Skip to main content

aiootp - an asynchronous pseudo one-time pad based crypto and anonymity library.

Project description

aiootp python package logo

aiootp - Asynchronous pseudo one-time pad based crypto and anonymity library.

aiootp is an asynchronous library providing access to cryptographic primatives and abstractions, transparently encrypted / decrypted file I/O and databases, as well as powerful, pythonic utilities that simplify data processing & cryptographic procedures in python code. This library’s online, salt reuse / misuse resistant, tweakable AEAD cipher, called Chunky2048, is an implementation of the pseudo one-time pad. The aim is to create a simple, standard, efficient implementation that’s indistinguishable from the unbreakable one-time pad cipher; to give users and applications access to user-friendly cryptographic tools; and, to increase the overall security, privacy, and anonymity on the web, and in the digital world. Users will find aiootp to be easy to write, easy to read, and fun.

Important Disclaimer

aiootp is experimental software that works with Python 3.7+. It’s a work in progress. The programming API could change with future updates, and it isn’t bug free. aiootp provides powerful security tools and misc utilities that’re designed to be developer-friendly and privacy preserving. As a security tool, aiootp needs to be tested and reviewed extensively by the programming and cryptography communities to ensure its implementations are sound. We provide no guarantees. This software hasn’t yet been audited by third-party security professionals.

version python-versions license code-style python-wheel-availability linux-build-status windows-build-status macos-build-status

Quick Install

$ sudo apt-get install python3-setuptools python3-pip

$ pip3 install --user --upgrade pip typing aiootp

Run Tests

$ cd ~/aiootp/tests

$ coverage run --source aiootp -m pytest -vv

Table Of Contents

Transparently Encrypted Databases ………….. Table Of Contents

The package’s AsyncDatabase & Database classes are very powerful data persistence utilities. They automatically handle encryption & decryption of user data & metadata, providing a pythonic interface for storing & retrieving any bytes or JSON serializable objects. They’re designed to seamlessly bring encrypted bytes at rest to users as dynamic objects in use.

Ideal Initialization ……………………… Table Of Contents

Make a new user key with a fast, cryptographically secure pseudo-random number generator. Then this strong 64-byte key can be used to create a database object.

from aiootp import acsprng, AsyncDatabase

key = await acsprng()

db = await AsyncDatabase(key)

User Profiles ……………………………. Table Of Contents

With User Profiles, passphrases may be used instead to open a database. Often, passwords & passphrases contain very little entropy. So, they aren’t recommended for that reason. However, profiles provide a succinct way to use passphrases more safely. They do this by deriving strong keys from low entropy user input using the memory/cpu hard passcrypt algorithm, & a secret salt which is automatically generated & stored on the user’s filesystem.

# Automatically convert any available user credentials into

# cryptographic tokens which help to safely open databases ->

db = await AsyncDatabase.agenerate_profile(

    b"",     # Here an unlimited number of bytes-type
                           # arguments can be passed as additional
    b"",  # optional credentials.



    salt=b"optional salt keyword argument",
              # Optional passcrypt configuration:
    mb=256,   # The memory cost in Mibibytes (MiB)

    cpu=2,    # The computational complexity & number of iterations

    cores=8,  # How many parallel processes passcrypt will utilize


Tags ……………………………………. Table Of Contents

Data within databases are primarily organized by Tags. Tags are simply string labels, and the data stored under them can be any bytes or JSON serializable objects.

async with db:

    # Using bracketed assignment adds tags to the cache

    db["tag"] = {"data": "can be any JSON serializable object"}

    db["hobby"] = b"fash smasher"

    db["bitcoin"] = "0bb6eee10d2f8f45f8a"

    db["lawyer"] = {"#": "555-555-1000", "$": 13000.50}

    db["safehouses"] = ["Dublin Forgery", "NY Insurrection"]

    # Changes in the cache are saved to disk when the context closes.

# View an instance's tags ->

>>> {'tag', 'hobby', 'bitcoin', 'lawyer', 'safehouses'}

# View the filenames that locate the data for each tag ->

>>> {'0z0l10btu_yd-n4quc8tsj9baqu8xmrxz87ix',

# There are various ways of working with tags ->

await db.aset_tag("new_tag", ["data", "goes", "here"])  # stored only in cache

await db.aquery_tag("new_tag")  # reads from disk if not in the cache
>>> ['data', 'goes', 'here']

tag_path = db.path / await db.afilename("new_tag")

"new_tag" in db
>>> True

tag_path.is_file()  # the tag is saved in the cache, not to disk yet
>>> False

await db.asave_tag("new_tag")

tag_path.is_file()  # now it's saved to disk
>>> True

# This removes the tag from cache, & any of its unsaved changes ->

await db.arollback_tag("new_tag")

# Or, the user can take the tag out of the database & the filesystem ->

await db.apop_tag("new_tag")
>>> ['data', 'goes', 'here']

"new_tag" in db
>>> False

>>> False

Access to data is open to the user, so care must be taken not to let external API calls touch the database without accounting for how that can go wrong.

Metatags ………………………………… Table Of Contents

Metatags are used to organize data by string names & domain separate cryptographic material. They are fully-fledged databases all on their own, with their own distinct key material too. They’re accessible from the parent through an attribute that’s added to the parent instance with the same name as the metatag. When the parent is saved, or deleted, then their descendants are also.

# Create a metatag database ->

molly = await db.ametatag("molly")

# They can contain their own sets of tags (and metatags) ->

molly["hobbies"] = ["skipping", "punching"]


# The returned metatag & the reference in the parent are the same ->

assert molly["hobbies"] is db.molly["hobbies"]

assert isinstance(molly, AsyncDatabase)

# All of an instance's metatags are viewable ->

>>> {'molly'}

# Delete a metatag from an instance ->

await db.adelete_metatag("molly")

>>> set()

assert not hasattr(db, "molly")

Basic Management …………………………. Table Of Contents

There’s a few settings & public methods on databases for users to manage their instances & data. This includes general utilities for saving & deleting databases to & from the filesystem, as well as fine-grained controls for how data is handled.

# The path attribute is set within the instance's __init__

# using a keyword-only argument. It's the directory where the

# instance will store all of its files.

>>> PosixPath('site-packages/aiootp/aiootp/databases')

# Write database changes to disk with transparent encryption ->

await db.asave_database()

# Entering the instance's context also saves data to disk ->

async with db:

    print("Saving to disk...")

# Delete a database from the filesystem ->

await db.adelete_database()

As databases grow in the number of tags, metatags & the size of data within, it becomes desireable to load data from them as needed, instead of all at once into the cache during initialization. This is why the preload boolean keyword-only argument is set to False by default.

# Let's create some test values to show the impact preloading has ->

async with (await AsyncDatabase(key, preload=True)) as db:

    db["favorite_foods"] = ["justice", "community"]

    await db.ametatag("exercise_routines")

    db.exercise_routines["gardening"] = {"days": ["moday", "wednesday"]}

    db.exercise_routines["swimming"] = {"days": ["thursday", "saturday"]}

# Again, preloading into the cache is toggled off by default ->

uncached_db = await AsyncDatabase(key)

# To retrieve elements, ``aquery_tag`` isn't necessary when

# preloading is used, since the tag is already in the cache ->

async with uncached_db:

    >>> ["justice", "community"]

    >>> None

    value = await uncached_db.aquery_tag("favorite_foods", cache=True)

    assert value == ["justice", "community"]

    assert uncached_db["favorite_foods"] == ["justice", "community"]

    # Metatags will be loaded, but their tags won't be ->

    assert type(uncached_db.exercise_routines) == AsyncDatabase

    >>> None

    await uncached_db.exercise_routines.aquery_tag("gardening", cache=True)
    >>> {"days": ["moday", "wednesday"]}

    >>> {"days": ["moday", "wednesday"]}

    # But, tags can also be queried without caching their values,

    value = await uncached_db.exercise_routines.aquery_tag("swimming")

    >>> {"days": ["thursday", "saturday"]}

    >>> None

    # However, changes to mutable values won't be transmitted to the

    # database if they aren't retrieved from the cache ->


    >>> {"days": ["thursday", "saturday", "sunday"]}

    await uncached_db.exercise_routines.aquery_tag("swimming")
    >>> {"days": ["thursday", "saturday"]}

Mirrors …………………………………. Table Of Contents

Database mirrors allow users to make copies of all files within a database under new encryption keys. This is useful if users simply want to make backups, or if they’d like to update / change their database keys.

# A unique login key / credentials are needed to create a new

# database ->

new_key = await acsprng()

new_db = await AsyncDatabase(new_key)

# Mirroring an existing database is done like this ->

await new_db.amirror_database(db)

assert (

    await new_db.aquery_tag("favorite_foods")

    is await db.aquery_tag("favorite_foods")


# If the user is just updating their database keys, then the old

# database should be deleted ->

await db.adelete_database()

# Now, the new database can be saved to disk & given an appropriate

# name ->

async with new_db as db:


Public Cryptographic Functions …………….. Table Of Contents

Although databases handle encryption & decryption automatically, users may want to utilize their databases’ keys to do custom cryptographic procedures manually. There are a few public functions available to users if they should want such functionality.

Encrypt / Decrypt ………………………… Table Of Contents
# Either JSON serializable or bytes-type data can be encrypted ->

json_plaintext = {"some": "JSON data can go here..."}

bytes_plaintext = b"some bytes plaintext goes here..."

token_plaintext = b"some token data goes here..."

json_ciphertext = await db.ajson_encrypt(json_plaintext)

bytes_ciphertext = await db.abytes_encrypt(bytes_plaintext)

token_ciphertext = await db.amake_token(token_plaintext)

# Those values can just as easily be decrypted ->

assert json_plaintext == await db.ajson_decrypt(json_ciphertext)

assert bytes_plaintext == await db.abytes_decrypt(bytes_ciphertext)

assert token_plaintext == await db.aread_token(token_ciphertext)

# Filenames may be added to classify ciphertexts. They also alter the

# key material used during encryption in such a way, that without the

# correct filename, the data cannot be decrypted ->

filename = "grocery-list"

groceries = ["carrots", "taytoes", "rice", "beans"]

ciphertext = await db.ajson_encrypt(groceries, filename=filename)

assert groceries == await db.ajson_decrypt(ciphertext, filename=filename)

await db.ajson_decrypt(ciphertext, filename="wrong filename")
>>> "InvalidSHMAC: Invalid StreamHMAC hash for the given ciphertext."

# Time-based expiration of ciphertexts is also available for all

# encrypted data this package produces ->

from aiootp.asynchs import asleep

await asleep(6)

await db.ajson_decrypt(json_ciphertext, ttl=1)
>>> "TimestampExpired: Timestamp expired by <5> seconds."

await db.abytes_decrypt(bytes_ciphertext, ttl=1)
>>> "TimestampExpired: Timestamp expired by <5> seconds."

await db.aread_token(token_ciphertext, ttl=1)
>>> "TimestampExpired: Timestamp expired by <5> seconds."

# The number of seconds that are exceeded may be helpful to know. In

# which case, this is how to retrieve that integer value ->


    await db.abytes_decrypt(bytes_ciphertext, ttl=1)

except db.TimestampExpired as error:

    assert error.expired_by == 5
HMACs …………………………………… Table Of Contents

Besides encryption & decryption, databases can also be used to manually verify the authenticity of bytes-type data with HMACs.

# Creating an HMAC of some data with a database is done this way ->

data = b"validate this data!"

hmac = await db.amake_hmac(data)

await db.atest_hmac(hmac, data)  # Runs without incident

# Data that is not the same will be caught ->

altered_data = b"valiZate this data!"

await db.atest_hmac(hmac, altered_data)
>>> "InvalidHMAC: Invalid HMAC hash for the given data."

# Any number of bytes-type arguments can be run thorugh the function,

# the collection of items is canonically encoded automagically ->

arbitrary_data = (b"uid_\x0f\x12", b"session_id_\xa1")

hmac = await db.amake_hmac(*arbitrary_data)

await db.atest_hmac(hmac, *arbitrary_data)  # Runs without incident

# Additional qualifying information can be specified with the ``aad``

# keyword argument ->

from time import time

timestamp = int(time()).to_bytes(8, "big")

hmac = await db.amake_hmac(*arbitrary_data, aad=timestamp)

await db.atest_hmac(hmac, *arbitrary_data)
>>> "InvalidHMAC: Invalid HMAC hash for the given data."

await db.atest_hmac(hmac, *arbitrary_data, aad=timestamp) # Runs fine

# This is most helpful for domain separation of the HMAC outputs.

# Each distinct setting & purpose of the HMAC should be specified


uuid = await db.amake_hmac(user_name, aad=b"uuid")

hmac = await db.amake_hmac(user_data, aad=b"data-authentication")


Chunky2048 Cipher ………………………… Table Of Contents

The Chunky2048 cipher is built from generators & SHA3-based key-derivation functions. It’s designed to be easy to use, difficult to misuse & future-proof with large security margins.

