X-Git-Url: https://git.njae.me.uk/?a=blobdiff_plain;f=cipher.py;h=e318aa46f8776fe24d9293e7d2b91def53d72663;hb=0e3e7f869cb661bd92d50a0407556e289c5e7bf9;hp=3e42da109840697938fe9fd16ca95f09a234574b;hpb=eaecd10e334e6d63d2fd222bc280b02febca5a1b;p=cipher-tools.git diff --git a/cipher.py b/cipher.py index 3e42da1..e318aa4 100644 --- a/cipher.py +++ b/cipher.py @@ -1,54 +1,37 @@ import string import collections -import logging -from itertools import zip_longest, cycle, chain +import math +from enum import Enum +from itertools import zip_longest, cycle, chain, count +import numpy as np +from numpy import matrix +from numpy import linalg from language_models import * +import pprint -logger = logging.getLogger(__name__) -logger.addHandler(logging.FileHandler('cipher.log')) -logger.setLevel(logging.WARNING) -#logger.setLevel(logging.INFO) -#logger.setLevel(logging.DEBUG) +## Utility functions +cat = ''.join +wcat = ' '.join +lcat = '\n'.join -modular_division_table = [[0]*26 for x in range(26)] +def pos(letter): + if letter in string.ascii_lowercase: + return ord(letter) - ord('a') + elif letter in string.ascii_uppercase: + return ord(letter) - ord('A') + else: + return '' + +def unpos(number): return chr(number % 26 + ord('a')) + + +modular_division_table = [[0]*26 for _ in range(26)] for a in range(26): for b in range(26): c = (a * b) % 26 modular_division_table[b][c] = a -def letters(text): - """Remove all non-alphabetic characters from a text - >>> letters('The Quick') - 'TheQuick' - >>> letters('The Quick BROWN fox jumped! over... the (9lazy) DOG') - 'TheQuickBROWNfoxjumpedoverthelazyDOG' - """ - return ''.join([c for c in text if c in string.ascii_letters]) - -def sanitise(text): - """Remove all non-alphabetic characters and convert the text to lowercase - - >>> sanitise('The Quick') - 'thequick' - >>> sanitise('The Quick BROWN fox jumped! over... the (9lazy) DOG') - 'thequickbrownfoxjumpedoverthelazydog' - """ - # sanitised = [c.lower() for c in text if c in string.ascii_letters] - # return ''.join(sanitised) - return letters(text).lower() - -def ngrams(text, n): - """Returns all n-grams of a text - - >>> ngrams(sanitise('the quick brown fox'), 2) # doctest: +NORMALIZE_WHITESPACE - ['th', 'he', 'eq', 'qu', 'ui', 'ic', 'ck', 'kb', 'br', 'ro', 'ow', 'wn', - 'nf', 'fo', 'ox'] - >>> ngrams(sanitise('the quick brown fox'), 4) # doctest: +NORMALIZE_WHITESPACE - ['theq', 'hequ', 'equi', 'quic', 'uick', 'ickb', 'ckbr', 'kbro', 'brow', - 'rown', 'ownf', 'wnfo', 'nfox'] - """ - return [text[i:i+n] for i in range(len(text)-n+1)] def every_nth(text, n, fillvalue=''): """Returns n strings, each of which consists of every nth character, @@ -64,8 +47,8 @@ def every_nth(text, n, fillvalue=''): >>> every_nth(string.ascii_lowercase, 5, fillvalue='!') ['afkpuz', 'bglqv!', 'chmrw!', 'dinsx!', 'ejoty!'] """ - split_text = [text[i:i+n] for i in range(0, len(text), n)] - return [''.join(l) for l in zip_longest(*split_text, fillvalue=fillvalue)] + split_text = chunks(text, n, fillvalue) + return [cat(l) for l in zip_longest(*split_text, fillvalue=fillvalue)] def combine_every_nth(split_text): """Reforms a text split into every_nth strings @@ -77,7 +60,7 @@ def combine_every_nth(split_text): >>> combine_every_nth(every_nth(string.ascii_lowercase, 26)) 'abcdefghijklmnopqrstuvwxyz' """ - return ''.join([''.join(l) + return cat([cat(l) for l in zip_longest(*split_text, fillvalue='')]) def chunks(text, n, fillvalue=None): @@ -130,7 +113,7 @@ def deduplicate(text): return list(collections.OrderedDict.fromkeys(text)) -def caesar_encipher_letter(letter, shift): +def caesar_encipher_letter(accented_letter, shift): """Encipher a letter, given a shift amount >>> caesar_encipher_letter('a', 1) @@ -149,14 +132,29 @@ def caesar_encipher_letter(letter, shift): 'y' >>> caesar_encipher_letter('a', -1) 'z' + >>> caesar_encipher_letter('A', 1) + 'B' + >>> caesar_encipher_letter('é', 1) + 'f' """ + # letter = unaccent(accented_letter) + # if letter in string.ascii_letters: + # if letter in string.ascii_uppercase: + # alphabet_start = ord('A') + # else: + # alphabet_start = ord('a') + # return chr(((ord(letter) - alphabet_start + shift) % 26) + + # alphabet_start) + # else: + # return letter + + letter = unaccent(accented_letter) if letter in string.ascii_letters: + cipherletter = unpos(pos(letter) + shift) if letter in string.ascii_uppercase: - alphabet_start = ord('A') + return cipherletter.upper() else: - alphabet_start = ord('a') - return chr(((ord(letter) - alphabet_start + shift) % 26) + - alphabet_start) + return cipherletter else: return letter @@ -181,12 +179,14 @@ def caesar_encipher(message, shift): 'cdezab' >>> caesar_encipher('ab cx yz', 2) 'cd ez ab' + >>> caesar_encipher('Héllo World!', 2) + 'Jgnnq Yqtnf!' """ enciphered = [caesar_encipher_letter(l, shift) for l in message] - return ''.join(enciphered) + return cat(enciphered) def caesar_decipher(message, shift): - """Encipher a message with the Caesar cipher of given shift + """Decipher a message with the Caesar cipher of given shift >>> caesar_decipher('bcd', 1) 'abc' @@ -194,54 +194,82 @@ def caesar_decipher(message, shift): 'abc' >>> caesar_decipher('cd ez ab', 2) 'ab cx