mirror of
https://github.com/TheAlgorithms/Python.git
synced 2024-11-30 16:31:08 +00:00
52a987ea2f
* Added docstrings,doctests and fixed a bug * Added docstrings,doctests and fixed a bug * Added docstrings,doctests and fixed a bug * Added docstrings and doctests with a bug fix * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Added docstrings and doctests with a bug fix * Update ciphers/trifid_cipher.py Co-authored-by: Christian Clauss <cclauss@me.com> * Update ciphers/trifid_cipher.py Co-authored-by: Christian Clauss <cclauss@me.com> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Docstrings edit * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Update trifid_cipher.py * Update pyproject.toml --------- Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Co-authored-by: Christian Clauss <cclauss@me.com>
211 lines
6.8 KiB
Python
211 lines
6.8 KiB
Python
"""
|
|
The trifid cipher uses a table to fractionate each plaintext letter into a trigram,
|
|
mixes the constituents of the trigrams, and then applies the table in reverse to turn
|
|
these mixed trigrams into ciphertext letters.
|
|
|
|
https://en.wikipedia.org/wiki/Trifid_cipher
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
# fmt: off
|
|
TEST_CHARACTER_TO_NUMBER = {
|
|
"A": "111", "B": "112", "C": "113", "D": "121", "E": "122", "F": "123", "G": "131",
|
|
"H": "132", "I": "133", "J": "211", "K": "212", "L": "213", "M": "221", "N": "222",
|
|
"O": "223", "P": "231", "Q": "232", "R": "233", "S": "311", "T": "312", "U": "313",
|
|
"V": "321", "W": "322", "X": "323", "Y": "331", "Z": "332", "+": "333",
|
|
}
|
|
# fmt: off
|
|
|
|
TEST_NUMBER_TO_CHARACTER = {val: key for key, val in TEST_CHARACTER_TO_NUMBER.items()}
|
|
|
|
|
|
def __encrypt_part(message_part: str, character_to_number: dict[str, str]) -> str:
|
|
"""
|
|
Arrange the triagram value of each letter of 'message_part' vertically and join
|
|
them horizontally.
|
|
|
|
>>> __encrypt_part('ASK', TEST_CHARACTER_TO_NUMBER)
|
|
'132111112'
|
|
"""
|
|
one, two, three = "", "", ""
|
|
for each in (character_to_number[character] for character in message_part):
|
|
one += each[0]
|
|
two += each[1]
|
|
three += each[2]
|
|
|
|
return one + two + three
|
|
|
|
|
|
def __decrypt_part(
|
|
message_part: str, character_to_number: dict[str, str]
|
|
) -> tuple[str, str, str]:
|
|
"""
|
|
Convert each letter of the input string into their respective trigram values, join
|
|
them and split them into three equal groups of strings which are returned.
|
|
|
|
>>> __decrypt_part('ABCDE', TEST_CHARACTER_TO_NUMBER)
|
|
('11111', '21131', '21122')
|
|
"""
|
|
this_part = "".join(character_to_number[character] for character in message_part)
|
|
result = []
|
|
tmp = ""
|
|
for digit in this_part:
|
|
tmp += digit
|
|
if len(tmp) == len(message_part):
|
|
result.append(tmp)
|
|
tmp = ""
|
|
|
|
return result[0], result[1], result[2]
|
|
|
|
|
|
def __prepare(
|
|
message: str, alphabet: str
|
|
) -> tuple[str, str, dict[str, str], dict[str, str]]:
|
|
"""
|
|
A helper function that generates the triagrams and assigns each letter of the
|
|
alphabet to its corresponding triagram and stores this in a dictionary
|
|
("character_to_number" and "number_to_character") after confirming if the
|
|
alphabet's length is 27.
|
|
|
|
>>> test = __prepare('I aM a BOy','abCdeFghijkLmnopqrStuVwxYZ+')
|
|
>>> expected = ('IAMABOY','ABCDEFGHIJKLMNOPQRSTUVWXYZ+',
|
|
... TEST_CHARACTER_TO_NUMBER, TEST_NUMBER_TO_CHARACTER)
|
|
>>> test == expected
|
|
True
|
|
|
|
Testing with incomplete alphabet
|
|
>>> __prepare('I aM a BOy','abCdeFghijkLmnopqrStuVw')
|
|
Traceback (most recent call last):
|
|
...
|
|
KeyError: 'Length of alphabet has to be 27.'
|
|
|
|
Testing with extra long alphabets
|
|
>>> __prepare('I aM a BOy','abCdeFghijkLmnopqrStuVwxyzzwwtyyujjgfd')
|
|
Traceback (most recent call last):
|
|
...
|
|
KeyError: 'Length of alphabet has to be 27.'
|
|
|
|
Testing with punctuations that are not in the given alphabet
|
|
>>> __prepare('am i a boy?','abCdeFghijkLmnopqrStuVwxYZ+')
|
|
Traceback (most recent call last):
|
|
...