High-level Functions …………………….. Table Of Contents

These premade recipes allow for the easiest usage of the cipher.

import aiootp

cipher = aiootp.Chunky2048(key)

# Symmetric encryption of JSON data ->

json_data = {"account": 33817, "names": ["queen b"], "id": None}

encrypted_json_data = cipher.json_encrypt(json_data, aad=b"demo")

decrypted_json_data = cipher.json_decrypt(

    encrypted_json_data, aad=b"demo", ttl=120


assert decrypted_json_data == json_data

# Symmetric encryption of binary data ->

binary_data = b"some plaintext data..."

encrypted_binary_data = cipher.bytes_encrypt(binary_data, aad=b"demo")

decrypted_binary_data = cipher.bytes_decrypt(

    encrypted_binary_data, aad=b"demo", ttl=30


assert decrypted_binary_data == binary_data

# encrypted URL-safe Base64 encoded tokens ->

token_data = b"some plaintext token data..."

encrypted_token_data = cipher.make_token(token_data, aad=b"demo")

decrypted_token_data = cipher.read_token(

    encrypted_token_data, aad=b"demo", ttl=3600


assert decrypted_token_data == token_data

High-level Generators …………………….. Table Of Contents

With these generators, the online nature of the Chunky2048 cipher can be utilized. This means that any arbitrary amount of data can be processed in streams of controllable, buffered chunks. These streaming interfaces automatically handle message padding & depadding, ciphertext validation & detection of out-of-order message blocks.


from aiootp import AsyncCipherStream

# Let's imagine we are serving some data over a network ->

receiver = SomeRemoteConnection(session).connect()

# This will manage encrypting a stream of data ->

stream = await AsyncCipherStream(key, aad=session.transcript)

# We'll have to send the salt & iv in some way ->

receiver.transmit(salt=stream.salt, iv=stream.iv)

# Now we can buffer the plaintext we are going to encrypt ->

for plaintext in receiver.upload.buffer(4 * stream.PACKETSIZE):

    await stream.abuffer(plaintext)

    # The stream will now produce encrypted blocks of ciphertext

    # as well as the block ID which authenticates each block ->

    async for block_id, ciphertext in stream:

        # The receiver needs both the block ID & ciphertext ->

        receiver.send_packet(block_id + ciphertext)

# Once done with buffering-in the plaintext, the ``afinalize``

# method is called so the remaining encrypted data will be

# flushed out of the buffer to the user ->

async for block_id, ciphertext in stream.afinalize():

    receiver.send_packet(block_id + ciphertext)

# Here we can give an optional check of further authenticity,

# also cryptographically asserts the stream is finished ->

receiver.transmit(shmac=await stream.shmac.afinalize())

Decryption / Authentication:

from aiootp import AsyncDecipherStream

# Here let's imagine we'll be downloading some data ->

source = SomeRemoteConnection(session).connect()

# The key, salt, aad & iv must be the same for both parties ->

stream = await AsyncDecipherStream(

    key, salt=source.salt, aad=session.transcript, iv=source.iv


# The downloaded ciphertext will now be buffered & the stream

# object will produce the plaintext ->

for ciphertext in * stream.PACKETSIZE):

    # Here stream.shmac.InvalidBlockID is raised if an invalid or

    # out-of-order block is detected within the last 4 packets ->

    await stream.abuffer(ciphertext)

    # If authentication succeeds, the plaintext is produced ->

    async for plaintext in stream:

        yield plaintext

# After all the ciphertext is downloaded, ``afinalize`` is called

# to finish processing the stream & flush out the plaintext ->

async for plaintext in stream.afinalize():

    yield plaintext

# An optional check for further authenticity which also

# cryptographically asserts the stream is finished ->

await stream.shmac.afinalize()

await stream.shmac.atest_shmac(source.shmac)


Passcrypt ………………………… Table Of Contents

The Passcrypt algorithm is a data independent memory & computationally hard password-based key derivation function. It’s built from a single primitive, the SHAKE-128 extendable output function from the SHA-3 family. Its resource costs are measured by three parameters: mb, which represents an integer number of Mibibytes (MiB); cpu, which is a linear integer measure of computational complexity & the number of iterations of the algorithm over the memory cache; and cores, which is an integer which directly assigns the number of separate processes that will be pooled to complete the algorithm. The number of bytes of the output tag are decided by the integer tag_size parameter. And, the number of bytes of the automatically generated salt are decided by the integer salt_size parameter.

Hashing & Verifying Passphrases …………………….. Table Of Contents

By far, the dominating measure of difficulty for Passcrypt is determined by the mb Mibibyte memory cost. It’s recommended that increases to desired difficulty are first translated into higher mb values, where resource limitations of the machines executing the algorithm permit. If more difficulty is desired than can be obtained by increasing mb, then increases to the cpu parameter should be used. The higher this parameter is the less likely an adversary is to benefit from expending less than the intended memory cost, & increases the execution time & complexity of the algorithm. The final option that should be considered, if still more difficulty is desired, is to lower the cores parallelization parameter, which will just cause each execution to take longer to complete.

from aiootp import Passcrypt, hash_bytes

# The class accepts an optional (but recommended) static "pepper"

# which is applied as additional randomness to all hashes computed

# by the class. It's a secret random bytes value of any size that is

# expected to be stored somewhere inaccessible by the database which

# contains the hashed passphrases ->

with open(SECRET_PEPPER_PATH, "rb") as pepper_file:

    Passcrypt.PEPPER =

# when preparing to hash passphrases, it's a good idea to use any &

# all of the static data / credentials available which are specific

# to the context of the registration ->

APPLICATION = b"my-application-name"

PRODUCT = b"the-product-being-accessed-by-this-registration"


# If the same difficulty settings are going to be used for every

# hash, then a ``Passcrypt`` instance can be initialized to

# automatically pass those static settings ->

pcrypt = Passcrypt(mb=1024, cpu=2, cores=8)  # 1 GiB, 8 cores

# Now that the static credentials / settings are ready to go, we

# can start hashing any user information that arrives ->

username = form["username"].encode()

passphrase = form["passphrase"].encode()

email_address = form["email_address"].encode()

# The ``hash_bytes`` function can then be used to automatically

# encode then hash the multi-input data so as to prevent the chance

# of canonicalization (&/or length extension) attacks ->

aad = hash_bytes(*STATIC_CONTEXT, username, email_address)

hashed_passphrase = pcrypt.hash_passphrase(passphrase, aad=aad)

assert type(hashed_passphrase) is bytes

assert len(hashed_passphrase) == 38

# Later, a hashed passphrase can be used to authenticate a user ->

untrusted_username = form["username"].encode()

untrusted_passphrase = form["passphrase"].encode()

untrusted_email_address = form["email_address"].encode()

aad = hash_bytes(

    *STATIC_CONTEXT, untrusted_username, untrusted_email_address




        hashed_passphrase, untrusted_passphrase, aad=aad, ttl=3600


except pcrypt.InvalidPassphrase as auth_fail:

    # If the passphrase does not hash to the same value as the

    # stored hash, then this exception is raised & can be handled

    # by the application ->


except pcrypt.TimestampExpired as registration_expired:

    # If the timestamp on the stored hash was created more than

    # ``ttl`` seconds before the current time, then this exception

    # is raised. This is helpful for automating registrations which

    # expire after a certain amount of time, which in this case was

    # 1 hour ->



    # If no exception was raised, then the user has been authenticated

    # by their passphrase, username, email address & the context of

    # the registration ->

    app.login_user(username, email_address)


Passcrypt Algorithm Overview …………………….. Table Of Contents

By being secret-independent, Passcrypt is resistant to side-channel attacks. This implementation is also written in pure python. Significant attention was paid to design the algorithm so as to suffer minimally from the performance inefficiencies of python, since doing so would help to equalize the cost of computation between regular users & dedicated attackers with custom hardware / software. Below is a diagram that depicts how an example execution works:

       ___________________ # of rows ___________________
      |                                                 |
      |              initial memory cache               |
      |  row  # of columns == 2 * max([1, cpu // 2])    |
      |   |   # of rows == ⌈1024*1024*mb/168*columns⌉   |
      v   v                                             v
column|---'-----------------------------------------'---| the initial cache
column|---'-----------------------------------------'---| of size ~`mb` is
column|---'-----------------------------------------'---| built very quickly
column|---'-----------------------------------------'---| using SHAKE-128.
column|---'-----------------------------------------'---| each (row, column)
column|---'-----------------------------------------'---| coordinate holds
column|---'-----------------------------------------'---| one element of
column|---'-----------------------------------------'---| 168-bytes.
                       reflection                  row
                      <-   |
      |--------------------'-------'--------------------| each row is
      |--------------------'-------'--------------------| hashed then has
      |--------------------'-------'--------------------| a new 168-byte
      |--------------------'-------'--------------------| digest overwrite
      |--------------------'-------'--------------------| the current pointer
      |--------------------'-------'--------------------| in an alternating
      |--------------------Xxxxxxxx'xxxxxxxxxxxxxxxxxxxx| sequence, first at
      |oooooooooooooooooooo'oooooooO--------------------| the index, then at
                                   |   ->                 its reflection.

      |--'-------------------------------------------'--| this continues
      |--'-------------------------------------------'--| until the entire
      |--'-------------------------------------------Xxx| cache has been
      |ooO-------------------------------------------'--| overwritten.
      |xx'xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx'xx| a single `shake_128`
      |oo'ooooooooooooooooooooooooooooooooooooooooooo'oo| object (H) is used
      |xx'xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx'xx| to do all of the
      |oo'ooooooooooooooooooooooooooooooooooooooooooo'oo| hashing.
         |   ->                                 <-   |
       index                                     reflection

      |xxxxxxxxxxx'xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| finally, the whole
      |ooooooooooo'ooooooooooooooooooooooooooooooooooooo| cache is quickly
      |xxxxxxxxxxx'xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| hashed `cpu` + 2
      |ooooooooooo'ooooooooooooooooooooooooooooooooooooo| number of times.
      |Fxxxxxxxxxx'xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| after each pass an
      |foooooooooo'ooooooooooooooooooooooooooooooooooooo| 84-byte digest is
      |fxxxxxxxxxx'xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| inserted into the
      |foooooooooo'ooooooooooooooooooooooooooooooooooooo| cache, ruling out
                  |   ->                                  hashing state cycles.
                  | hash cpu + 2 # of times               Then a `tag_size`-
                  v                                       byte tag is output.

      tag = H.digest(tag_size)


X25519 & Ed25519 …………………………. Table Of Contents

Asymmetric curve 25519 tools are available from these high-level interfaces over the cryptography package.

X25519 ………………………………….. Table Of Contents

Elliptic curve 25519 diffie-hellman exchange protocols.

from aiootp import X25519, DomainKDF, GUID, Domains

# Basic Elliptic Curve Diffie-Hellman ->

guid = GUID().new()

my_ecdhe_key = X25519().generate()

yield guid, my_ecdhe_key.public_bytes  # send this to Bob

raw_shared_secret =

shared_kdf = DomainKDF(  # Use this to create secret shared keys







# Triple ECDH Key Exchange client initialization ->

with ecdhe_key.dh3_client() as exchange:

    response =


clients_kdf = exchange.result()

# Triple ECDH Key Exchange for a receiving peer ->

identity_key, ephemeral_key = client_public_keys = internet.receive()

server = ecdhe_key.dh3_server(identity_key, ephemeral_key)

with server as exchange:

servers_kdf = exchange.result()

# Success! Now both the client & server peers share an identical

# ``DomainKDF`` hashing object to create shared keys ->

assert (


    == servers_kdf.sha3_512(context=b"test")


Ed25519 …………………………………. Table Of Contents

Edwards curve 25519 signing & verification.

from aiootp import Ed25519

# In a land, long ago ->

alices_key = Ed25519().generate()


# Alice wants to sign a document so that Bob can prove she wrote it.

# So, Alice sends the public key bytes of the key she wants to

# associate with her identity, the document & the signature ->

document = b"DesignDocument.cad"

signed_document = alices_key.sign(document)

message = {
    "document": document,
    "signature": signed_document,
    "public_key": alices_key.public_bytes,


# In a land far away ->

alices_message = internet.receive()

# Bob sees the message from Alice! Bob already knows Alice's public

# key & she has reason believe it is genuinely Alice's. So, she'll

# import Alice's known public key to verify the signed document ->

assert alices_message["public_key"] == alices_public_key

alice_verifier = Ed25519().import_public_key(alices_public_key)

    alices_message["signature"], alices_message["document"]

internet.send(b"Beautiful work, Alice! Thanks ^u^")

The verification didn’t throw an exception! So, Bob knows the file was signed by Alice.

Comprende ……………………………….. Table Of Contents

This magic with generators is made simple with the comprehension decorator. It wraps them in Comprende objects with access to myriad data processing pipeline utilities right out of the box.

Synchronous Generators ……………………. Table Of Contents

from aiootp.gentools import comprehension


def gen(x: int, y: int):

    z = yield x + y

    return x * y * z

# Drive the generator forward with a context manager ->

with gen(x=1, y=2) as example:

    z = 5

    # Calling the object will send ``None`` into the coroutine by default ->

    sum_of_x_y = example()

    assert sum_of_x_y == 3

    # Passing ``z`` will send it into the coroutine, cause it to reach the

    # return statement & exit the context manager ->


# The result returned from the generator is now available ->

product_of_x_y_z = example.result()

assert product_of_x_y_z == 10

# Here's another example ->


def one_byte_numbers():

    for number in range(256):

        yield number

# Chained ``Comprende`` generators are excellent inline data processors ->

base64_data = one_byte_numbers().int_to_bytes(1).to_base64().list()

# This converted each number to bytes then base64 encoded them into a list.

# We can wrap other iterables to add functionality to them ->


def unpack(iterable):

    for item in iterable:

        yield item

# This example just hashes each output then yields them

for digest in unpack(base64_data).sha3_256():


Asynchronous Generators …………………… Table Of Contents

Async Comprende coroutines have almost exactly the same interface as synchronous ones.

from aiootp.asynchs import asleep

from aiootp.gentools import Comprende, comprehension


async def gen(x: int, y: int):

    # Because having a return statement in an async generator is a

    # SyntaxError, the return value is expected to be passed into

    # Comprende.ReturnValue, and then raised to propagate upstream.

    # It's then available from the instance's ``aresult`` method ->

    z = yield x + y

    raise Comprende.ReturnValue(x * y * z)

# Drive the generator forward.

async with gen(x=1, y=2) as example:

    z = 5

    # Awaiting the ``__call__`` method will send ``None`` into the

    # coroutine by default ->

    sum_of_x_y = await example()

    assert sum_of_x_y == 3

    # Passing ``z`` will send it into the coroutine, cause it to reach the

    # raise statement which will exit the context manager gracefully ->

    await example(z)

# The result returned from the generator is now available ->

product_of_x_y_z = await example.aresult()

assert product_of_x_y_z == 10

# Let's see some other ways async generators mirror synchronous ones ->


async def one_byte_numbers():

    # It's probably a good idea to pass control to the event loop at

    # least once or twice, even if async sleeping after each iteration

    # may be excessive when no real work is being demanded by range(256).

    # This consideration is more or less significant depending on the

    # expectations placed on this generator by the calling code.

    await asleep()

    for number in range(256):

        yield number

    await asleep()

# This is asynchronous data processing ->

base64_data = await one_byte_numbers().aint_to_bytes(1).ato_base64().alist()

# This converted each number to bytes then base64 encoded them into a list.

# We can wrap other iterables to add asynchronous functionality to them ->


async def unpack(iterable):

    for item in iterable:

        yield item

# Want only the first twenty results? ->

async for digest in unpack(base64_data).asha3_256()[:20]:

    # Then you can slice the generator.


# Users can slice generators to receive more complex output rules, like:

# Getting every second result starting from the 4th result to the 50th ->

async for result in unpack(base64_data)[3:50:2]:


# Although, negative slice numbers are not supported.

Comprende generators have loads of tooling for users to explore. Play around with it and take a look at the other chainable generator methods in aiootp.Comprende.lazy_generators.