yz' + >>> caesar_decipher('Jgnnq Yqtnf!', 2) + 'Hello World!' """ return caesar_encipher(message, -shift) -def affine_encipher_letter(letter, multiplier=1, adder=0, one_based=True): +def affine_encipher_letter(accented_letter, multiplier=1, adder=0, one_based=True): """Encipher a letter, given a multiplier and adder - >>> ''.join([affine_encipher_letter(l, 3, 5, True) \ - for l in string.ascii_uppercase]) - 'HKNQTWZCFILORUXADGJMPSVYBE' - >>> ''.join([affine_encipher_letter(l, 3, 5, False) \ - for l in string.ascii_uppercase]) - 'FILORUXADGJMPSVYBEHKNQTWZC' + >>> cat(affine_encipher_letter(l, 3, 5, True) \ + for l in string.ascii_letters) + 'hknqtwzcfiloruxadgjmpsvybeHKNQTWZCFILORUXADGJMPSVYBE' + >>> cat(affine_encipher_letter(l, 3, 5, False) \ + for l in string.ascii_letters) + 'filoruxadgjmpsvybehknqtwzcFILORUXADGJMPSVYBEHKNQTWZC' """ + # letter = unaccent(accented_letter) + # if letter in string.ascii_letters: + # if letter in string.ascii_uppercase: + # alphabet_start = ord('A') + # else: + # alphabet_start = ord('a') + # letter_number = ord(letter) - alphabet_start + # if one_based: letter_number += 1 + # cipher_number = (letter_number * multiplier + adder) % 26 + # if one_based: cipher_number -= 1 + # return chr(cipher_number % 26 + alphabet_start) + # else: + # return letter + letter = unaccent(accented_letter) if letter in string.ascii_letters: - if letter in string.ascii_uppercase: - alphabet_start = ord('A') - else: - alphabet_start = ord('a') - letter_number = ord(letter) - alphabet_start + letter_number = pos(letter) if one_based: letter_number += 1 cipher_number = (letter_number * multiplier + adder) % 26 if one_based: cipher_number -= 1 - return chr(cipher_number % 26 + alphabet_start) + if letter in string.ascii_uppercase: + return unpos(cipher_number).upper() + else: + return unpos(cipher_number) else: return letter def affine_decipher_letter(letter, multiplier=1, adder=0, one_based=True): """Encipher a letter, given a multiplier and adder - >>> ''.join([affine_decipher_letter(l, 3, 5, True) \ - for l in 'HKNQTWZCFILORUXADGJMPSVYBE']) - 'ABCDEFGHIJKLMNOPQRSTUVWXYZ' - >>> ''.join([affine_decipher_letter(l, 3, 5, False) \ - for l in 'FILORUXADGJMPSVYBEHKNQTWZC']) - 'ABCDEFGHIJKLMNOPQRSTUVWXYZ' + >>> cat(affine_decipher_letter(l, 3, 5, True) \ + for l in 'hknqtwzcfiloruxadgjmpsvybeHKNQTWZCFILORUXADGJMPSVYBE') + 'abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ' + >>> cat(affine_decipher_letter(l, 3, 5, False) \ + for l in 'filoruxadgjmpsvybehknqtwzcFILORUXADGJMPSVYBEHKNQTWZC') + 'abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ' """ + # if letter in string.ascii_letters: + # if letter in string.ascii_uppercase: + # alphabet_start = ord('A') + # else: + # alphabet_start = ord('a') + # cipher_number = ord(letter) - alphabet_start + # if one_based: cipher_number += 1 + # plaintext_number = ( + # modular_division_table[multiplier] + # [(cipher_number - adder) % 26]) + # if one_based: plaintext_number -= 1 + # return chr(plaintext_number % 26 + alphabet_start) + # else: + # return letter if letter in string.ascii_letters: - if letter in string.ascii_uppercase: - alphabet_start = ord('A') - else: - alphabet_start = ord('a') - cipher_number = ord(letter) - alphabet_start + cipher_number = pos(letter) if one_based: cipher_number += 1 plaintext_number = ( modular_division_table[multiplier] - [(cipher_number - adder) % 26] ) + [(cipher_number - adder) % 26]) if one_based: plaintext_number -= 1 - return chr(plaintext_number % 26 + alphabet_start) + if letter in string.ascii_uppercase: + return unpos(plaintext_number).upper() + else: + return unpos(plaintext_number) else: return letter @@ -254,7 +282,7 @@ def affine_encipher(message, multiplier=1, adder=0, one_based=True): """ enciphered = [affine_encipher_letter(l, multiplier, adder, one_based) for l in message] - return ''.join(enciphered) + return cat(enciphered) def affine_decipher(message, multiplier=1, adder=0, one_based=True): """Decipher a message @@ -265,44 +293,47 @@ def affine_decipher(message, multiplier=1, adder=0, one_based=True): """ enciphered = [affine_decipher_letter(l, multiplier, adder, one_based) for l in message] - return ''.join(enciphered) + return cat(enciphered) + + +class KeywordWrapAlphabet(Enum): + from_a = 1 + from_last = 2 + from_largest = 3 -def keyword_cipher_alphabet_of(keyword, wrap_alphabet=0): +def keyword_cipher_alphabet_of(keyword, wrap_alphabet=KeywordWrapAlphabet.from_a): """Find the cipher alphabet given a keyword. wrap_alphabet controls how the rest of the alphabet is added after the keyword. - 0 : from 'a' - 1 : from the last letter in the sanitised keyword - 2 : from the largest letter in the sanitised keyword >>> keyword_cipher_alphabet_of('bayes') 'bayescdfghijklmnopqrtuvwxz' - >>> keyword_cipher_alphabet_of('bayes', 0) + >>> keyword_cipher_alphabet_of('bayes', KeywordWrapAlphabet.from_a) 'bayescdfghijklmnopqrtuvwxz' - >>> keyword_cipher_alphabet_of('bayes', 1) + >>> keyword_cipher_alphabet_of('bayes', KeywordWrapAlphabet.from_last) 'bayestuvwxzcdfghijklmnopqr' - >>> keyword_cipher_alphabet_of('bayes', 2) + >>> keyword_cipher_alphabet_of('bayes', KeywordWrapAlphabet.from_largest) 'bayeszcdfghijklmnopqrtuvwx' """ - if wrap_alphabet == 0: - cipher_alphabet = ''.join(deduplicate(sanitise(keyword) + + if wrap_alphabet == KeywordWrapAlphabet.from_a: + cipher_alphabet = cat(deduplicate(sanitise(keyword) + string.ascii_lowercase)) else: - if wrap_alphabet == 1: + if wrap_alphabet == KeywordWrapAlphabet.from_last: last_keyword_letter = deduplicate(sanitise(keyword))[-1] else: last_keyword_letter = sorted(sanitise(keyword))[-1] last_keyword_position = string.ascii_lowercase.find( last_keyword_letter) + 1 - cipher_alphabet = ''.join( + cipher_alphabet = cat( deduplicate(sanitise(keyword) + string.ascii_lowercase[last_keyword_position:] + string.ascii_lowercase)) return cipher_alphabet -def keyword_encipher(message, keyword, wrap_alphabet=0): +def keyword_encipher(message, keyword, wrap_alphabet=KeywordWrapAlphabet.from_a): """Enciphers a message with a keyword substitution cipher. wrap_alphabet controls how the rest of the alphabet is added after the keyword. @@ -312,18 +343,18 @@ def keyword_encipher(message, keyword, wrap_alphabet=0): >>> keyword_encipher('test message', 'bayes') 'rsqr ksqqbds' - >>> keyword_encipher('test message', 'bayes', 0) + >>> keyword_encipher('test message', 'bayes', KeywordWrapAlphabet.from_a) 'rsqr ksqqbds' - >>> keyword_encipher('test message', 'bayes', 1) + >>> keyword_encipher('test message', 'bayes', KeywordWrapAlphabet.from_last) 'lskl dskkbus' - >>> keyword_encipher('test message', 'bayes', 2) + >>> keyword_encipher('test message', 'bayes', KeywordWrapAlphabet.from_largest) 'qspq jsppbcs' """ cipher_alphabet = keyword_cipher_alphabet_of(keyword, wrap_alphabet) cipher_translation = ''.maketrans(string.ascii_lowercase, cipher_alphabet) - return message.lower().translate(cipher_translation) + return unaccent(message).lower().translate(cipher_translation) -def keyword_decipher(message, keyword, wrap_alphabet=0): +def keyword_decipher(message, keyword, wrap_alphabet=KeywordWrapAlphabet.from_a): """Deciphers a message with a keyword substitution cipher. wrap_alphabet controls how the rest of the alphabet is added after the keyword. @@ -333,57 +364,157 @@ def keyword_decipher(message, keyword, wrap_alphabet=0): >>> keyword_decipher('rsqr ksqqbds', 'bayes') 'test message' - >>> keyword_decipher('rsqr ksqqbds', 'bayes', 0) + >>> keyword_decipher('rsqr ksqqbds', 'bayes', KeywordWrapAlphabet.from_a) 'test message' - >>> keyword_decipher('lskl dskkbus', 'bayes', 1) + >>> keyword_decipher('lskl dskkbus', 'bayes', KeywordWrapAlphabet.from_last) 'test message' - >>> keyword_decipher('qspq jsppbcs', 'bayes', 2) + >>> keyword_decipher('qspq jsppbcs', 'bayes', KeywordWrapAlphabet.from_largest) 'test message' """ cipher_alphabet = keyword_cipher_alphabet_of(keyword, wrap_alphabet) cipher_translation = ''.maketrans(cipher_alphabet, string.ascii_lowercase) return message.lower().translate(cipher_translation) -def scytale_encipher(message, rows): - """Enciphers using the scytale transposition cipher. - Message is padded with spaces to allow all rows to be the same length. - >>> scytale_encipher('thequickbrownfox', 3) - 'tcnhkfeboqrxuo iw ' - >>> scytale_encipher('thequickbrownfox', 4) - 'tubnhirfecooqkwx' - >>> scytale_encipher('thequickbrownfox', 5) - 'tubn hirf ecoo qkwx ' - >>> scytale_encipher('thequickbrownfox', 6) - 'tqcrnxhukof eibwo ' - >>> scytale_encipher('thequickbrownfox', 7) - 'tqcrnx hukof eibwo ' +def vigenere_encipher(message, keyword): + """Vigenere encipher + + >>> vigenere_encipher('hello', 'abc') + 'hfnlp' """ - if len(message) % rows != 0: - message += ' '*(rows - len(message) % rows) - row_length = round(len(message) / rows) - slices = [message[i:i+row_length] - for i in range(0, len(message), row_length)] - return ''.join([''.join(r) for r in zip_longest(*slices, fillvalue='')]) + shifts = [pos(l) for l in sanitise(keyword)] + pairs = zip(message, cycle(shifts)) + return cat([caesar_encipher_letter(l, k) for l, k in pairs]) -def scytale_decipher(message, rows): - """Deciphers using the scytale transposition cipher. - Assumes the message is padded so that all rows are the same length. - - >>> scytale_decipher('tcnhkfeboqrxuo iw ', 3) - 'thequickbrownfox ' - >>> scytale_decipher('tubnhirfecooqkwx', 4) - 'thequickbrownfox' - >>> scytale_decipher('tubn hirf ecoo qkwx ', 5) - 'thequickbrownfox ' - >>> scytale_decipher('tqcrnxhukof eibwo ', 6) - 'thequickbrownfox ' - >>> scytale_decipher('tqcrnx hukof eibwo ', 7) - 'thequickbrownfox ' +def vigenere_decipher(message, keyword): + """Vigenere decipher + + >>> vigenere_decipher('hfnlp', 'abc') + 'hello' + """ + shifts = [pos(l) for l in sanitise(keyword)] + pairs = zip(message, cycle(shifts)) + return cat([caesar_decipher_letter(l, k) for l, k in pairs]) + + +def beaufort_encipher(message, keyword): + """Beaufort encipher + + >>> beaufort_encipher('inhisjournaldatedtheidesofoctober', 'arcanaimperii') + 'sevsvrusyrrxfayyxuteemazudmpjmmwr' + """ + shifts = [pos(l) for l in sanitise(keyword)] + pairs = zip(message, cycle(shifts)) + return cat([unpos(k - pos(l)) for l, k in pairs]) + +beaufort_decipher = beaufort_encipher + +beaufort_variant_encipher=vigenere_decipher +beaufort_variant_decipher=vigenere_encipher + + +def