|
|
ValueError: Each message character has to be included in alphabet!
|
|
|
|
Testing with numbers
|
|
>>> __prepare(500,'abCdeFghijkLmnopqrStuVwxYZ+')
|
|
Traceback (most recent call last):
|
|
...
|
|
AttributeError: 'int' object has no attribute 'replace'
|
|
"""
|
|
# Validate message and alphabet, set to upper and remove spaces
|
|
alphabet = alphabet.replace(" ", "").upper()
|
|
message = message.replace(" ", "").upper()
|
|
|
|
# Check length and characters
|
|
if len(alphabet) != 27:
|
|
raise KeyError("Length of alphabet has to be 27.")
|
|
if any(char not in alphabet for char in message):
|
|
raise ValueError("Each message character has to be included in alphabet!")
|
|
|
|
# Generate dictionares
|
|
character_to_number = dict(zip(alphabet, TEST_CHARACTER_TO_NUMBER.values()))
|
|
number_to_character = {
|
|
number: letter for letter, number in character_to_number.items()
|
|
}
|
|
|
|
return message, alphabet, character_to_number, number_to_character
|
|
|
|
|
|
def encrypt_message(
|
|
message: str, alphabet: str = "ABCDEFGHIJKLMNOPQRSTUVWXYZ.", period: int = 5
|
|
) -> str:
|
|
"""
|
|
encrypt_message
|
|
===============
|
|
|
|
Encrypts a message using the trifid_cipher. Any punctuatuions that
|
|
would be used should be added to the alphabet.
|
|
|
|
PARAMETERS
|
|
----------
|
|
|
|
* message: The message you want to encrypt.
|
|
* alphabet (optional): The characters to be used for the cipher .
|
|
* period (optional): The number of characters you want in a group whilst
|
|
encrypting.
|
|
|
|
>>> encrypt_message('I am a boy')
|
|
'BCDGBQY'
|
|
|
|
>>> encrypt_message(' ')
|
|
''
|
|
|
|
>>> encrypt_message(' aide toi le c iel ta id era ',
|
|
... 'FELIXMARDSTBCGHJKNOPQUVWYZ+',5)
|
|
'FMJFVOISSUFTFPUFEQQC'
|
|
|
|
"""
|
|
message, alphabet, character_to_number, number_to_character = __prepare(
|
|
message, alphabet
|
|
)
|
|
|
|
encrypted_numeric = ""
|
|
for i in range(0, len(message) + 1, period):
|
|
encrypted_numeric += __encrypt_part(
|
|
message[i : i + period], character_to_number
|
|
)
|
|
|
|
encrypted = ""
|
|
for i in range(0, len(encrypted_numeric), 3):
|
|
encrypted += number_to_character[encrypted_numeric[i : i + 3]]
|
|
return encrypted
|
|
|
|
|
|
def decrypt_message(
|
|
message: str, alphabet: str = "ABCDEFGHIJKLMNOPQRSTUVWXYZ.", period: int = 5
|
|
) -> str:
|
|
"""
|
|
decrypt_message
|
|
===============
|
|
|
|
Decrypts a trifid_cipher encrypted message .
|
|
|
|
PARAMETERS
|
|
----------
|
|
|
|
* message: The message you want to decrypt .
|
|
* alphabet (optional): The characters used for the cipher.
|
|
* period (optional): The number of characters used in grouping when it
|
|
was encrypted.
|
|
|
|
>>> decrypt_message('BCDGBQY')
|
|
'IAMABOY'
|
|
|
|
Decrypting with your own alphabet and period
|
|
>>> decrypt_message('FMJFVOISSUFTFPUFEQQC','FELIXMARDSTBCGHJKNOPQUVWYZ+',5)
|
|
'AIDETOILECIELTAIDERA'
|
|
"""
|
|
message, alphabet, character_to_number, number_to_character = __prepare(
|
|
message, alphabet
|
|
)
|
|
|
|
decrypted_numeric = []
|
|
for i in range(0, len(message), period):
|
|
a, b, c = __decrypt_part(message[i : i + period], character_to_number)
|
|
|
|
for j in range(len(a)):
|
|
decrypted_numeric.append(a[j] + b[j] + c[j])
|
|
|
|
return "".join(number_to_character[each] for each in decrypted_numeric)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
import doctest
|
|
|
|
doctest.testmod()
|
|
msg = "DEFEND THE EAST WALL OF THE CASTLE."
|
|
encrypted = encrypt_message(msg, "EPSDUCVWYM.ZLKXNBTFGORIJHAQ")
|
|
decrypted = decrypt_message(encrypted, "EPSDUCVWYM.ZLKXNBTFGORIJHAQ")
|
|
print(f"Encrypted: {encrypted}\nDecrypted: {decrypted}")
|