Module Overview ………………………….. Table Of Contents

Here’s a quick overview of this package’s modules:

import aiootp

# Commonly used constants, datasets & functionality across all modules ->


# The basic utilities & abstractions of the package's architecture ->


# A collection of the package's generator utilities ->


# This module is responsible for providing entropy to the package ->


# The high & low level abstractions used to implement the Chunky2048 cipher ->


# The higher-level abstractions used to create / manage key material ->


# Global async / concurrency functionalities & abstractions ->



FAQ …………………………………….. Table Of Contents

Q: What is the one-time pad?

A: It’s a cipher which provides an information theoretic guarantee of confidentiality. It’s typically thought to be too cumbersome a cipher for generalized application because it conveys strict, and well, cumbersome, requirements onto its users. The need for its keys to be at least as large as all the messages it’s ever used to encrypt is one such requirement. Our goal is to design a cipher which immitates the one-time pad through clever algorithms, in such a way as to minimize its inconveniences & still provide some form of information theoretic confidentiality guarantees or, at a minimum, be able to make non-trivial statements about its security against even computationally unbounded adversaries. In this effort, we’ve built what we hope to be a candidate cipher, which we’ve called Chunky2048.

Q: How fast is this ``Chunky2048`` cipher?

A: Well, because it relies on hashlib.shake_128 hashing to build key material streams, it’s rather efficient. It can process about 24 MB/s on a ~1.5 GHz core for both encrypting & decrypting. This is still slow relative to other stream ciphers, but this package is written in pure Python & without hardware optimizations. Using SHA3 ASICs, specific chipset instructions, or a lower-level language implementation, could make this algorithm competitively fast.

Q: What size keys does the ``Chunky2048`` cipher use?

A: It’s been designed to work with any size of key >= 64 bytes.

Q: What’s up with the ``AsyncDatabase`` / ``Database``?

A: The idea is to create an intuitive, pythonic interface to a transparently encrypted and decrypted persistence tool that also cryptographically obscures metadata. It’s designed to persist raw bytes or JSON serializable data, which gives it native support for some of the most important basic python datatypes. It’s still a work in progress, albeit a very nifty one.

Q: Why are the modules transformed into ``Namespace`` objects?

A: We overwrite our modules in this package to have a more fine-grained control over what part of the package’s internal state is exposed to users & applications. The goal is make it more difficult for users to inadvertently jeopardize their security tools, & minimize the attack surface available to adversaries. The Namespace class also makes it easier to coordinate and decide the library’s UI/UX across the package.

Changelog ……………………………….. Table Of Contents

Changes for version 0.22.1

Major Changes

  • The top-level DomainKDF class’ hashing methods can now accept an arbitrary amount of additional data arguments which do not change the internal state of its objects.

  • Switch the order of the internal raw guids with the node_number in the GUID class. This is intended to induce the most variability possible in output guids by interpreting the variable raw guids as more significant bits.

Minor Changes

  • The default cpu cost for Passcrypt was lowered from 2 to 1.

  • Ensured raw guid byte values used by GUID class are interpreted as big-endian integers.

  • The top-level (a)csprng functions now don’t bother to convert a falsey, non-bytes, user-supplied entropy argument to bytes. Instead they just use a value from an internal entropy pool as additional entropy for that invocation of the function.

  • Code clean-ups.

  • Documentation fixes.

  • Added tests for DomainKDF, GUID & SyntheticIV, & improved clarity of some existing tests.

  • Packaging changes to create coherent wheel files.

  • Explicitly declare use of big-endian encoding throughout the package.

  • Conduct a more comprehensive addition of the package’s types to the Typing class.

Changes for version 0.22.0

(Major Rewrite: Backwards Incompatible)

Security Advisory:

  • The top-level (a)csprng functions were found to be unsafe in concurrent code, leading to the possibilty of producing identical outputs from distinct calls if run in quick succession from concurrently running threads & coroutines. The classification of this vulnerability is severe because: 1) users should be able to expect the output of a 64-byte cryptographically secure pseudo-random number generator to always produce unique outputs; and, 2) much of the package utilizes them to produce cryptographic material. This vulnerability does not effect users of the library which are not running it in multiple concurrent threads or coroutines. The vulnerability has been patched & all users are highly encouraged to upgrade to v0.22.0+.

Major Changes

  • Support for python 3.6 was dropped. The package now supports python versions 3.7+.

  • Chunky2048: A new version of the cipher has been developed which implements algorithms & interfaces that offer improvements in multiple regards: smaller size overhead of ciphertexts, faster execution time for large messages & large keys, more robust salt reuse/misue resistance, fewer aspects harming deniability & better domain separation. Many of the changes are described here:

    • The (a)bytes_keys generators were updated to use shake_128-based KDF objects instead of sha3_512, yielding 256-bytes on each iteration instead of 128, now requiring only a single iteration to produce a keystream key for each block, instead of two. This choice was made during the process of analyzing the use of the user’s encryption key to seed the seed_kdf on each iteration. We wanted to stop doing that essentially, because it slowed down the cipher too much when used with large keys. And because it seems like a bad idea to use the same key repeatedly while also not incorporating the uniqueness or entropy from the message’s salt, siv or aad.

      But still, we somehow wanted to come up with an idea which could efficiently & continually extract entropy from the user key if it did happen to be large. An answer came in the form of expanding on an earlier implemented idea which used the key multiple times to create unique seeds during initialization. In this case, however, instead of creating unique seeds with the single seed_kdf, each of the three KDFs & the MAC object used by the cipher will be given the whole key once at initialization, with proper domain separation, & including the message salt & aad (The siv can’t be used because its creation happens after initialization during encryption). This gives each of their (SHA3) 200-byte internal states independent access to the full entropy of the key.

      Then, the problem was that, by using sha3_512 internally, a maximum of 64-bytes of entropy could be communicated between KDFs at each round (and only 32-bytes from the StreamHMAC (shmac) object’s sha3_256 MAC). But the blocksize of each round is 256-bytes. So, the idea became to attempt to communicate more entropy between the KDFs & MAC each round than there exists possible messages in the message space of each round. It seems plausible, that by only assuming the independence of each of the KDFs / MAC & that they can indeed efficiently pass entropy to one another, that for large keys we could argue the relevant key space is that of the 800-byte internal state of the cipher at each round (which happens to be more than three times the size of the message space of each round). This is to say, we conjecture, that by efficiently communicating more entropy from independent sources than there exists possible messages, & in fact incorporating the entropy of each message block into the cipher’s state at the start of each round, that the entropy of the internal keyspace is continually being refreshed in a way which is negligibly distinguishable from using a fresh random key each round the length of the blocksize. This seems like at least a feasible way to begin the argument that it is possible to meaningfully relate the information theoretic security of the one-time pad to a pseudo one-time pad in a measurable way.

      Efficiently Pass Entropy: By this we mean, the rate of bits extracted from one state object, to the rate of bits of actual entropy absorbed by a receiveing state object, up to its XORable state size, being different by only a negligible amount. Here, we can conservatively assume the limit of this efficiency is the XORable state size, since we know that in the ideal setting, XORing n uniform random bits with an unknown message of <= n bits is perfectly hiding, which implies perfectly efficient conveyance of entropy. By using shake_128 as each of the cipher’s state objects, & its larger rate of 168-bytes, more than twice the number of bytes can be passed to & extracted from each, per round & per call to their internal f permutation, as compared with sha3_512. If they can efficiently pass entropy, then any secret state exposed by the left_kdf or right_kdf in the creation of ciphertext, can then be efficiently displaced by the introduction of new entropy from the other state objects. This follows from the theory that a finite sized pool of entropy which is already maximally filled with entropy, cannot incorporate more entropy without fundamentally erasing internal information. From this we arrived at the new design for Chunky2048. In this new design, the shmac feeds 168-bytes to the seed_kdf, the seed_kdf creates 336-bytes to feed 168-bytes each to the left_kdf & right_kdf, the left_kdf & right_kdf each produce 128-byte keys which XOR the 256-byte plaintext, then this ciphertext feeds the shmac & the cycle repeats.

      More work needs to be done to formalize these definitions & analyze their properties. We would be grateful for any help from those with expertise in formal proofs of security in tearing apart this design as we move closer to the first stable release of the package.

    • The SyntheticIV class’ algorithm has been updated as a result of analyzing how we could improve the salt reuse / misuse resistance of the cipher without attesting to plaintext contents in the form of an siv attached to ciphertexts. This plaintext attestation worked counter to our goal of wanting to be able to say something non-trivial about the key-deniability of the cipher. It was noticed that the plaintext padding already incorporated an 8-byte timestamp (now reduced to 4-bytes) & 16-bytes of ephemeral randomness as part of the prepended inner-header, & that these values were not at all used to seed the cipher’s state during decryption. Instead a keyed-hash was calculated over the first block of plaintext during encryption to create the 24-byte siv. But, this is actually less effective at producing salt reuse / misuse resistance than using the timestamp & ephemeral randomness directly in seeding the seed_kdf, because the timestamp is a unique & global counter that does not suffer from collisions. This understanding came while also trying to find a good use for the initial primer_key generated by the keystream generator when sending in the first obligatory None value. In the previous version it was used to initialize the shmac, but now that the shmac would be initialized directly with the user key, it was searching for a use. So the idea was to pair them.

      The new 256-byte primer_key would be XORed with the 256-byte first block of plaintext to mask the inner-header. The unmasked inner-header & 148-bytes of the shmac’s digest will seed the keystream, & the freshly seeded keystream output would be truncated to XOR the part of the masked plaintext which doesn’t include the inner-header. There’s no need now to attach the siv to the ciphertext. Instead, during decryption, the decipher algorithm has access to the inner-header, because it has access to the primer_key & the masked inner-header. The actual plaintext contents of the first block are only accessible after unmasking the inner-header & seeding the keystream. This combination alone of protection from a timestamp & 16-bytes of randomness should give a salt reuse / misuse resistance of at least ~2 ^ 64 messages per second!

      However, even with this new scheme, it would still be problematic to repeat a combination of key, salt & aad, since it would leak the XORs of timestamp information. With all of this in mind, the new formulation would include a 16-byte salt & a newly introduced 16-byte iv, both of which are attached to ciphertexts. This is a header size reduction of 16-bytes, since prior salt & siv sizes were 24-bytes each. The difference between the salt & iv is that the salt is available for the user to choose, but the iv is always generated randomly. Since the iv isn’t dependent on message data the way that the siv was, it too can now be incorporated into all of the state objects during initialization. The iv ensures that even if a key, salt & aad tuple repeats, the timestamp is still protected. Below is a diagram of the procedure:

      |                                     |
      |    Algorithm Diagram: Encryption    |
       ------------------------------------------------------------------     #
      |      inner-header      |        first block of plaintext         |    #
      | timestamp |  siv-key   |                                         |    #
      |  4-bytes  |  16-bytes  |               236-bytes                 |    #
       ------------------------------------------------------------------     #
      |---------------------- entire first block ------------------------|    #
                                       |                                      #
                                       |                                      #
      first 256-byte keystream key ----                                      #
                                       |                                      #
                                       |                                      #
                                       V                                      #
                            masked plaintext block                            #
       ------------------------------------------------------------------     #
      |  masked inner-header   |     first block of masked plaintext     |    #
       ------------------------------------------------------------------     #
                               |----- the 236-byte masked plaintext -----|    #
                                                    |                         #
                                                    |                         #
      siv = inner-header + shmac.digest(148)        |                         #
      keystream(siv)[10:246] -----------------------                         #
                                                    |                         #
                                                    |                         #
                                                    V                         #
       ------------------------------------------------------------------     #
      |  masked inner-header   |       first block of ciphertext         |    #
       ------------------------------------------------------------------     #
      |                                     |
      |    Algorithm Diagram: Decryption    |
       ------------------------------------------------------------------     #
      |  masked inner-header   |        first block of ciphertext        |    #
       ------------------------------------------------------------------     #
      |---------------------- entire first block ------------------------|    #
                                       |                                      #
                                       |                                      #
      first 256-byte keystream key ----                                      #
                                       |                                      #
                                       |                                      #
                                       V                                      #
                          unmasked ciphertext block                           #
       ------------------------------------------------------------------     #
      |      inner-header      |   first block of unmasked ciphertext    |    #
       ------------------------------------------------------------------     #
                               |--- the 236-byte unmasked ciphertext ----|    #
                                                    |                         #
                                                    |                         #
      siv = inner-header + shmac.digest(148)        |                         #
      keystream(siv)[10:246] -----------------------                         #
                                                    |                         #
                                                    |                         #
                                                    V                         #
       ------------------------------------------------------------------     #
      |      inner-header      |         first block of plaintext        |    #
      | timestamp |  siv-key   |                                         |    #
      |  4-bytes  |  16-bytes  |               236-bytes                 |    #
       ------------------------------------------------------------------     #
    • The Padding class has seen some changes. Firstly, the 8-byte timestamp in the inner-header was reduced to 4-bytes. Furthermore, to get the full 136 years out of the 4-byte timestamps, the epoch used to calculate them was changed to unix timestamp 1672531200 (Sun, 01 Jan 2023 00:00:00 UTC). This is the new default 0 date for the package’s timestamps. This saves some space & aims to provided fewer bits of confirmable attestation & correlation in proof games which simulate attacks on the key-deniability of the cipher. To explain: the plaintext padding includes random padding. That padding is intended to leave an adversary which attempts to brute force a ciphertext’s encryption key, even with unbounded computational resources, in a state where it cannot decide with better accuracy than random chance between the exponentially large number of keys which create the same shmac tag (the variable keyspace is much larger than the 32-byte tag) with their accompanying exponentially large number of plausible plaintexts (any reasonable plaintext with any variable length random padding between 16 & 272 bytes), & the actual user key & plaintext.

      We also got rid of the use of a padding_key to indicate the end of a plaintext message. It used to be sliced off the primer_key, but the primer_key has a new use now. Also, the padding_key was another form of plaintext / key attestation harming deniability that we wanted to get rid of. Instead, a simpler method is now employed: The final byte of the final block of padded plaintext is a number which tells the decryptor exactly how many bytes of random padding were added to the plaintext to fill the block. This saves a lot of space, is simpler, minimizes unnecessary key attestation, & eliminates the need for the Padding class to know anything about user secrets in order to do the padding, which is an improvment all around.