polybius_grid(keyword, column_order, row_order, letters_to_merge=None, + wrap_alphabet=KeywordWrapAlphabet.from_a): + """Grid for a Polybius cipher, using a keyword to rearrange the + alphabet. + + + >>> polybius_grid('a', 'abcde', 'abcde')['x'] == ('e', 'c') + True + >>> polybius_grid('elephant', 'abcde', 'abcde')['e'] == ('a', 'a') + True + >>> polybius_grid('elephant', 'abcde', 'abcde')['b'] == ('b', 'c') + True + """ + alphabet = keyword_cipher_alphabet_of(keyword, wrap_alphabet=wrap_alphabet) + if letters_to_merge is None: + letters_to_merge = {'j': 'i'} + grid = {l: k + for k, l in zip([(c, r) for c in column_order for r in row_order], + [l for l in alphabet if l not in letters_to_merge])} + for l in letters_to_merge: + grid[l] = grid[letters_to_merge[l]] + return grid + +def polybius_reverse_grid(keyword, column_order, row_order, letters_to_merge=None, + wrap_alphabet=KeywordWrapAlphabet.from_a): + """Grid for decrypting using a Polybius cipher, using a keyword to + rearrange the alphabet. + + >>> polybius_reverse_grid('a', 'abcde', 'abcde')['e', 'c'] == 'x' + True + >>> polybius_reverse_grid('elephant', 'abcde', 'abcde')['a', 'a'] == 'e' + True + >>> polybius_reverse_grid('elephant', 'abcde', 'abcde')['b', 'c'] == 'b' + True + """ + alphabet = keyword_cipher_alphabet_of(keyword, wrap_alphabet=wrap_alphabet) + if letters_to_merge is None: + letters_to_merge = {'j': 'i'} + grid = {k: l + for k, l in zip([(c, r) for c in column_order for r in row_order], + [l for l in alphabet if l not in letters_to_merge])} + return grid + + +def polybius_flatten(pair, column_first): + """Convert a series of pairs into a single list of characters""" + if column_first: + return str(pair[1]) + str(pair[0]) + else: + return str(pair[0]) + str(pair[1]) + +def polybius_encipher(message, keyword, column_order, row_order, + column_first=False, + letters_to_merge=None, wrap_alphabet=KeywordWrapAlphabet.from_a): + """Encipher a message with Polybius cipher, using a keyword to rearrange + the alphabet + + + >>> polybius_encipher('this is a test message for the ' \ + 'polybius decipherment', 'elephant', \ + [1, 2, 3, 4, 5], [1, 2, 3, 4, 5], \ + wrap_alphabet=KeywordWrapAlphabet.from_last) + '2214445544551522115522511155551543114252542214111352123234442355411135441314115451112122' + >>> polybius_encipher('this is a test message for the ' \ + 'polybius decipherment', 'elephant', 'abcde', 'abcde', \ + column_first=False) + 'bbadccddccddaebbaaddbbceaaddddaecbaacadadcbbadaaacdaabedbcccdeddbeaabdccacadaadcceaababb' + >>> polybius_encipher('this is a test message for the ' \ + 'polybius decipherment', 'elephant', 'abcde', 'abcde', \ + column_first=True) + 'bbdaccddccddeabbaaddbbecaaddddeabcaaacadcdbbdaaacaadbadecbccedddebaadbcccadaaacdecaaabbb' + """ + grid = polybius_grid(keyword, column_order, row_order, letters_to_merge, wrap_alphabet) + return cat(polybius_flatten(grid[l], column_first) + for l in message + if l in grid) + + +def polybius_decipher(message, keyword, column_order, row_order, + column_first=False, + letters_to_merge=None, wrap_alphabet=KeywordWrapAlphabet.from_a): + """Decipher a message with a Polybius cipher, using a keyword to rearrange + the alphabet + + >>> polybius_decipher('bbdaccddccddeabbaaddbbecaaddddeabcaaacadcdbbdaaaca'\ + 'adbadecbccedddebaadbcccadaaacdecaaabbb', 'elephant', 'abcde', 'abcde', \ + column_first=False) + 'toisisvtestxessvbephktoefhnugiysweqifoekxelt' + + >>> polybius_decipher('bbdaccddccddeabbaaddbbecaaddddeabcaaacadcdbbdaaaca'\ + 'adbadecbccedddebaadbcccadaaacdecaaabbb', 'elephant', 'abcde', 'abcde', \ + column_first=True) + 'thisisatestmessageforthepolybiusdecipherment' """ - cols = round(len(message) / rows) - columns = [message[i:i+rows] for i in range(0, cols * rows, rows)] - return ''.join([''.join(c) for c in zip_longest(*columns, fillvalue='')]) + grid = polybius_reverse_grid(keyword, column_order, row_order, letters_to_merge, wrap_alphabet) + column_index_type = type(column_order[0]) + row_index_type = type(row_order[0]) + if column_first: + pairs = [(column_index_type(p[1]), row_index_type(p[0])) for p in chunks(message, 2)] + else: + pairs = [(row_index_type(p[0]), column_index_type(p[1])) for p in chunks(message, 2)] + return cat(grid[p] for p in pairs if p in grid) def transpositions_of(keyword): @@ -424,72 +555,772 @@ def column_transposition_encipher(message, keyword, fillvalue=' ', emptycolumnwise=False): """Enciphers using the column transposition cipher. Message is padded to allow all rows to be the same length. + + >>> column_transposition_encipher('hellothere', 'abcdef', fillcolumnwise=True) + 'hlohr eltee ' + >>> column_transposition_encipher('hellothere', 'abcdef', fillcolumnwise=True, emptycolumnwise=True) + 'hellothere ' + >>> column_transposition_encipher('hellothere', 'abcdef') + 'hellothere ' + >>> column_transposition_encipher('hellothere', 'abcde') + 'hellothere' + >>> column_transposition_encipher('hellothere', 'abcde', fillcolumnwise=True, emptycolumnwise=True) + 'hellothere' + >>> column_transposition_encipher('hellothere', 'abcde', fillcolumnwise=True, emptycolumnwise=False) + 'hlohreltee' + >>> column_transposition_encipher('hellothere', 'abcde', fillcolumnwise=False, emptycolumnwise=True) + 