable object (e.g. int, str …) having isimmutable(x) \=\= True, freeze(x) returns x and iswrapped(freeze(x)) will be False.

For all other objects x, having isimmutable(x) \=\= False, freeze(x) will return a Frozen object having iswrapped(freeze(x)) == True

For all other wrapped objects w, created with private(x) or protect(x), freeze(w) will always return a Wrapped object with iswrapped(w) == True

Python rules for attributes of type ‘property’:

Properties are defined in the CLASS, and cannot be changed in the object INSTANCE
Properties cannot be DELETED
Properties cannot be WRITTEN to unless property has a ‘setter’ method defined in the CLASS
These rules are implemented by the python language (interpreter) and Protected class does not enforce or check

What kind of python objects can be wrapped?

Pretty much anything. pyprotect only mediates attribute access using object.__getattribute__, object.__setattr__ and object.__delatr__. If these methods work on your object, your object can be wrapped

Work in progress

Uploading to pypi.org
Test cases required

Changelog

Jan-20-2022

Project name on github changed to pyprotect to match the name of the module. This has been long-pending. The old github link redirects to the new project name.
Jan-15-2022
Started signing commits with my GPG key and displaying ‘verified’ for signed commits
Dec-08-2022
A number of parameters to protect() have been discontinued. See list and reasons below, as well as how to achieve the same effect without thos parameters (sometimes, it takes more work). Most of them would be realistically useful very rarely, and / or do not align with what I call ‘idiomatic python’.

hide_all, hide_method, hide_dunder
So-called ‘special methods’ in Python serve an important functional roles - especially in emulating containers (tuples, lists, sets, dicts), emulating numeric types supporting arithmetic operators, numeric comparisons etc. If such specific ‘dunder methods’ were hidden, it would definitely affect the behavior of the wrapped object. hide_dunder would hide all such special methods. hide_method would in addition hide all methods in the objest. hide_all would hide all object attributes, making the object virtually useless, and the option useful in testing (if at all).

hide_data
In most cases, hiding all non-method data attributes will make the object less useful / cripple the expected usage of the object. Specific use-cases can be achieved using the ‘hide’ parameter.

ro_dunder
Seems like it can be replaced by using ro_method’ Mostly, in ‘idiomatic python’, methods of a class / instance are not mutated from outside the class / instance. This expected ‘idiomatic python’ behavior can be achieved with ‘ro_method’.

ro_all
Is unnecessary, since ‘frozen’ can be used instead.

add
In my opinion, ‘add’ does not align with ‘idiomatic python’. While Python allows users of a class / instance adding attributes to the class / instance, that is not the expected style. Based on this, I have decided to promote that ‘idiomatic style’, and prevent adding / deleting any attributes in a Private or Protected object.

show
It is unnecessary, since all attributes are visible by default in Python. Only ‘hide’ or ‘hide_private’ will hide any attributes.

This leaves the following (incidentally also reducing testing load):

- Visibility:
    - hide_private
    - hide
- Mutability:
    - ro_data
    - ro_method
    - ro
    - rw
    - frozen
- Behavior:
    - dynamic

Restrict visibility or mutability of Python object attributes

Features

Table of Contents

Quick start

Classes

Features of key classes

Wrapped

Frozen

Private

FrozenPrivate

Protected

FrozenProtected

API

Wrapping API

freeze

private

protect

wrap

Checking types of wrapped objects

isfrozen

isimmutable

isprivate

isprotected

Checking properties of objects inside wrapped objects

contains

help_protected

id_protected

isinstance_protected

isreadonly

instance_of_protected

isvisible

same_class_protected

subclass_of_protected

pyprotect module metadata

immutable_builtin_attributes

always_delegated_attributes

attribute_protected

hidden_pickle_attributes

never_writeable

never_writeable_private

Calling wrap operations multiple times

Python rules for attributes of type ‘property’:

What kind of python objects can be wrapped?

Work in progress

Changelog

Jan-20-2022

Jan-15-2022

Dec-08-2022