  • New (Async)CipherStream & (Async)DecipherStream classes were introduced which allow users to utilize the online nature of the Chunky2048 cipher, ciphering & deciphering data in bufferable chunks, without needing to know about or instantiate all of the low-level classes. They automatically handle the required plaintext padding, ciphertext authentication, & detection of out-of-order message blocks. This greatly simplifies the safe usage of Chunky2048 in online mode, provides robustness, & gets rid of the need for users to worry about the dangers of release of unverified plaintexts.

  • The Passcrypt algorithm was redesigned to be data-independent, more efficiently acheive its security goals, & allow for more compact hashes which include its difficulty settings metadata. The kb parameter was changed to mb, & now measures Mibibytes (MiB). A new cores parallelization parameter was added, which indicates the number of parallel processes to use to complete the procedure. And the cpu parameter now measures the number of iterations over the memory cache that are done, as well as the computational complexity of the algorithm. Passcrypt now uses shake_128 instead of sha3_512 internally. This also allows for users to specify a tag_size number of bytes to produce as an output tag. A salt_size parameter can now also be supplied to the (a)hash_passphrase methods. The (a)hash_passphrase methods now produce raw-bytes outputs & the (a)hash_passphrase_raw & (a)verify_raw methods were removed. (a)verify methods now also accept range-type objects as mb_allowed, cpu_allowed, & cores_allowed keyword argument inputs. These range objects can be used to specify the exact amount of resources which the user allows for difficulty settings, which can mitigate adversarial (or unintentional) DOS attacks on machines doing hash verification.

  • Type annotations were added to most of the library, including return types, which were completely neglected in prior versions. They are still not functioning with mypy, & are serving right now as documentation & auto-complete helpers.

  • Many unnecesssary, low-level or badly designed features, functions & classes were either deleted or pulled into private namespaces, along with major reorganization & cleanup of the codebase. The tangled mess of internal module imports was also cleaned up. The goal is to provide access to only the highest level, simplest, & safest by default interfaces which can actually help users in their data processing & cryptographic tasks. These changes aim to improve maintainability, readability, correctness & safety.

  • New top-level (a)hash_bytes functions were added to the package, which accept an unlimited number bytes-type inputs as positional arguments & automatically canonically encode all inputs before being hashed (which aims to prevent canonicalization attacks & length-extension attacks). A key keyword-only argument can also be supplied to optionally produce keyed hashes.

  • A new top-level GUID class was added. It creates objects which produce variable length, obfuscated, pseudo-random bytes-type globally unique identifiers based on a user-defined integer node_number, a user-defined uniform bytes salt, a nanosecond timestamp, random entropy bytes & a 1-byte counter. The benefits of its novel design explained: 1) the namespace separation of user-defined salts (like name-based uuids); 2) guaranteed output uniqueness for all instances using the same salt & node_number which occur on a different nanosecond (like time-based uuids, but with higher precision); 3) guaranteed output uniqueness between all instances which use the same salt but a different node_number, even if produced on the same nanosecond; 4) guaranteed output uniqueness for any unique instance using the same salt & node_number if it produces 256 or fewer outputs every nanosecond; 5) probabilistic output uniqueness for any unique instance using the same salt & node_number if it produces >256 outputs per-nanosecond, exponentially proportional to the number of random entropy bytes (which in turn are proportional to the output size of the GUIDs); 6) output invertability, meaning outputs can be unmasked & sorted according to timestamp, node_number & counter; 7) random-appearing outputs, with the marginal amount of privacy which can be afforded by obfuscated affine-group operations. Admittedly, point 7) still leaves much room for improvement, as the privacy of the design could instead be ensured by strong hardness assumptions given by other types of invertible permutations or group operations. The goal was to create something efficient (below 3µs per guid), which met the above criterion, & that produced output bit sequences which passed basic randomness tests. We’d be excited to accept pull requests which use strong invertable permutations or group operations that are also about as efficient, & that for n-byte declared output sizes, outputs do not repeat for fewer than ~256 ** n sequential input values.

  • The top-level DomainKDF class now also creates KDF objects which automatically canonically encode all inputs.

  • The X25519 protocols now return DomainKDF results instead of plain sha3_512 objects.

  • The (Base)Comprende classes were greatly simplified, & the caching & messages features were removed.

  • The top-level (a)mnemonic functions now return lists of bytes-type words, instead of str-type, & can now be used to quickly generate lists of randomly selected words without providing a (now optional) passphrase.

  • The (Async)Database classes’ (a)generate_profile methods no longer require tokens to first be created by the user. That is now handled internally, & the external API accepts raw bytes inputs for credentials from the user.

  • The PackageSigner & PackageVerifier now use sha384 for digests instead of sha512. The verifier now by default recomputes & verifies the digests of files from the filesystem using the path keyword argument to the constructor as the root directory for the relative filepaths declared in the “checksums” entry of the signature summary.

Minor Changes

  • A new Clock class was added to the module which provides a very intuitive API for handling time & timestamp functionalities for various time units.

  • The test suite was reorganized, cleaned up & extended significantly, & now also utilizes pytest-asyncio to run async tests. This led to many found & fixed bugs in code that was not being tested. There’s still a substantial amount of tests that need to be written. We would greatly appreciate contributions which extend our test coverage.

  • Many improvements to the correctness, completeness & aesthetic beauty of the code documentation with the addition of visual aides, diagrams & usage examples.

  • A top-level report_security_issue function was added, which provides a terminal application for users to automatically encrypt security reports to us using our new X25519 public key.

  • We lost access to our signing keys in encrypted drives which were damaged in flooding. So we decided to shred them & start fresh. Our new Ed25519 signing key is “70d1740f2a439da98243c43a4d7ef1cf993b87a75f3bb0851ae79de675af5b3b”. Contact us via email or twitter if you’d like to confirm that the key you are seeing is really ours.

Changes for version 0.21.1

Minor Changes

  • Fix usage of the wrong package signing key.

Changes for version 0.21.0

Major Changes

  • Non-backwards compatible changes:

  • Altered the Chunky2048 cipher’s key derivation to continuously extract entropy from users’ main encryption key. The design goal of the cipher is to be as close as possible to a one-time pad, but because we use key derivations to mix together all the relevant values used by the cipher, there’s a limited amount of entropy that can be extracted from the main key no matter how large it is. The changes feed the main key into the internal seed KDF multiple times when creating the cipher’s initial seeds, & once on every iteration of the (a)bytes_keys generators.

  • Merged two internal KDFs used by the cipher into the one seed KDF. This also now means that using the (a)update_key methods of the StreamHMAC class updates the KDF used to ratchet the encryption keystream.

  • Use sha3_512 instead of sha3_256 for the StreamHMAC final HMAC & slice the first bytes designated by the package’s module. This allows the HMAC length to be specified & changed easily. It’s highly discouraged to use anything less than 32-bytes.

Minor Changes

  • Internal refactorings.

  • Updates to tests.

Changes for version 0.20.7

Major Changes

  • Changed the way the Padding.(a)end_padding methods calculate the required padding length. The change causes the methods to now assume that the plaintext has already been prepended with the start padding.

  • The various test_* & verify_* functions/methods throughout the package have been changed to return None on successful validation instead of True, which more closely matches the convention for exception-raising validators.

  • The default block_id length was changed from 16-bytes to 24-bytes.

Minor Changes

  • Make the (a)end_padding methods of the Padding class assume the supplied data has already been prepended with the start padding. This better integrates with streams of plaintext (online usage).

  • Small internal refactorings.

  • Documentation fixes.

Changes for version 0.20.6

Major Changes

  • The (Async)Database classes now support storing raw bytes type tag entries! This is a huge boon to time/space efficiency when needing to store large binary files, since they don’t need to be converted to & from base64. This feature was made possible with only very minor changes to the classes, & they’re fully backwards-compatible! Older versions will not be able handle raw bytes entries, but old JSON serializable entries work the same way they did.

Minor Changes

  • Docfixes.

  • Small refactorings.

  • Add new tests & make existing tests complete faster.

  • Support empty strings to be passed to the (Async)Database constructors’ directory kwarg, signifying the current directory. Now None is the only falsey value which triggers the constructors to use the default database directory.

  • Fixed a bug in the AsyncDatabase class’ aset_tag method, which would throw an attribute error when passed the cache=False flag.

  • Add Windows support to the CI tests.

Changes for version 0.20.5

Minor Changes

  • Include the missing changelog entries for v0.20.4.

Changes for version 0.20.4

Major Changes

  • Add python3.10 support by copying the async_lru package’s main module from their more up-to-date github repository instead of from PyPI.

Minor Changes

  • Small refactorings & code cleanups.

  • Documentation updates.

  • Type-hinting updates.

  • Cleanups to the package’s module API.

  • Improve CI & extend to python3.10.

Changes for version 0.20.3

Minor Changes

  • Small refactorings.

  • Documentation updates.

  • Type-hinting updates.

  • Additional tests.

Changes for version 0.20.2

Major Changes

  • Changed the Padding class’ (a)check_timestamp methods to (a)test_timestamp, to better match the naming convention in the rest of the package.

  • Removed the (a)sum_sha3__(256/512) chainable generator methods from the Comprende class.

  • Removed the os.urandom based functions in the module.

Minor Changes

  • Fixes & improvements to out of date documentation.

  • Small fixes to type-hints.

  • Small refactorings.

  • Add (a)generate_key functions to the package & (Async)Keys classes.

  • Fix some exception messages.

Changes for version 0.20.1

Minor Changes

  • Small fixes & improvements to documentation.

  • Small fixes & improvements to tests.

  • Small fixes to type-hints.

  • Small re-organization of source file contents.

  • Small bug fixes.

Changes for version 0.20.0 (Backwards incompatible updates)

Major Changes

  • The (a)json_(en/de)crypt & (a)bytes_(en/de)crypt functions & methods now only expect to work with bytes type ciphertext. And, the low-level cipher generators expect iterables of bytes where they used to expect iterables of integers.

  • The pid keyword-only argument throughout the package was changed to aad to more clearly communicate its purpose as authenticated additional data.

  • The key, salt & aad values throughout the package are now expected to be bytes type values.

  • The key must now be at least 32-bytes for use within the Chunky2048 cipher & its interfaces.

  • The salt, for use in the Chunky2048 cipher & its interfaces, was decreased from needing to be 32-bytes to 24-bytes.

  • The siv, for use in the Chunky2048 cipher & its interfaces, was increased from needing to be 16-bytes to 24-bytes.

  • The new KeyAADBundle class was created as the primary interface for consuming key, salt, aad & siv values. This class’ objects are the only ones that are used to pass around these values in low-level Chunky2048 cipher functionalities. The higher-level cipher functions are the only public interfaces that still receive these key, salt, & aad values.

  • The KeyAADBundle now manages the new initial key derivation of the Chunky2048 cipher. This new algorithm is much more efficient, utilizing the output of the keystream’s first priming call instead of throwing it away, removing the need for several other previously used hashing calls.

  • The bytes_keys & abytes_keys keystream generator algorithms were improved & made more efficient. They also now only receive bytes type coroutine values or None.

  • The StreamHMAC algorithms were improved & made more efficient.

  • The Chunky2048 class now creates instance’s that initialize, & who’s methods are callable, much more efficiently by reducing its previously dynamic structure. Its now reasonable to use these instances in code that has strict performance requirements.

  • The Keys & AsyncKeys classes were trimmed of all instance behaviour. They are now strictly namespaces which contain static or class methods.

  • All instance’s of the word password throughout the package have been replaced with the word passphrase. The Passcrypt class now only accepts bytes type passphrase & salt values. The returned hashes are also now always bytes.

  • The Padding & BytesIO classes’ functionalities were made more efficient & cleaned up their implementations.

  • New PackageSigner & PackageVerifier classes were added to the module to provide an intuituve API for users to sign their own packages. This package now also uses these classes to sign itself.

  • The new module was created to organize the generator utilities that were previously scattered throughout the package’s top-level namespaces.

  • The new module was created to help organize the exceptions raised throughout the package, improving readability & maintainability.

  • The new module was added to assist in the long process of adding functional type-hinting throughout the package. For now, the type hints that have been added primarily function as documentation.

  • A new Slots base class was added to the module to simplify the creation of more memory efficient & performant container classes. The new module was made for such classes for use throughout the package. And, most classes throughout the package were given __slots__ attributes.

  • A new OpenNamespace class was added, which is a subclass of Namespace, with the only difference being that instances do not omit attributes from their repr’s.

  • The new (a)bytes_are_equal functions, which are pointers to hmac.compare_digest from the standard library, have replaced the (a)time_safe_equality functions.

  • The (a)sha_256(_hmac) & (a)sha_512(_hmac) functions have had their names changed to (a)sha3__256(_hmac) & (a)sha3__512(_hmac). This was done to communicate that they are actually SHA3 functions, but the double underscore is to keep them differentiable from the standard library’s hashlib objects. They can now also return bytes instead of hex strings if their hex keyword argument is truthy.

  • The base functionality of the Comprende class was refactored out into a BaseComprende class. The chainable data processor generator methods remain in the Comprende class. Their endpoint methods (such as (a)list & (a)join) have also been changed so they don’t cache results by default.

  • The Passcrypt class’ kb & hardness can now be set to values independently from one another. The algorithm runs on the new (a)bytes_keys coroutines, & a slightly more effective cache building procedure.

  • The databases classes now don’t preload their values by default. And, various methods which work with tags & metatags have been given a cache keyword-only argument to toggle on/off the control of using the cache for each operation.

  • New method additions/changes to the database classes:

    • (a)rollback_tag, (a)clear_cache, & a filenames property were added.

    • (a)hmac was changed to (a)make_hmac, & now returns bytes hashes.

    • (a)save was changed to (a)save_database.

    • (a)query was changed to (a)query_tag.

    • (a)set was changed to (a)set_tag.

    • (a)pop was changed to (a)pop_tag.

    • The tags, metatags & filenames properties now return sets instead of lists.

  • The Ropake class has been removed from the package pending changes to the protocol & its implementation.

  • The (a)generate_salt function now returns bytes type values, & takes a size keyword-only argument, with no default, that determines the number of bytes returned between [8, 64].

  • The (a)random_512 & (a)random_256 public functions can now cause their underlying random number generators to fill their entropy pools when either the rounds or refresh keyword arguments are specified.