'htehlelroe' + >>> column_transposition_encipher('hellothere', 'abcde', fillcolumnwise=False, emptycolumnwise=False) + 'hellothere' + >>> column_transposition_encipher('hellothere', 'clever', fillcolumnwise=True, emptycolumnwise=True) + 'heotllrehe' + >>> column_transposition_encipher('hellothere', 'clever', fillcolumnwise=True, emptycolumnwise=False) + 'holrhetlee' + >>> column_transposition_encipher('hellothere', 'clever', fillcolumnwise=False, emptycolumnwise=True) + 'htleehoelr' + >>> column_transposition_encipher('hellothere', 'clever', fillcolumnwise=False, emptycolumnwise=False) + 'hleolteher' + >>> column_transposition_encipher('hellothere', 'cleverly') + 'hleolthre e ' + >>> column_transposition_encipher('hellothere', 'cleverly', fillvalue='!') + 'hleolthre!e!' + >>> column_transposition_encipher('hellothere', 'cleverly', fillvalue=lambda: '*') + 'hleolthre*e*' """ - # >>> column_transposition_encipher('hellothere', 'clever') - # 'hleolteher' - # >>> column_transposition_encipher('hellothere', 'cleverly', fillvalue='!') - # 'hleolthre!e!' - # >>> column_transposition_encipher('hellothere', 'clever', columnwise=True) - # 'htleehoelr' - # """ transpositions = transpositions_of(keyword) message += pad(len(message), len(transpositions), fillvalue) if fillcolumnwise: - rows = every_nth(message, len(transpositions)) + rows = every_nth(message, len(message) // len(transpositions)) else: - rows = chunks(mesage, len(transpositions)) - columns = every_nth(message, len(transpositions), fillvalue=fillvalue) + rows = chunks(message, len(transpositions)) transposed = [transpose(r, transpositions) for r in rows] if emptycolumnwise: return combine_every_nth(transposed) else: - return ''.join(chain(*transposed)) + return cat(chain(*transposed)) def column_transposition_decipher(message, keyword, fillvalue=' ', - columnwise=False): + fillcolumnwise=False, + emptycolumnwise=False): """Deciphers using the column transposition cipher. Message is padded to allow all rows to be the same length. + + >>> column_transposition_decipher('hellothere', 'abcde', fillcolumnwise=True, emptycolumnwise=True) + 'hellothere' + >>> column_transposition_decipher('hlohreltee', 'abcde', fillcolumnwise=True, emptycolumnwise=False) + 'hellothere' + >>> column_transposition_decipher('htehlelroe', 'abcde', fillcolumnwise=False, emptycolumnwise=True) + 'hellothere' + >>> column_transposition_decipher('hellothere', 'abcde', fillcolumnwise=False, emptycolumnwise=False) + 'hellothere' + >>> column_transposition_decipher('heotllrehe', 'clever', fillcolumnwise=True, emptycolumnwise=True) + 'hellothere' + >>> column_transposition_decipher('holrhetlee', 'clever', fillcolumnwise=True, emptycolumnwise=False) + 'hellothere' + >>> column_transposition_decipher('htleehoelr', 'clever', fillcolumnwise=False, emptycolumnwise=True) + 'hellothere' + >>> column_transposition_decipher('hleolteher', 'clever', fillcolumnwise=False, emptycolumnwise=False) + 'hellothere' """ - # >>> column_transposition_decipher('hleolteher', 'clever') - # 'hellothere' - # >>> column_transposition_decipher('hleolthre!e!', 'cleverly', fillvalue='?') - # 'hellothere!!' - # >>> column_transposition_decipher('htleehoelr', 'clever', columnwise=True) - # 'hellothere' - # """ transpositions = transpositions_of(keyword) - if columnwise: - columns = chunks(message, int(len(message) / len(transpositions))) + message += pad(len(message), len(transpositions), fillvalue) + if emptycolumnwise: + rows = every_nth(message, len(message) // len(transpositions)) else: - columns = every_nth(message, len(transpositions), fillvalue=fillvalue) - untransposed_columns = untranspose(columns, transpositions) - return combine_every_nth(untransposed_columns) + rows = chunks(message, len(transpositions)) + untransposed = [untranspose(r, transpositions) for r in rows] + if fillcolumnwise: + return combine_every_nth(untransposed) + else: + return cat(chain(*untransposed)) +def scytale_encipher(message, rows, fillvalue=' '): + """Enciphers using the scytale transposition cipher. + Message is padded with spaces to allow all rows to be the same length. -def vigenere_encipher(message, keyword): - """Vigenere encipher + >>> scytale_encipher('thequickbrownfox', 3) + 'tcnhkfeboqrxuo iw ' + >>> scytale_encipher('thequickbrownfox', 4) + 'tubnhirfecooqkwx' + >>> scytale_encipher('thequickbrownfox', 5) + 'tubn hirf ecoo qkwx ' + >>> scytale_encipher('thequickbrownfox', 6) + 'tqcrnxhukof eibwo ' + >>> scytale_encipher('thequickbrownfox', 7) + 'tqcrnx hukof eibwo ' + """ + # transpositions = [i for i in range(math.ceil(len(message) / rows))] + # return column_transposition_encipher(message, transpositions, + # fillvalue=fillvalue, fillcolumnwise=False, emptycolumnwise=True) + transpositions = [i for i in range(rows)] + return column_transposition_encipher(message, transpositions, + fillvalue=fillvalue, fillcolumnwise=True, emptycolumnwise=False) - >>> vigenere_encipher('hello', 'abc') - 'hfnlp' +def scytale_decipher(message, rows): + """Deciphers using the scytale transposition cipher. + Assumes the message is padded so that all rows are the same length. + + >>> scytale_decipher('tcnhkfeboqrxuo iw ', 3) + 