  • The following variables were removed from the package:

    • (a)keys, (a)passcrypt, (a)seeder, (a)time_safe_equality, Datastream, bits, (a)seedrange, (a)build_tree, (a)customize_parameters, convert_class_method_to_member, convert_static_method_to_member, (a)xor, (a)padding_key, (a)prime_table, (a)unique_range_gen, (a)non_0_digits, (a)bytes_digits, (a)digits, (a)permute, (a)shuffle, (a)unshuffle, (a)create_namespace, ((a)depad_plaintext, (a)pad_plaintext & their generator forms. Only the non-generator forms remain in the Padding class), (The (a)passcrypt, (a)uuids, (a)into_namespace methods from the database classes), (The (a)csprbg functions were removed & instead the (a)csprng functions produce bytes type values.)

  • Thorough & deep refactorings of modules, classes & methods. Many methods & functions were made private, cleaning up the APIs of the package, focusing on bringing the highest-level functionalities to top level namespaces accessible to users. Some purely private functionalities were entirely moved to private namespaces not readily accessible to users.

  • Most of the constants which determine the functionalities throughout the package were refactored out into This allows for easy changes to protocols & data formats.

Minor Changes

  • Many documentation improvements, fixes, trimmings & updates.

  • Added a WeakEntropy class to the module.

Changes for version 0.19.4

Major Changes

  • Created a private EntropyDaemon class to run a thread in the background which feeds into & extracts entropy from some of the package’s entropy pools. Also moved the separate private _cache entropy pools from the parameters to the random number generators. They’re now a single private _pool shared global that’s asynchronously & continuously updated by the background daemon thread.

  • Switched the random portion of function names in the module to read unique instead. This was done to the functions which are actually pseudo-random. This should give users a better idea of which functions do what. The exception is that the random_sleep & arandom_sleep functions have kept their names even though they sleep a pseudo-randomly variable amount of time. Their names may cause more confusion if they were either (a)unique_sleep or (a)urandom_sleep. Because they don’t use os.urandom & what is a unique_sleep? When / if a better name is found these function names will be updated as well.

Minor Changes

  • Various docstring / documentation fixes & refactorings.

Changes for version 0.19.3

Major Changes

  • Removed ascii_encipher, ascii_decipher, aascii_encipher & aascii_decipher generators from the Chunky2048 & Comprende classes, & the package. It was unnecessary, didn’t fit well with the intended use of the Padding class, & users would be much better served by converting their ascii to bytes to use the bytes_ generators instead.

  • Removed the map_encipher, map_decipher, amap_encipher & amap_decipher generators from the Chunky2048 & Comprende classes, & the package. They were not being used internally to the package anymore, & their functionality, security & efficiency could not be guaranteed to track well with the changes in the rest of the library.

  • Added domain specificity to the X25519 protocols’ key derivations.

  • Renamed the database classes’ (a)encrypt & (a)decrypt methods to (a)json_encrypt & (a)json_decrypt for clarity & consistency with the rest of the package. Their signatures, as well as those in (a)bytes_encrypt & (a)bytes_decrypt, were also altered to receive plaintext & ciphertext as their only positional arguments. The filename argument is now a keyword-only argument with a default None value. This allows databases to be used more succinctly for manual encryption & decryption by making the filename tweak optional.

  • The runs keyword argument for the functions in was renamed to rounds. It seems more clear that it is controlling the number of rounds are internally run within the (a)random_number_generator functions when deriving new entropy.

Minor Changes

  • Fixes to docstrings & tutorials. Rewrite & reorganization of the PREADME.rst & README.rst. More updates to the readme’s are still on the way.

  • Slight fix to the Passcrypt docstring’s algorithm diagram.

  • Moved the default passcrypt settings to variables in the Passcrypt class.

  • Added the ability to send passcrypt settings into the mnemonic & amnemonic coroutines, which call the algorithm internally but previously could only use the default settings.

  • Some code cleanups & refactorings.

Changes for version 0.19.2

Minor Changes

  • Made the output lengths of the Padding class’ generator functions uniform. When the footer padding on a stream of plaintext needs to exceed the 256-byte blocksize (i.e. when the last unpadded plaintext block’s length L is 232 < L < 256), then another full block of padding is produced. The generators now yield 256-byte blocks consistently (except during depadding when the last block of plaintext may be smaller than the blocksize), instead of sometimes producing a final padded block which is 512 bytes.

Changes for version 0.19.1

Minor Changes

  • Fixed a bug where database classes were evaluating as falsey when they didn’t have any tags saved in them. They should be considered truthy if they’re instantiated & ready to store data, even if they’re currently empty & not saved to disk. This was reflected in their __bool__ methods. The bug caused empty metatags not to be loaded when an instance loads, even when preload is toggled True.

  • Removed the coroutine-receiving logic from the Padding class’ Comprende generators. Since they buffer data, the received values aren’t ever going to coincide with the correct iteration & will be susceptible to bugs

  • Fixed a bug in the Padding class’ Comprende generators which cut iteration short because not enough data was available from the underlying generators upfront. Now, if used correctly to pad/depad chunks of plaintext 256 bytes at a time, then they work as expected.

  • The update, aupdate, update_key & aupdate_key methods in both the StreamHMAC & DomainKDF classes now return self to allow inline updates.

  • Added acsprng & csprng function pointers to the Chunky2048 class.

  • Updates to docstrings which didn’t get updated with info on the new synthetic IV feature.

  • Some other docstring fixes.

  • Some small code cleanups & refactorings.

Changes for version 0.19.0

Major Changes

  • Security Upgrade: The package’s cipher was changed to an online, authenticated scheme with salt reuse / misuse resistance. This was acheived through a few backwards incompatible techniques:

    1. A synthetic IV (SIV) is calculated from the keyed-hash of the first 256-byte block of plaintext. The SIV is then used to seed the keystream generator, & is used to update the validator object. This ensures that if the first block is unique, then the whole ciphertext will be unique.

    2. A 16-byte ephemeral & random SIV-key is also prepended to the first block of plaintext during message padding. Since this value is also hashed to derive the SIV, this key gives a strong guarantee that a given message will produce a globally unique ciphertext.

    3. An 8-byte timestamp is prepended to the first block of plaintext during padding. Timestamps are inherently sequential, they can be verified by a user within some bounds, & can also be used to mitigate replay attacks. Since it’s hashed to make the SIV, then it helps make the entire ciphertext unique.

    4. After being updated with each block of ciphertext, the validator’s current state is again fed into the keystream generator as a new rotating seed. This mitigation is limited to ensuring only that every following block of ciphertext to a block which is unique will also be unique. More specifically this means that: if all other mitigations fail to be unique, or are missing, then the first block which is unique will appear the same, except for the bits which have changed, but, all following blocks will be randomized. This limitation could be avoided with a linear expansion in the ciphertext size by generating an SIV for each block of plaintext. This linear expansion is prohibitive as a default setting, but the block level secrecy, even when all other mitigations fail, is enticing. This option may be added in the future as a type of padding mode on the plaintext.

    The SIV-key is by far the most important mitigation, as it isn’t feasibly forgeable by an adversary, & therefore also protects against attacks using encryption oracles. These changes can be found in the SyntheticIV class, the (en/de)cipher & xor generators, & the StreamHMAC class in the module. The padding changes can also be found in the new Padding class in the module. The SIV is attached in the clear with ciphertexts & was designed to function with minimal user interaction. It needs only to be passed into the StreamHMAC class during decryption – during encryption it’s automatically generated & stored in the StreamHMAC validator object’s siv property attribute.

  • Security Patch: The internal sha3_512 kdf’s to the akeys, keys, abytes_keys & bytes_keys keystream generators are now updated with 72 bytes of (64 key material + 8 padding), instead of just 64 bytes of key material. 72 bytes is the bitrate of the sha3_512 object. This change causes the internal state of the object to be permuted for each iteration update & before releasing a chunk of key material. Frequency analysis of ciphertext bytes didn’t smooth out to the cumulative distribution expected for all large ciphertexts prior to this change. But after the change the distribution does normalize as expected. This indicates that the key material streams were biased away from random in a small but measurable way. Although, no particular byte values seem to have been preferred by this bias, this is a huge shortcoming with unknown potential impact on the strength of the package’s cipher. This update is strongly recommended & is backwards incompatible.

  • This update gives a name to the package’s pseudo-one-time-pad cipher implementation. It’s now called Chunky2048! The OneTimePad class’ name was updated to Chunky2048 to match the change.

  • The PreemptiveHMACValidation class & its related logic in the StreamHMAC class was removed. The chaining of validator output into the keystream makes running the validator over the ciphertext separately or prior to the decryption process very costly. It would either mean recalculating the full hash of the ciphertext a second time to reproduce the correct outputs during each block, or a large linear memory increase to hold all of its digests to be fed in some time after preemtive validation. It’s much simpler to remove that functionality & potentially replace it with something else that fits the user’s applications better. For instance, the current_digest & acurrent_digest methods can produce secure, 32-byte authentication tags at any arbitrary blocks throughout the cipher’s runtime, which validate the cipehrtext up to that point. Or, the next_block_id & anext_block_id methods, which are a more robust option because each id they produce validates the next ciphertext block before updating the internal state of the validator. This acts as an automatic message ordering algorithm, & leaves the deciphering party’s state unharmed by dropped packets or manipulated ciphertext.

  • The update_key & aupdate_key methods were also added to the StreamHMAC class. They allow the user to update the validators’ internal key with new entropy or context information during its runtime.

  • The Comprende class now takes a chained keyword-only argument which flags an instance as a chained generator. This flag allows instances to communicate up & down their generator chain using the shared Namespace object accessible by their messages attribute.

  • The chainable Comprende generator functions had their internals altered to allow them to receive, & pass down their chain, values sent from a user using the standard coroutine send & asend method syntax.

  • Comprende instances no longer automatically reset themselves every time they enter their context managers or when they are iterated over. This makes their interface more closely immitate the behavior of async/sync generator objects. To get them to reset, the areset or reset methods must be used. The message chaining introduced in this update allows chains of Comprende async/sync generators to inform each other when the user instructs one of them to reset.

  • The standard library’s hmac module is now used internally to the module’s sha_512_hmac, sha_256_hmac, asha_512_hmac & asha_256_hmac functions. They still allow any type of data to be hashed, but also now default to hashing bytes type objects as they are given.

  • The new Domains class, found in, is now used to encode constants into deterministic pseudo-random 8-byte values for helping turn hash function outputs into domain-specific hashes. Its use was included throughout the library. This method has an added benefit with respect to this package’s usage of SHA-3. That being, the bitrate for both sha3_512 & sha3_256 are (2 * 32 * k) + 8 bytes, where k = 1 for sha3_512 & k = 2 for sha3_256. This means that prepending an 8-byte domain string to their inputs also makes it more efficient to add some multiple of key material to make the input data precisely equal the bitrate. More info on domain-specific hashing can be found here.

  • A new DomainsKDF class in was added to create a more standard & secure method of key derivation to the library which also incorporates domain separation. Its use was integrated thoughout the AsyncDatabase & Database classes to mitigate any further vulnerabilities of their internal key-derivation functions. The database classes now also use bytes-type keys internally, instead of hex strings.

  • The Passcrypt class now contains methods which create & validate passcrypt hashes which have their settings & salt attached to them. Instances can now also be created with persistent settings that are automatically sent into instance methods.

Minor Changes

  • Many fixes of docstrings, typos & tutorials.

  • Many refactorings: name changes, extracted classes / functions, reorderings & moves.

  • Various code clean-ups, efficiency & usability improvements.

  • Many constants used throughout the library were given names defined in the module.

  • Removed extraneous functions throughout the library.

  • The asymmetric key generation & exchange functions/protocols were moved from the module to

  • Add missing modules to the MANIFEST.rst file.

  • Added a UniformPrimes class to the __datasets module for efficient access to primes that aren’t either mostly 1 or 0 bits, as is the case for the primes helper table. These primes are now used in the Hasher class’ amask_byte_order & mask_byte_order methods.

  • The time_safe_equality & atime_safe_equality methods are now standalone functions available from the module.

  • Added reset_pool to the Processes & Threads classes. Also fixed a missing piece of logic in their submit method.

  • Added various conversion values & timing functions to the module.

  • The make_uuid & amake_uuid coroutines had their names changed to make_uuids & amake_uuids.

  • Created a new Datastream class in to handle buffering & resizing iterable streams of data. It enables simplifying logic that must happen some number of iterations before the end of a stream. It’s utilized in the Padding class’ generator functions available as chainable Comprende methods.

  • The data & adata generators can now produce a precise number of size-length blocks as specified by a user. This gets rid of the confusing usage of the old stop keyword-only argument, which stopped a stream after approximately size number of elements.

  • Improved the efficiency & safety of entropy production in the module.

Changes for version 0.18.1

Major Changes

  • Security Patch: Deprecated & replaced an internal kdf for saving database tags due to a vulnerability. If an adversary can get a user to reveal the value returned by the hmac method when fed the tag file’s filename & the salt used for that encrypted tag, then they could deduce the decryption key for the tag. A version check was added only for backwards compatibility & will be removed on the next update. All databases should continue functioning as normal, though all users are advised to re-save their databases after upgrading so the new kdf can be used. This will not overwrite the old files, so they’ll need to be deleted manually.

  • Replaced usage of the async switch coroutine with asyncio.sleep because it was not allowing tasks to switch as it was designed to. Many improvements were made related to this change to make the package behave better in async contexts.

  • Removed the private method in the database classes which held a reference to the root salt. It’s now held in a private attribute. This change simplifies the code a bit & allows instances to be pickleable.

  • The atimeout & timeout chainable Comprende generator methods can now stop the generators’ executions mid-iteration. They run them in separate async tasks or thread pools, respectively, to acheive this.

  • The await_on & wait_on generators now restart their timeout counters after every successful iteration that detected a new value in their queue. The delay keyword argument was changed to probe_frequency, a keyword-only argument.

  • Removed the package’s dependency on the aioitertools package.

  • Made the sympy package an optional import. If any of its functionalities are used by the user, the package is only then imported & this is done automatically.

  • Various streamlining efforts were made to the imports & entropy initialization to reduce the package’s import & startup time.

Minor Changes

  • Fixes of various typos, docstrings & tutorials.

  • Various cleanups, refactorings & efficiency improvements.

  • Added new tests for detecting malformed or modified ciphertexts.

  • Removed extraneous functions in

  • Add a UNIFORM_PRIME_512 value to for use in the Hasher.mask_byte_order & Hasher.amask_byte_order methods. Those methods were also altered to produce more uniform looking results. The returned masked values are now also 64 bytes by default.

  • Added an automate_key_use keyword-only boolean argument to the init for the OneTimePad, Keys & AsyncKeys classes. It can be toggled to stop the classes from overwriting class methods so they automatically read the instance’s key attribute. This optionally speeds up instantiation by an order of magnitude at the cost of convenience.

  • Fixed asynchs.Threads class’ wrongful use of a multiprocessing Manager.list object instead of a regular list.

  • Changed the _delay keyword-only argument in Processes & Threads classes’ methods to probe_freqeuncy so users can specify how often results will be checked for after firing off a process, thread, or associated pool submission.

  • Now the asubmit & submit methods in Processes & Threads can accept keyword arguments.

  • Added agather & gather methods to the Threads & Processes classes. They receive any number of functions, & args &/or kwargs to pass to those functions when submitting them to their associated pools.

  • Changed the runsum instance IDs from hex strings to bytes & cleaned up the instance caching & cleaning logic.

  • Altered & made private the asalted_multiply & salted_multiply functions in the module.

  • Started a new event loop specific to the module which should prevent the RuntimeError when random_number_generator is called from within the user’s running event loop.

  • Added a ValueError check to the (a)cspr(b/n)g functions in This will allow simultaneously running tasks to request entropy from the function by returning a result from a newly instantiated generator object.

  • Added checks in the *_encipher & *_decipher generators to help assure users correctly declare the mode for their StreamHMAC validator instances.

  • Fixed the __len__ function in the database classes to count the number of tags in the database & exclude their internal maintenaince files.

  • The TimeoutError raised after decrypting a ciphertext with an expired timestamp now contains the seconds it has exceeded the ttl in a value attribute.

  • The timestamp used to sign the package now displays the day of signing instead of the second of signing.

  • The (a)sum_sha_* & (a)sum_passcrypt generators were altered to reapply the supplied salt on every iteration.

  • Stabilized the usability of the stop keyword-only argument in the adata & data generators. It now directly decides the total number of elements in a sequence allowed to be yielded.

Changes for version 0.18.0

Major Changes

  • Security Patch: Rewrote the HMAC-like creation & authentication process for all of the package’s ciphers. Now, the *_encipher & *_decipher Comprende generators must be passed a validator object to hash the ciphertext as it’s being created / decrypted. The StreamHMAC class was created for this purpose. It’s initalized with the user’s long-term key, the ephemeral salt & the pid value. The pid value can now effectively be used to validate additional data. These changes force the package’s cipher to be used as an AEAD cipher.

  • Security Patch: The package’s *_hmac hash functions & the Comprende class’ hash generators were rewritten to prepend salts & keys to data prior to hashing instead of appending. This is better for several important reasons, such as: reducing the amortizability of costs in trying to brute-force hashes, & more closely following the reasoning behind the HMAC spec even though sha3 has a different security profile.

  • Algorithm Patch: The akeys, keys, abytes_keys, & bytes_keys algorithms have been patched to differentiate each iteration’s two sha3_512 hashes from one another in perpetuity. They contained a design flaw which would, if both sha3_512 objects landed upon the same 1600-bit internal state, then they would produce the same keystreams from then on. This change in backwards incompatible. This flaw is infeasible to exploit in practice, but since the package’s hashes & ciphertext validations were already channging this release, there was no reason to not fix this flaw so that it’s self-healing if they ever do land on the same internal states.

  • The Passcrypt class & its algorithm were made more efficient to better equalize the cost for users & adversaries & simplifies the algorithm. Any inefficiencies in an implementation would likely cause the adversary to be able to construct optimized implementations to put users at an even greater disadvantage at protecting their inputs to the passcrypt algorithm. It used the sum_sha_256 hash function internally, & since it was also changing in a non-backwards compatible way with this update, it was the best time to clean up the implementation.

  • Updated the package’s description & its docstrings that refer to the package’s cipher as an implementation of the one-time-pad. It’s not accurate since the package uses pseudo-random hash functions to produce key material. Instead, the package’s goal is to create a pseudo-one-time-pad that’s indistinguishable from a one-time-pad. The OneTimePad class will keep its name for succinctness.

  • New amake_token, make_token, aread_token & read_token class & instance methods added to the OneTimePad class. These tokens are urlsafe base64 encoded, are encrypted, authenticated & contain timestamps that can enforce a time-to-live for each token.

  • Non-backwards compatible changes to the database classes’ filenames, encryption keys & HMACs. The *_hmac hash functions that the databases rely on were changing with this update, so additionally the filenames table used to encode the filenames was switched from the BASE_36_TABLE to the BASE_38_TABLE. Both tables are safe for uri’s across all platforms, but the new table can encode information slightly more efficiently.

  • Major refactorings & signature changes across the package to make passing keys & salts to *_hmac functions & the Comprende class’ hash generators explicit.

  • Removed the of keyword argument from all of the Comprende class’ generators. It was overly complicating the code, & was not entirely clear or useful for settings outside of the tags & atags generators.

  • Removed pybase64 from the package & its dependencies list. The built-in python base64 module works just fine.

  • Sorted the WORDS_LIST, ASCII_ALPHANUMERIC, & BASE_64_TABLE datasets.

  • The salt & asalt functions have been renamed to generate_salt & agenerate_salt for clarity’s sake, & to reduce naming collisions.

  • Added another redundancy to the arandom_number_generator & random_number_generator functions. Now the async tasks it prepares into a list are pseudo-randomly shuffled before being passed into asyncio.gather.

Minor Changes

  • Added a logo image to the package.

  • Separated the FAQ section from PREADME.rst.

  • The primes & bits datasets are now represented in hex in the source code.

  • Added a BASE_38_TABLE dataset to the package.

  • The database classes now fill an ephemeral dictionary of filenames that couldn’t be used to successfully load a tag file, available from within the _corrupted_files attribute.

  • The Comprende class’ acache_check & cache_check context manager methods are now called aauto_cache & auto_cache.

  • Added new bytes_count & abytes_count generators to module which increment each iteration & yield the results as bytes.

  • Removed the akeypair & keypair functions from the package. Their successors are the asingle_use_key & single_use_key methods in the AsyncKeys & Keys classes. The attempt is to clarify & put constraints on the interface for creating a bundle of key material that has a single-use-only salt attached, as well as the pid value.

  • Moved ciphertext encoding functions into the BytesIO class from the global module.

  • Split PrimeGroups into two classes, one higher-level class by the same name & a BasePrimeGroups class. The former also has some added functionality for masking the order of bytes in a sequence using an modular exponentiation.

  • The Hasher class now has functionality added to mask the order of a bytes sequence with a modular multiplication.

  • Fixed the name of the project in the attribution lines in several source files.

  • Reconciled tests with the major changes in this release.

  • The old identity key for the package that was signed by the gnupg identity key was shredded & replaced with a new signed key.

  • Several bug fixes to the automated code signing.

Changes for version 0.17.0

Major Changes

  • Security Patch: The HMAC verifiers on ciphertexts did not include the salt or pid values when deriving the HMAC. This associated data can therefore be changed to cause a party to decrypt a past ciphertext with a salt or pid of an attacker’s choosing. This is a critical vulnerability & it is highly recommended all users update. The fix is to hash the ciphertext, salt & pid together & sending that hash into the validator to have the HMAC created / tested. This change will cause all prior ciphertexts to be marked invalid by the validator.

  • Refactored the names of the Comprende cipher methods to better communicate their intended use as lower level tools that cannot be used on their own to obtain authenticated, CCA or CPA secure encryption.

  • Added more comprehensive tests for X25519 & Ed25519 classes, as well as the protocols that utilize the X25519 ecdh exchange. Fixed some bugs in the process.

  • X25519 instances that contain a secret key now have access to protocol methods which automatically pass their key in as a keyword argument. This simplifies their usage further.

  • Incorporated the new Hasher class into the package’s random number generator to improve its entropy production.

Minor Changes

  • Various fixes to typos, docstrings & tutorials.

  • New tutorials & docs added.

  • Changed the default table in ByteIO ‘s json_to_ascii, ajson_to_ascii, ascii_to_json & aascii_to_json to the URL_SAFE_TABLE to facilitate the creation of urlsafe_tokens.

  • Removed all code in the Ropake class that was used to create a default database to store a default salt for users. All of that functionality is expected to be handled by the database classes’ token & profile creation tools.

  • Fixed bug in package signing script that called hex from a string.

  • Updated the package signing script to include these metadata in the signatures of the ephemeral keys: name of the package, version, the date in seconds.

  • Added metadata to the setup.cfg file.

  • Make passcrypt objects available from the keygens module.

  • Add more consistent ability within Ropake class to specify a time-to-live for protocol messages.

  • Added check to make sure instances of X25519 & Ed25519 are not trying to import a new secret key once they already have one. This won’t be allowed in favor of creating a new object for a new secret key.

  • Fixed bug in database classes’ bytes ciphers which called themselves recursively instead of calling the global functions of the same name.

Changes for version 0.16.0

Major Changes

  • All Database & AsyncDatabase filenames have been converted to base36 to aid in making the manifest files & the databases as a whole more space efficient. These changes are not backwards compatible.

  • More work was done to clean up the databases & make them more efficient, as well as equalize the sizes of the database files to mitigate leaking metadata about what they might contain.

  • Added new X25519 & Ed25519 classes that greatly simplify the usage of the cryptography module’s 25519 based tools. They also help organize the codebase better – where Ropake was holding onto all of the asymmetric tooling even though those tools were not part of the Ropake protocol.

  • New base & helper Asymmetric25519 & BaseEllipticCurve classes were added as well to facilitate the reorganization.

  • Many methods in Ropake were turned private to simplify & clean up the interface so its intended use as a protocol is more clear for users.

  • Added the time-to-live functionality to Ropake decryption functions. The TIMEOUT attribute on the class can also be changed to import a global time-to-live for all Ropake ciphertexts.

  • Removed all nc_ hash functions from the package/ module.

  • The Namespace class now has a keys method so that namespaces can be unpacked using star-star syntax.

  • Because of the ongoing failures of gnupg, we are moving away from signing our packages with gnupg. Our new Ed25519 keys will be from the cryptography package, & we’ll sign those with our gnupg key as a secondary form of attestation. Our package signing will be automated in the file & the methods we use will be transparent in the code. The new signatures for each package version will be placed in the file SIGNATURES.txt.

Minor Changes

  • Many fixes & additions to docstrings & tutorials.

  • Massive refactorings, cleanups & typo fixes across the library, especially in the database classes, Ropake & the ciphers module.

  • Added comprehensive functional tests for the Ropake class.

  • Added BASE_36_TABLE to the commons module.

  • Fixed metadata issues in that caused upload issues to pypi.

  • The generate_profile, load_profile, agenerate_profile & aload_profile database methods now accept arbitrary keyword arguments that get passed into the database’s __init__ constructor.

  • username & password are now required keyword-only arguments to the agenerate_profile_tokens & generate_profile_tokens classmethods.

  • The aload & load database methods now take a manifest kwarg that when toggled True will also refresh the manifest file from disk.

  • Now when a database object is ordered to delete itself, the entirety of the instance’s caches & attribute values are cleared & deleted.

  • Filled out the references to strong key generators & protocols in the keygens module.

Changes for version 0.15.0

Major Changes

  • Security Patch: The previous update left the default salt stored by the Ropake class on the user filesystem as an empty string for new files that were created since the asalt & salt functions were switched to producing 256-bit values instead of 512-bits. This bug has now been fixed.

  • An 8 byte timestamp is now prepended to each plaintext during the padding step. The decryption functions now take a ttl kwarg which will measure & enforce a time-to-live for ciphertexts under threat of TimeoutError.

  • Added new profile feature to the database classes. This standardizes & simplifies the process for users to open databases using only low-entropy “profile” information such as username, password, *credentials & an optional salt a user may have access to. The new agenerate_profile_tokens, generate_profile_tokens, agenerate_profile, generate_profile, aprofile_exists, profile_exists, aload_profile, load_profile, adelete_profile & delete_profile functions are the public part of this new feature.

  • Some more database class attributes have been turned private to clean up the api.

  • Fixed typo in __exit__ method of Database class which referenced a method which had its name refactored, leading to a crash.

  • Shifted the values in the primes dictionary such that the key for each element in the dictionary is the exclusive maximum of each prime in that element. Ex: primes[512][-1].to_bytes(64, “big”) is now valid. Whereas before, primes[512] was filled with primes that were 64 bytes and 1 bit long, making them 65 byte primes. This changes some of the values of constants in the package & therefore some values derived from those constants.

  • Slimmed down the number of elements in the primes & bits dictionaries, reducing the size of the package a great deal. primes now contains two primes in each element, the first is the minimum prime of that bit length, the latter the maximum.

  • Added URLSAFE_TABLE to the package.

  • Made salt & pid & ttl keyword only arguments in key generators & encryption / decryption functions, further tighening up the api.

Minor Changes

  • Added this_second function to asynchs module for integer time.

  • Added apadding_key, padding_key, aplaintext_stream & plaintext_stream functions to the ciphers module.

  • Added apadding_key, padding_key to the keygens module & AsyncKeys & Keys classes.

  • Added axi_mix, xi_mix, acheck_timestamp, check_timestamp, to the generics module.

  • Added acsprbg, csprbg, asalt, salt, apadding_key, padding_key, aplaintext_stream & plaintext_stream functions to OneTimePad class as staticmethod & instance methods.

  • Added acheck_timestamp & check_timestamp functions to the BytesIO class.

  • Added adeniable_filename & deniable_filename to the paths module.

  • Removed check for falsey data in encryption functions. Empty data is & should be treated as valid plaintext.

  • Various refactorings, docstring fixes & efficiency improvements.

  • Added some new tests for database profiles.

Changes for version 0.14.0

Major Changes

  • Security patch: The apad_bytes, pad_bytes, adepad_bytes & depad_bytes functions were changed internally to execute in a more constant time. The variations were small for 256-byte buffers (the default), but can grow very wide with larger buffers. The salt in the package’s encryption utilities is now used to derive the plaintext’s padding, making each padding unique.

  • Unified the types of encodings the library’s encryption functions utilize for producing ciphertext. This includes databases. They now all use the LIST_ENCODING. This greatly increases the efficiency of the databases’ encryption/decryption, save/load times. And this encoding is more space efficient. This change is backwards incompatible.

  • The LIST_ENCODING specification was also changed to produce smaller ciphertexts. The salt is no longer encrypted & included as the first 256 byte chunk of ciphertext. It is now packaged along with ciphertext in the clear & is restricted to being a 256-bit hex string.