'thequickbrownfox ' + >>> scytale_decipher('tubnhirfecooqkwx', 4) + 'thequickbrownfox' + >>> scytale_decipher('tubn hirf ecoo qkwx ', 5) + 'thequickbrownfox ' + >>> scytale_decipher('tqcrnxhukof eibwo ', 6) + 'thequickbrownfox ' + >>> scytale_decipher('tqcrnx hukof eibwo ', 7) + 'thequickbrownfox ' """ - shifts = [ord(l) - ord('a') for l in sanitise(keyword)] - pairs = zip(message, cycle(shifts)) - return ''.join([caesar_encipher_letter(l, k) for l, k in pairs]) + # transpositions = [i for i in range(math.ceil(len(message) / rows))] + # return column_transposition_decipher(message, transpositions, + # fillcolumnwise=False, emptycolumnwise=True) + transpositions = [i for i in range(rows)] + return column_transposition_decipher(message, transpositions, + fillcolumnwise=True, emptycolumnwise=False) -def vigenere_decipher(message, keyword): - """Vigenere decipher - >>> vigenere_decipher('hfnlp', 'abc') - 'hello' +def railfence_encipher(message, height, fillvalue=''): + """Railfence cipher. + Works by splitting the text into sections, then reading across them to + generate the rows in the cipher. The rows are then combined to form the + ciphertext. + + Example: the plaintext "hellotherefriends", with a height of four, written + out in the railfence as + h h i + etere* + lorfns + l e d + (with the * showing the one character to finish the last section). + Each 'section' is two columns, but unfolded. In the example, the first + section is 'hellot'. + + >>> railfence_encipher('hellothereavastmeheartiesthisisalongpieceoftextfortestingrailfenceciphers', 2, fillvalue='!') + 'hlohraateerishsslnpeefetotsigaleccpeselteevsmhatetiiaogicotxfretnrifneihr!' + >>> railfence_encipher('hellothereavastmeheartiesthisisalongpieceoftextfortestingrailfenceciphers', 3, fillvalue='!') + 'horaersslpeeosglcpselteevsmhatetiiaogicotxfretnrifneihr!!lhateihsnefttiaece!' + >>> railfence_encipher('hellothereavastmeheartiesthisisalongpieceoftextfortestingrailfenceciphers', 5, fillvalue='!') + 'hresleogcseeemhetaocofrnrner!!lhateihsnefttiaece!!ltvsatiigitxetifih!!oarspeslp!' + >>> railfence_encipher('hellothereavastmeheartiesthisisalongpieceoftextfortestingrailfenceciphers', 10, fillvalue='!') + 'hepisehagitnr!!lernesge!!lmtocerh!!otiletap!!tseaorii!!hassfolc!!evtitffe!!rahsetec!!eixn!' + >>> railfence_encipher('hellothereavastmeheartiesthisisalongpieceoftextfortestingrailfenceciphers', 3) + 'horaersslpeeosglcpselteevsmhatetiiaogicotxfretnrifneihrlhateihsnefttiaece' + >>> railfence_encipher('hellothereavastmeheartiesthisisalongpieceoftextfortestingrailfenceciphers', 5) + 'hresleogcseeemhetaocofrnrnerlhateihsnefttiaeceltvsatiigitxetifihoarspeslp' + >>> railfence_encipher('hellothereavastmeheartiesthisisalongpieceoftextfortestingrailfenceciphers', 7) + 'haspolsevsetgifrifrlatihnettaeelemtiocxernhorersleesgcptehaiaottneihesfic' """ - shifts = [ord(l) - ord('a') for l in sanitise(keyword)] - pairs = zip(message, cycle(shifts)) - return ''.join([caesar_decipher_letter(l, k) for l, k in pairs]) + sections = chunks(message, (height - 1) * 2, fillvalue=fillvalue) + n_sections = len(sections) + # Add the top row + rows = [cat([s[0] for s in sections])] + # process the middle rows of the grid + for r in range(1, height-1): + rows += [cat([s[r:r+1] + s[height*2-r-2:height*2-r-1] for s in sections])] + # process the bottom row + rows += [cat([s[height - 1:height] for s in sections])] + # rows += [wcat([s[height - 1] for s in sections])] + return cat(rows) + +def railfence_decipher(message, height, fillvalue=''): + """Railfence decipher. + Works by reconstructing the grid used to generate the ciphertext, then + unfolding the sections so the text can be concatenated together. + + Example: given the ciphertext 'hhieterelorfnsled' and a height of 4, first + work out that the second row has a character missing, find the rows of the + grid, then split the section into its two columns. + + 'hhieterelorfnsled' is split into + h h i + etere + lorfns + l e d + (spaces added for clarity), which is stored in 'rows'. This is then split + into 'down_rows' and 'up_rows': + + down_rows: + hhi + eee + lrn + led + + up_rows: + tr + ofs + + These are then zipped together (after the up_rows are reversed) to recover + the plaintext. + + Most of the procedure is about finding the correct lengths for each row then + splitting the ciphertext into those rows. + + >>> railfence_decipher('hlohraateerishsslnpeefetotsigaleccpeselteevsmhatetiiaogicotxfretnrifneihr!', 2).strip('!') + 'hellothereavastmeheartiesthisisalongpieceoftextfortestingrailfenceciphers' + >>> railfence_decipher('horaersslpeeosglcpselteevsmhatetiiaogicotxfretnrifneihr!!lhateihsnefttiaece!', 3).strip('!') + 'hellothereavastmeheartiesthisisalongpieceoftextfortestingrailfenceciphers' + >>> railfence_decipher('hresleogcseeemhetaocofrnrner!!lhateihsnefttiaece!!ltvsatiigitxetifih!!oarspeslp!', 5).strip('!') + 'hellothereavastmeheartiesthisisalongpieceoftextfortestingrailfenceciphers' + >>> railfence_decipher('hepisehagitnr!!lernesge!!lmtocerh!!otiletap!!tseaorii!!hassfolc!!evtitffe!!rahsetec!!eixn!', 10).strip('!') + 'hellothereavastmeheartiesthisisalongpieceoftextfortestingrailfenceciphers' + >>> railfence_decipher('horaersslpeeosglcpselteevsmhatetiiaogicotxfretnrifneihrlhateihsnefttiaece', 