  • The interfaces for the Database & AsyncDatabase were cleaned up. Many attributes & functions that were not intended as the public interface of the classes were made “private”. Also, the no longer used utilities for encrypting & decrypting under the MAP_ENCODING were removed.

  • Updated the abytes_xor, bytes_xor, axor & xor generators to shrink the size of the seed that’s fed into the keystream. This allows the one-time-pad cipher to be more cpu efficient.

Minor Changes

  • Fixed various typos, docstrings & tutorials that have no kept up with the pace of changes.

  • Various refactorings throughout.

  • The akeypair & keypair functions now produce a Namespace populated with a 512-bit hex key & a 256-bit hex salt to be more consistent with their intended use-case with the one-time-pad cipher.

  • Removed aencode_salt, encode_salt, adecode_salt & decode_salt functions since they are no longer used in conjunction with LIST_ENCODING ciphertexts.

  • Updated tests to recognize these changes.

  • Gave the OneTimePad class access to a BytesIO object under a new io attribute.

Changes for version 0.13.0

Major Changes

  • Security Patch: xor & axor functions that define the one-time-pad cipher had a vulnerability fixed that can leak <1-bit of plaintext. The issue was in the way keys were built, where the multiplicative products of two key segments were xor’d together. This lead to keys being slightly more likely to be positive integers, meaning the final bit had a greater than 1/2 probability of being a 0. The fix is accompanied with an overhaul of the one-time-pad cipher which is more efficient, faster, & designed with a better understanding of the way bytes are processed & represented. The key chunks now do not, & must not, surpass 256 bytes & neither should any chunk of plaintext output. Making each chunk deterministically 256 bytes allows for reversibly formatting ciphertext to & from bytes-like strings. These changes are backwards incompatible with prior versions of this package & are strongly recommended.

  • Added bytes_xor & abytes_xor functions which take in key generators which produce key segments of type bytes instead of hex strings.

  • AsyncDatabase & Database now save files in bytes format, making them much more efficient on disk space. They use the new BytesIO class in the generics module to transparently convert to & from json & bytes. This change is also not backwards compatible.

  • Removed acipher, cipher, adecipher, decipher, aorganize_encryption_streams, organize_encryption_streams, aorganize_decryption_streams, organize_decryption_streams, aencrypt, encrypt, adecrypt, decrypt, asubkeys & subkeys generators from the ciphers module & package to slim down the code, remove repetition & focus on the cipher tools that include hmac authentication.

  • Removed deprecated diffie-hellman methods in Ropake class.

  • Removed the static power10 dictionary from the package.

  • The default secret salt for the Ropake class is now derived from the contents of a file that’s in the databases directory which is chmod’d to 0o000 unless needed.

  • Made aclient_message_key, client_message_key, aserver_message_key, & server_message_key Ropake class methods to help distinguish client-to-server & server-to-client message keys which prevents replay attacks on the one-message ROPAKE protocol.

  • Added protocol coroutines to the Ropake class which allow for easily engaging in 2DH & 3DH elliptic curve exchanges for servers & clients.

  • Efficiency improvements to the aseeder & seeder generator functions in the randoms module. This affects the acsprng & csprng objects & all the areas in the library that utilize those objects.

  • Changed the repr behavior of Comprende instances to redact all args & kwargs by default to protect cryptographic material from unintentionally being displayed on user systems. The repr can display full contents by calling the enable_debugging method of the DebugControl class.

  • All generator functions decorated with comprehension are now given a root attribute. This allows direct access to the function without needing to instantiate or run it as a Comprende object. This saves a good deal of cpu & time in the overhead that would otherwise be incurred by the class. This is specifically more helpful in tight &/or lower-level looping.

Minor Changes

  • Various refactorings across the library.

  • Fixed various typos, bugs & inaccurate docstrings throughout the library.

  • Add chown & chmod functions to the asynchs.aos module.

  • Now makes new multiprocessing.Manager objects in the asynchs.Processes & asynchs.Threads classes to avoid errors that occur when using a stale object whose socket connections are closed.

  • Changed Ropake class’ adb_login & db_login methods to adatabase_login_key & database_login_key. Also, fix a crash bug in those methods.

  • Changed Ropake class’ aec25519_pub, ec25519_pub, aec25519_priv & ec25519_priv methods to aec25519_public_bytes, ec25519_public_bytes, aec25519_private_bytes & ec25519_private_bytes.

  • Added low-level private methods to Ropake class which do derivation & querying of the default class key & salt.

  • Behavior changes to the ainverse_int & inverse_int functions in the generics module to allow handling bases represented in str or bytes type strings.

  • Behavior & name changes to the abinary_tree & binary_tree functions in the generics module to abuild_tree & build_tree. They now allow making uniform trees of any width & depth, limited only by the memory in a user’s machine.

  • Provided new acsprbg & csprbg objects to the library that return 512-bits of cryptographically secure pseudo-random bytes type strings. They are made by the new abytes_seeder & bytes_seeder generators.

  • The csprng, acsprng, csprbg & acsprbg objects were wrapped in functions that automatically restart the generators if they’re stalled / interrupted during a call. This keeps the package from melting down if it can no longer call the CSPRNGs for new entropy.

  • Cleaned up & simplified table_key functions in the keygens module.

  • Changed the output of asafe_symm_keypair & safe_symm_keypair functions to contain bytes values not their hex-only representation. Also removed these functions from the main imports of the package since they are slow & their main contribution is calling arandom_number_generator & random_number_generator to utilize a large entropy pool when starting CSPRNGs.

  • Added new values to the bits dictionary.

  • Added apad_bytes, pad_bytes, adepad_bytes & depad_bytes functions which use shake_256 to pad/depad plaintext bytes to & from multiples of 256 bytes. They take in a key to create the padding. This method is intended to also aid in protecting against padding oracle attacks.

Changes for version 0.12.0

Major Changes

  • The OPAKE protocol was renamed to ROPAKE, an acronym for Ratcheting Opaque Password Authenticated Key Exchange. This change was necessary since OPAKE is already a name for an existing PAKE protocol. This change also means the Opake class name was changed to Ropake.

  • The Ropake class’ registration algorithm was slightly modified to use the generated Curve25519 shared_key an extra time in the key derivation process. This shouldn’t break any currently authenticated sessions.

  • The asyncio_contextmanager package is no longer a listed dependency in The main file from that package was copied over into the /aiootp directory in order to remove the piece of code that caused warnings to crop up when return values were retrieved from async generators. This change will put an end to this whack-a-mole process of trying to stop the warnings with try blocks scattered about the codebase.

  • Added asave_tag, save_tag, asave_file & save_file methods to the database classes so that specific entries can be saved to disk without having to save the entire database which is much more costly. The manifest file isn’t saved to disk when these methods are used, so if a tag file isn’t already saved in the database, then the saved files will not be present in the manifest or in the cache upon subsequent loads of the database. The saved file will still however be saved on the filesystem, though unbeknownst to the database instance.

  • The Namespace class now redacts all obvious key material in instance repr’s, which is any 64+ hex character string, or any number with 64+ decimal digits.

  • Removed the experimental recursive value retrieval within Comprende’s __aexamine_sent_exceptions & __examine_sent_exceptions methods. This change leads to more reliable & faster code, in exchange for an unnecessary feature being removed.

  • Bug fix of the auuids & uuids methods by editing the code in the asyncio_contextmanager dependency & using the patched package instead of the comprehension decorator for the arelay & relay methods of Comprende. Their internal algorithms was also updated to be simpler, but are incompatible with the outputs of past versions of these methods.

Minor Changes

  • Various refactorings & documentation additions / modifications throughout the library.

  • Various small bug fixes.

  • The shared keys derived from the Ropake protocol are now returned in a Namespace object instead of a raw dictionary, which allows the values to be retrieved by dotted &/or bracketed lookup.

  • The atest_hmac & test_hmac algorithms / methods were made more efficient & were refactored. Now they call atime_safe_equality & time_safe_equality internally, which are new methods that can apply the non-constant time but randomized timing comparisons on any pairs of values.

Changes for version 0.11.0

Major Changes

  • The Opake protocol was made greatly more efficient. This was done by replacing the diffie-hellman verifiers with a hash & xor commit & reveal system. Most hashing was made more efficient my using quicker & smaller sha_512 function instead of nc_512, & streamlining the protocol.

  • The Opake.client & Opake.client_registration methods now take an instantiated client database instead of client credentials which improves security, efficiency & usability. This change reduces the amount of exposure received by user passwords & other credentials. It also simplifies usage of the protocol by only needing to carry around a database instead of a slew of credentials, which is also faster, since the credentials are passed through the cpu & memory hard passcrypt function everytime to open the database.

Minor Changes

  • Heavy refactorings & documentation additions / modifications of the Opake class. Removed the Opake.ainit_database & Opake.init_database methods, & made the salt default argument parameter in Opake.aclient_database, Opake.client_database, Opake.adb_login & Opake.db_login into a keyword only argument so any extra user defined credentials are able to be passed without specifying a salt.

  • The decorators for the Comprende.arelay & Comprende.relay methods were changed from @asyncio_contextmanager.async_contextmanager to @comprehension() to stop that package from raising exceptions when we retrieve return values from async generators.

Changes for version 0.10.1

Major Changes

  • Added Processes & Threads classes to which abstract spawning & getting return values from async & sync functions intended to be run in threads, processes or pools of the former types. This simplifies & adds time control to usages of processes & threads throughout the library.

  • Reduced the effectiveness of timing analysis of the modular exponentiation in the Opake class’ verifiers by making the process return values only after discrete intervals of time. Timing attacks on that part of the protocol may still be viable, but should be significantly reduced.

  • Bug fix in Comprende which should take care of warnings raised from the aiocontext package when retrieving async generator values by raising UserWarning within them.

Minor Changes

  • Heavy refactorings of the Opake class.

  • Various refactorings & cleanups around the package.

  • Further add return_exceptions=True flag to gather calls in

  • Added is_registration & is_authentication which take a client hello message that begin the Opake protocol, & return False if the message is not either a registration or authentication message, respectively, & return "Maybe" otherwise, since these functions can’t determine without running the protocol whether or not the message is valid.

Changes for version 0.10.0

Major Changes

  • Added a new oblivious, one-message, password authenticated key exchange protocol class in aiootp.ciphers.Opake. It is a first attempt at the protocol, which works rather well, but may be changed or cleaned up in a future update.

  • Added the cryptography package as a dependency for elliptic curve 25519 diffie-hellman key exchange in the Opake protocol.

  • Fix buggy data processing functions in module.

  • Added silent flag to AsyncDatabase & Database methods, which allows their instances to finish initializing even if a file is missing from the filesystem, normally causing a FileNotFoundError. This makes trouble-shooting corrupted databases easier.

  • Added new aiootp.paths.SecurePath function which returns the path to a unique directory within the database’s default directory. The name of the returned directory is a cryptographic value used to create & open the default database used by the Opake class to store the cryptographic salt that secures the class’ client passwords. It’s highly recommended to override this default database by instantiating the Opake class with a custom user-defined key. The instance doesn’t need to be saved, since all the class’ methods are either class or static methods. The __init__ method only changes the class’ default database to one opened with the user-defined key &/or directory kwargs, & should really only be done once at the beginning of an application.

Minor Changes

  • Various refactorings & cleanups around the package.

  • Added Comprende class feature to return the values from even the generators within an instance’s arguments. This change better returns values to the caller from chains of Comprende generators.

  • Fixed commons.BYTES_TABLE missing values.

  • Added commons.DH_PRIME_4096_BIT_GROUP_16 & commons.DH_GENERATOR_4096_BIT_GROUP_16 constants for use in the Opake protocol’s public key verifiers.

  • Added other values to the module.

  • Added new very large no-collision hash functions to the module used to xor with diffie-hellman public keys in the Opake class.

  • Added new wait_on & await_on Comprende generators to which waits for a queue or container to be populated & yields it whenever it isn’t empty.

Changes for version 0.9.3

Major Changes

  • Speed & efficiency improvements in the Comprende class & azip.

Minor Changes

  • Various refactorings & code cleanups.

  • Added apop & pop Comprende generators to the library.

  • Switched the default character table in the ato_base, to_base, afrom_base, & from_base chainable generator methods from the 62 character ASCII_ALPHANUMERIC table, to the 95 character ASCII_TABLE.

  • Made the digits generators in automatically create a new cryptographically secure key if a key isn’t passed by a user.

  • Some extra data processing functions added to

Changes for version 0.9.2

Major Changes

  • Added passcrypt & apasscrypt instance methods to OneTimePad, Keys, & AsyncKeys classes. They produce password hashes that are not just secured by the salt & passcrypt algorithm settings, but also by their main symmetric instance keys. This makes passwords infeasible to crack without also compromising the instance’s 512-bit key.

Minor Changes

  • Further improvements to the random number generator in Made its internals less sequential thereby raising the bar of work needed by an attacker to successfully carry out an order prediction attack.

  • Added checks in the Passcrypt class to make sure both a salt & password were passed into the algorithm.

  • Switched PermissionError exceptions in Passcrypt._validate_args to ValueError to be more consistent with the rest of the class.

  • Documentation updates / fixes.

Changes for version 0.9.1

Minor Changes

  • Now any falsey values for the salt keyword argument in the library’s keys, akeys, bytes_keys, abytes_keys, subkeys, & asubkeys infinite keystream generators, & other functions around the library, will cause them to generate a new cryptographically secure pseudo-random value for the salt. It formerly only did this when salt was None.

  • The seeder & aseeder generators have been updated to introduce 512 new bits of entropy from secrets.token_bytes on every iteration to ensure that the CSPRNG will produce secure outputs even if its internal state is somehow discovered. This also allows for simply calling the CSPRNG is enough, there’s no longer a strong reason to pass new entropy into it manually, except to add even more entropy as desired.

  • Made size the last keywordCHECKSUMS.txt argument in encrypt & aencrypt to better mirror the signatures for rest of the library.

  • Added token_bits & atoken_bits functions to which are renamings of secrets.randbits.

  • Refactored & improved the security og’s random number generator.

Changes for version 0.9.0

Major Changes

  • Added hmac codes to ciphertext for the following functions: json_encrypt, ajson_encrypt, bytes_encrypt, abytes_encrypt, Database.encrypt & AsyncDatabase.aencrypt. This change greatly increases the security of ciphertext by ensuring it hasn’t been modified or tampered with maliciously. One-time pad ciphertext is maleable, so without hmac validation it can be changed to successfully allow decryption but return the wrong plaintext. These functions are the highest level abstractions of the library for encryption/decryption, which made them excellent targets for this important security update. As well, it isn’t easily possible for the library to provide hmac codes for generators that produce ciphertext, because the end of a stream of ciphertext isn’t known until after the results have left the scope of library code. So users will need to produce their own hmac codes for generator ciphertext unless we find an elegant solution to this issue. These functions now all return dictionaries with the associated hmac stored in the "hmac" entry. The bytes functions formerly returned lists, now their ciphertext is available from the "ciphertext" entry. And, all database files will have an hmac attached to them now. These changes were designed to still be compatible with old ciphertexts but they’ll likely be made incompatible by the v0.11.x major release.

  • Only truthy values are now valid key keyword arguments in the library’s keys, akeys, bytes_keys, abytes_keys, subkeys, & asubkeys infinite keystream generators. Also now seeding extra entropy into csprng & acsprng when salt is falsey within them.

  • Only truthy values are now valid for password & salt arguments in apasscrypt, passcrypt & their variants.

Minor Changes

  • Updates to documentation & README.rst tutorials.

  • The kb, cpu, & hardness arguments in sum_passcrypt & asum_passcrypt chainable generator methods were switched to keyword only arguments.

Changes for version 0.8.1

Major Changes

  • Added sum_passcrypt & asum_passcrypt chainable generator methods to Comprende class. They cumulatively apply the passcrypt algorithm to each yielded value from an underlying generator with the passcrypt’d prior yielded result used as a salt. This allows making proofs of work, memory & space-time out of iterations of the passcrypt algorithm very simple.

Minor Changes

  • Various inaccurate docstrings fixed.

  • Various refactorings of the codebase.

  • Made kb, cpu, & hardness arguments into keyword only arguments in AsyncDatabase & Database classes.

  • The length keyword argument in functions around the library was changed to size to be consistent across the whole package. Reducing the cognitive burden of memorizing more than one name for the same concept.

  • Various efficiency boosts.

  • Edits to README.rst.

  • Added encode_salt, aencode_salt, decode_salt & adecode_salt functions to the library, which gives access to the procedure used to encrypt & decrypt the random salt which is often the first element produced in one-time pad ciphertexts.

  • Added cryptographically secure pseudo-random values as default keys in encryption functions to safeguard against users accidentally encrypting data without specifying a key. This way, such mistakes will produce ciphertext with an unrecoverable key, instead of without a key at all.

Changes for version 0.8.0

Major Changes

  • Fix test_hmac, atest_hmac functions in the keys & database classes. The new non-constant-time algorithm needs a random salt to be added before doing the secondary hmac to prevent some potential exotic forms of chosen plaintext/ciphertext attacks on the algorithm. The last version of the algorithm should not be used.

  • The Keys & AsyncKeys interfaces were overhauled to remove the persistance of instance salts. They were intended to be updated by users with the reset & areset methods, but that cannot be guaranteed easily through the class, so it is an inappropriate interface since reusing salts for encryption is completely insecure. The instances do still maintain state of their main encryption key, & new stateful methods for key generation, like mnemonic & table_key, have been added. The state & astate methods have been removed.

  • Gave OneTimePad instances new stateful methods from the module & keys classes. Its instances now remember the main symmetric key behind the key property & automatically passes it as a keyword argument to the methods in OneTimePad.instance_methods.

Minor Changes

  • Update CHANGES.rst file with the updates that were not logged for v0.7.1.

  • BYTES_TABLE was turned into a list so that the byte characters can be retrieved instead of their ordinal numbers.

Changes for version 0.7.1

Major Changes

  • Fix a mistake in the signatures of passcrypt & apasscrypt. The args ``kb, cpu & hardness were changed into keyword only arguments to mitigate user mistakes, but the internal calls to those functions were still using positional function calls, which broke the api. This issue is now fixed.

Changes for version 0.7.0

Major Changes

  • Replaced usage of bare random module functions, to usage of an instance of random.Random to keep from messing with user’s settings on that module.

  • Finalized the algorithm for the passcrypt & apasscrypt functions. The algorithm is now provably memory & cpu hard with a wide security margin with adequate settings. The algorithm isn’t likely change with upcoming versions unless a major flaw is found.

  • The default value for the cpu argument in passcrypt & apasscrypt is now 3 & now directly determines how many hash iterations are done for each element in the memory cache. This provides much more responsiveness to users & increases the capacity to impact resource cost with less tinkering.

  • Switched the AsyncKeys.atest_hmac & Keys.test_hmac methods to a scheme which is not constant time, but which instead does not leak useful information. It does this by not comparing the hmacs of the data, but of a pair of secondary hmacs. The timing analysis itself is now dependant on knowledge of the key, since any conclusions of such an analysis would be unable correlate its findings with any supplied hmac without it.

  • Added test_hmac & atest_hmac to the database classes, & changed their hmac algorithm from sha3_512 to sha3_256.

Minor Changes

  • Various code cleanups, refactorings & speedups.

  • Several fixes to inaccurate documentation.

  • Several fixes to inaccurate function signatures.

  • Added mnemonic & amnemonic key generators to with a wordlist 2048 entries long. A custom wordlist can also be passed in.

  • Minor changes in Comprende to track down a bug in the functions that use the asyncio_contextmanager package. It causes a warning when asking async generators to return (not yield) values.

  • Some refactoring of random_number_generator & arandom_number_generator.

Changes for version 0.6.0

Major Changes

  • Replaced the usage of os.urandom within the package with secrets.token_bytes to be more reliable across platforms.

  • Replaced several usages of random.randrange within to calls to secrets.token_bytes which is faster & more secure. It now also seeds random module periodically prior to usage.

  • Changed the internal cache sorting algorithm of passcrypt & apasscrypt functions. The key function passed to list.sort(key=key) now not only updates the hashlib.sha3_512 proof object with each element in the cache, but with it’s own current output. This change is incompatible with previous versions of the functions. The key function is also trimmed down of unnecessary value checking.

  • The default value for the cpu argument in passcrypt & apasscrypt is now 40_000. This is right at the edge of when the argument begins impacting the cpu work needed to comptute the password hash when the kb argument is the default of 1024.

  • Switched the AsyncKeys.atest_hmac & Keys.test_hmac methods to a constant time algorithm.

Minor Changes

  • Various code cleanups, refactorings & speedups.

  • Added a concurrent.futures.ThreadPoolExecutor instance to the asynchs module for easily spinning off threads. It’s available under asynchs.thread_pool.

  • Added sort & asort chainable generator method to the Comprende class. They support sorting by a key sorting function as well.

  • Changed the name of asynchs.executor_wrapper to asynchs.wrap_in_executor.

  • Changed the name of randoms.non0_digit_stream, randoms.anon0_digit_stream, randoms.digit_stream & randoms.adigit_stream to randoms.non_0_digits, randoms.anon_0_digits, randoms.digits & randoms.adigits.

  • Several fixes to inaccurate documentation.

  • apasscrypt & Passcrypt.anew now use the synchronous version of the algorithm internally because it’s faster & it doesn’t change the parallelization properties of the function since it’s already run automatically in another process.

  • Added shuffle, ashuffle, unshuffle, & aunshuffle functions to that reorder sequences pseudo-randomly based on their key & salt keyword arguments.

  • Fixed bugs in AsyncKeys &

  • Added debugger & adebugger chainable generator methods to the Comprende class which benchmarks & inspects running generators with an inline syntax.

Changes for version 0.5.1

Major Changes

  • Fixed a bug in the methods auuids & uuids of the database classes that assigned to a variable within a closure that was nonlocal but which wasn’t declared non-local. This caused an error which made the methods unusable.

  • Added passcrypt & apasscrypt functions which are designed to be tunably memory & cpu hard password-based key derivation function. It was inspired by the scrypt protocol but internally uses the library’s tools. It is a first attempt at the protocol, it’s internal details will likely change in future updates.

  • Added bytes_keys & abytes_keys generators, which are just like the library’s keys generator, except they yield the concatenated sha3_512.digest instead of the sha3_512.hexdigest.

  • Added new chainable generator methods to the Comprende class for processing bytes, integers, & hex strings into one another.

Minor Changes

  • Various code cleanups.

  • New tests added to the test suite for passcrypt & apasscrypt.

  • The Comprende class’ alist & list methods can now be passed a boolean argument to return either a mutable list directly from the lru_cache, or a copy of the cached list. This list is used by the generator itself to yield its values, so wilely magic can be done on the list to mutate the underlying generator’s results.

Changes for version 0.5.0

Major Changes

  • Added interfaces in Database & AsyncDatabase to handle encrypting & decrypting streams (Comprende generators) instead of just raw json data. They’re methods called encrypt_stream, decrypt_stream, aencrypt_stream, & adecrypt_stream.

  • Changed the attribute _METATAG used by Database & AsyncDatabase to name the metatags entry in the database. This name is smaller, cleaner & is used to prevent naming collisions between user entered values & the metadata the classes need to organize themselves internally. This change will break databases from older versions keeping them from accessing their metatag child databases.

  • Added the methods auuids & uuids to AsyncDatabase & Database which return coroutines that accept potentially sensitive identifiers & turns them into salted size length hashes distinguished by a salt & a category.

Minor Changes

  • Various code & logic cleanups / speedups.

  • Refactorings of the Database & AsyncDatabase classes.

  • Various inaccurate docstrings fixed.

Changes for version 0.4.0

Major Changes

  • Fixed bug in aiootp.abytes_encrypt function which inaccurately called a synchronous Comprende end-point method on the underlying async generator, causing an exception and failure to function.

  • Changed the procedures in akeys & keys that generate their internal key derivation functions. They’re now slightly faster to initialize & more theoretically secure since each internal state is fed by a seed which isn’t returned to the user. This encryption algorithm change is incompatible with the encryption algorithms of past versions.

Minor Changes

  • Various code cleanups.

  • Various inaccurate docstrings fixed.

  • Keyword arguments in Keys().test_hmac & AsyncKeys().atest_hmac had their order switched to be slightly more friendly to use.

  • Added documentation to README.rst on the inner workings of the one-time-pad algorithm’s implementation.

  • Made Compende.arandom_sleep & Compende.random_sleep chainable generator methods.

  • Changed the Compende.adelimit_resize & Compende.delimit_resize algorithms to not yield inbetween two joined delimiters in a sequence being resized.

Changes for version 0.3.1

Minor Changes

  • Fixed bug where a static method in AsyncDatabase & Database was wrongly labelled a class method causing a failure to initialize.

Changes for version 0.3.0

Major Changes

  • The AsyncDatabase & Database now use the more secure afilename & filename methods to derive the hashmap name and encryption streams from a user-defined tag internal to their aencrypt / adecrypt / encrypt / decrypt methods, as well as, prior to them getting called. This will break past versions of databases’ ability to open their files.

  • The package now has built-in functions for using the one-time-pad algorithm to encrypt & decrypt binary data instead of just strings or integers. They are available in aiootp.abytes_encrypt, aiootp.abytes_decrypt, aiootp.bytes_encrypt & aiootp.bytes_decrypt.

  • The Comprende class now has generators that do encryption & decryption of binary data as well. They are available from any Comprende generator by the abytes_encrypt, abytes_decrypt, bytes_encrypt & bytes_decrypt chainable method calls.

Minor Changes

  • Fixed typos and inaccuracies in various docstrings.

  • Added a module to handle inserting functionality from higher-level to lower-level modules and classes.

  • Various code clean ups and redundancy eliminations.

  • AsyncKeys & Keys classes now only update their self.salt key by default when their areset & reset methods are called. This aligns more closely with their intended use.

  • Added arandom_sleep & random_sleep chainable methods to the Comprende class which yields outputs of generators after a random sleep for each iteration.

  • Added several other chainable methods to the Comprende class for string & bytes data processing. They’re viewable in Comprende.lazy_generators.

  • Added new, initial tests to the test suite.

Changes for version 0.2.0

Major Changes

  • Added ephemeral salts to the AsyncDatabase & Database file encryption procedures. This is a major security fix, as re-encryption of files with the same tag in a database with the same open key would use the same streams of key material each time, breaking encryption if two different versions of a tag file’s ciphertext stored to disk were available to an adversary. The database methods encrypt, decrypt, aencrypt & adecrypt will now produce and decipher true one-time pad ciphertext with these ephemeral salts.

  • The aiootp.subkeys & aiootp.asubkeys generators were revamped to use the keys & akeys generators internally instead of using their own, slower algorithm.

  • AsyncDatabase file deletion is now asynchronous by running the builtins.os.remove function in an async thread executor. The decorator which does the magic is available at aiootp.asynchs.executor_wrapper.

Minor Changes

  • Fix typos in __root_salt & __aroot_salt docstrings. Also replaced the hash(self) argument for their lru_cache & alru_cache with a secure hmac instead.

  • add gi_frame, gi_running, gi_code, gi_yieldfrom, ag_frame, ag_running, ag_code & ag_await properties to Comprende class to mirror async/sync generators more closely.

  • Remove ajson_encrypt, ajson_decrypt, json_encrypt, json_decrypt functions’ internal creation of dicts to contain the plaintext. It was unnecessary & therefore wasteful.

  • Fix docstrings in OneTimePad methods mentioning parent kwarg which is a reference to deleted, refactored code.

  • Fix incorrect docstrings in databases namestream & anamestream methods.

  • Added ASYNC_GEN_THROWN constant to Comprende class to try to stop an infrequent & difficult to debug RuntimeError when async generators do not stop after receiving an athrow.

  • Database tags are now fully loaded when they’re copied using the methods into_namespace & ainto_namespace.

  • Updated inaccurate docstrings in map_encrypt, amap_encrypt, map_decrypt & amap_decrypt OneTimePad methods.

  • Added acustomize_parameters async function to aiootp.generics module.

  • Various code clean ups.

Changes for version 0.1.0

Minor Changes

  • Initial version.

Major Changes

  • Initial version.

Known Issues …………………………….. Table Of Contents

  • The test suite for this software is under construction, & what tests have been published are currently inadequate to the needs of cryptography software.

  • This package is currently in beta testing & active development, meaning major changes are still possible when there are really good reasons to do so. Contributions are welcome. Send us a message if you spot a bug or security vulnerability:

Project details

Download files

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

Source Distribution

aiootp-0.22.1.tar.gz (418.7 kB view hashes)

Uploaded source

Built Distribution

aiootp-0.22.1-py37.py38.py39.py310.py311-none-any.whl (213.0 kB view hashes)

Uploaded py37 py38 py39 py310 py311

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page