3) + 'hellothereavastmeheartiesthisisalongpieceoftextfortestingrailfenceciphers' + >>> railfence_decipher('hresleogcseeemhetaocofrnrnerlhateihsnefttiaeceltvsatiigitxetifihoarspeslp', 5) + 'hellothereavastmeheartiesthisisalongpieceoftextfortestingrailfenceciphers' + >>> railfence_decipher('haspolsevsetgifrifrlatihnettaeelemtiocxernhorersleesgcptehaiaottneihesfic', 7) + 'hellothereavastmeheartiesthisisalongpieceoftextfortestingrailfenceciphers' + """ + # find the number and size of the sections, including how many characters + # are missing for a full grid + n_sections = math.ceil(len(message) / ((height - 1) * 2)) + padding_to_add = n_sections * (height - 1) * 2 - len(message) + # row_lengths are for the both up rows and down rows + row_lengths = [n_sections] * (height - 1) * 2 + for i in range((height - 1) * 2 - 1, (height - 1) * 2 - (padding_to_add + 1), -1): + row_lengths[i] -= 1 + # folded_rows are the combined row lengths in the middle of the railfence + folded_row_lengths = [row_lengths[0]] + for i in range(1, height-1): + folded_row_lengths += [row_lengths[i] + row_lengths[-i]] + folded_row_lengths += [row_lengths[height - 1]] + # find the rows that form the railfence grid + rows = [] + row_start = 0 + for i in folded_row_lengths: + rows += [message[row_start:row_start + i]] + row_start += i + # split the rows into the 'down_rows' (those that form the first column of + # a section) and the 'up_rows' (those that ofrm the second column of a + # section). + down_rows = [rows[0]] + up_rows = [] + for i in range(1, height-1): + down_rows += [cat([c for n, c in enumerate(rows[i]) if n % 2 == 0])] + up_rows += [cat([c for n, c in enumerate(rows[i]) if n % 2 == 1])] + down_rows += [rows[-1]] + up_rows.reverse() + return cat(c for r in zip_longest(*(down_rows + up_rows), fillvalue='') for c in r) + +def make_cadenus_keycolumn(doubled_letters = 'vw', start='a', reverse=False): + """Makes the key column for a Cadenus cipher (the column down between the + rows of letters) + + >>> make_cadenus_keycolumn()['a'] + 0 + >>> make_cadenus_keycolumn()['b'] + 1 + >>> make_cadenus_keycolumn()['c'] + 2 + >>> make_cadenus_keycolumn()['v'] + 21 + >>> make_cadenus_keycolumn()['w'] + 21 + >>> make_cadenus_keycolumn()['z'] + 24 + >>> make_cadenus_keycolumn(doubled_letters='ij', start='b', reverse=True)['a'] + 1 + >>> make_cadenus_keycolumn(doubled_letters='ij', start='b', reverse=True)['b'] + 0 + >>> make_cadenus_keycolumn(doubled_letters='ij', start='b', reverse=True)['c'] + 24 + >>> make_cadenus_keycolumn(doubled_letters='ij', start='b', reverse=True)['i'] + 18 + >>> make_cadenus_keycolumn(doubled_letters='ij', start='b', reverse=True)['j'] + 18 + >>> make_cadenus_keycolumn(doubled_letters='ij', start='b', reverse=True)['v'] + 6 + >>> make_cadenus_keycolumn(doubled_letters='ij', start='b', reverse=True)['z'] + 2 + """ + index_to_remove = string.ascii_lowercase.find(doubled_letters[0]) + short_alphabet = string.ascii_lowercase[:index_to_remove] + string.ascii_lowercase[index_to_remove+1:] + if reverse: + short_alphabet = cat(reversed(short_alphabet)) + start_pos = short_alphabet.find(start) + rotated_alphabet = short_alphabet[start_pos:] + short_alphabet[:start_pos] + keycolumn = {l: i for i, l in enumerate(rotated_alphabet)} + keycolumn[doubled_letters[0]] = keycolumn[doubled_letters[1]] + return keycolumn + +def cadenus_encipher(message, keyword, keycolumn, fillvalue='a'): + """Encipher with the Cadenus cipher + + >>> cadenus_encipher(sanitise('Whoever has made a voyage up the Hudson ' \ + 'must remember the Kaatskill mountains. ' \ + 'They are a dismembered branch of the great'), \ + 'wink', \ + make_cadenus_keycolumn(doubled_letters='vw', start='a', reverse=True)) + 'antodeleeeuhrsidrbhmhdrrhnimefmthgeaetakseomehetyaasuvoyegrastmmuuaeenabbtpchehtarorikswosmvaleatned' + >>> cadenus_encipher(sanitise('a severe limitation on the usefulness of ' \ + 'the cadenus is that every message must be ' \ + 'a multiple of twenty-five letters long'), \ + 'easy', \ + make_cadenus_keycolumn(doubled_letters='vw', start='a', reverse=True)) + 'systretomtattlusoatleeesfiyheasdfnmschbhneuvsnpmtofarenuseieeieltarlmentieetogevesitfaisltngeeuvowul' + """ + rows = chunks(message, len(message) // 25, fillvalue=fillvalue) + columns = zip(*rows) + rotated_columns = [col[start:] + col[:start] for start, col in zip([keycolumn[l] for l in keyword], columns)] + rotated_rows = zip(*rotated_columns) + transpositions = transpositions_of(keyword) + transposed = [transpose(r, transpositions) for r in rotated_rows] + return cat(chain(*transposed)) + +def cadenus_decipher(message, keyword, keycolumn, fillvalue='a'): + """ + >>> cadenus_decipher('antodeleeeuhrsidrbhmhdrrhnimefmthgeaetakseomehetyaa' \ + 'suvoyegrastmmuuaeenabbtpchehtarorikswosmvaleatned', \ + 'wink', \ + make_cadenus_keycolumn(reverse=True)) + 'whoeverhasmadeavoyageupthehudsonmustrememberthekaatskillmountainstheyareadismemberedbranchofthegreat' + >>> cadenus_decipher('systretomtattlusoatleeesfiyheasdfnmschbhneuvsnpmtof' \ + 'arenuseieeieltarlmentieetogevesitfaisltngeeuvowul', \ + 'easy', \ + make_cadenus_keycolumn(reverse=True)) + 'aseverelimitationontheusefulnessofthecadenusisthateverymessagemustbeamultipleoftwentyfiveletterslong' + """ + rows = chunks(message, len(message) // 25, fillvalue=fillvalue) + transpositions = transpositions_of(keyword) + untransposed_rows = [untranspose(r, transpositions) for r in rows] + columns = zip(*untransposed_rows) + rotated_columns = [col[-start:] + col[:-start] for start, col in zip([keycolumn[l] for l in keyword], columns)] + rotated_rows = zip(*rotated_columns) + # return rotated_columns + return cat(chain(*rotated_rows)) + + +def hill_encipher(matrix, message_letters, fillvalue='a'): + """Hill cipher + + >>> hill_encipher(np.matrix([[7,8], [11,11]]), 'hellothere') + 'drjiqzdrvx' + >>> hill_encipher(np.matrix([[6, 24, 1], [13, 16, 10], [20, 17, 15]]), \ + 'hello there') + 'tfjflpznvyac' + """ + n = len(matrix) + sanitised_message = sanitise(message_letters) + if len(sanitised_message) % n != 0: + padding = fillvalue[0] * (n - len(sanitised_message) % n) + else: + padding = '' + message = [pos(c) for c in sanitised_message + padding] + message_chunks = [message[i:i+n] for i in range(0, len(message), n)] + # message_chunks = chunks(message, len(matrix), fillvalue=None) + enciphered_chunks = [((matrix * np.matrix(c).T).T).tolist()[0] + for c in message_chunks] + return cat([unpos(round(l)) + for l in sum(enciphered_chunks, [])]) + +def hill_decipher(matrix, message, fillvalue='a'): + """Hill cipher + + >>> hill_decipher(np.matrix([[7,8], [11,11]]), 'drjiqzdrvx') + 'hellothere' + >>> hill_decipher(np.matrix([[6, 24, 1], [13, 16, 10], [20, 17, 15]]), \ + 'tfjflpznvyac') + 'hellothereaa' + """ + adjoint = linalg.det(matrix)*linalg.inv(matrix) + inverse_determinant = modular_division_table[int(round(linalg.det(matrix))) % 26][1] + inverse_matrix = (inverse_determinant * adjoint) % 26 + return hill_encipher(inverse_matrix, message, fillvalue) + + +# Where each piece of text ends up in the AMSCO transpositon cipher. +# 'index' shows where the slice appears in the plaintext, with the slice +# from 'start' to 'end' +AmscoSlice = collections.namedtuple('AmscoSlice', ['index', 'start', 'end']) + +class AmscoFillStyle(Enum): + continuous = 1 + same_each_row = 2 + reverse_each_row = 3 + +def amsco_transposition_positions(message, keyword, + fillpattern=(1, 2), + fillstyle=AmscoFillStyle.continuous, + fillcolumnwise=False, + emptycolumnwise=True): + """Creates the grid for the AMSCO transposition cipher. Each element in the + grid shows the index of that slice and the start and end positions of the + plaintext that go to make it up. + + >>> amsco_transposition_positions(string.ascii_lowercase, 'freddy', \ + fillpattern=(1, 2)) # doctest: +NORMALIZE_WHITESPACE + [[AmscoSlice(index=3, start=4, end=6), + AmscoSlice(index=2, start=3, end=4), + AmscoSlice(index=0, start=0, end=1), + AmscoSlice(index=1, start=1, end=3), + AmscoSlice(index=4, start=6, end=7)], + [AmscoSlice(index=8, start=12, end=13), + AmscoSlice(index=7, start=10, end=12), + AmscoSlice(index=5, start=7, end=9), + AmscoSlice(index=6, start=9, end=10), + AmscoSlice(index=9, start=13, end=15)], + [AmscoSlice(index=13, start=19, end=21), + AmscoSlice(index=12, start=18, end=19), + AmscoSlice(index=10, start=15, end=16), + AmscoSlice(index=11, start=16, end=18), + AmscoSlice(index=14, start=21, end=22)], + [AmscoSlice(index=18, start=27, end=28), + AmscoSlice(index=17, start=25, end=27), + AmscoSlice(index=15, start=22, end=24), + AmscoSlice(index=16, start=24, end=25), + AmscoSlice(index=19, start=28, end=30)]] + """ + transpositions = transpositions_of(keyword) + fill_iterator = cycle(fillpattern) + indices = count() + message_length = len(message) + + current_position = 0 + grid = [] + current_fillpattern = fillpattern + while current_position < message_length: + row = [] + if fillstyle == AmscoFillStyle.same_each_row: + fill_iterator = cycle(fillpattern) + if fillstyle == AmscoFillStyle.reverse_each_row: + fill_iterator = cycle(current_fillpattern) + for _ in range(len(transpositions)): + index = next(indices) + gap = next(fill_iterator) + row += [AmscoSlice(index, current_position, current_position + gap)] + current_position += gap + grid += [row] + if fillstyle == AmscoFillStyle.reverse_each_row: + current_fillpattern = list(reversed(current_fillpattern)) + return [transpose(r, transpositions) for r in grid] + +def amsco_transposition_encipher(message, keyword, + fillpattern=(1,2), fillstyle=AmscoFillStyle.reverse_each_row): + """AMSCO transposition encipher. + + >>> amsco_transposition_encipher('hellothere', 'abc', fillpattern=(1, 2)) + 'hoteelhler' + >>> amsco_transposition_encipher('hellothere', 'abc', fillpattern=(2, 1)) + 'hetelhelor' + >>> amsco_transposition_encipher('hellothere', 'acb', fillpattern=(1, 2)) + 'hotelerelh' + >>> amsco_transposition_encipher('hellothere', 'acb', fillpattern=(2, 1)) + 'hetelorlhe' + >>> amsco_transposition_encipher('hereissometexttoencipher', 'encode') + 'etecstthhomoerereenisxip' + >>> amsco_transposition_encipher('hereissometexttoencipher', 'cipher', fillpattern=(1, 2)) + 'hetcsoeisterereipexthomn' + >>> amsco_transposition_encipher('hereissometexttoencipher', 'cipher', fillpattern=(1, 2), fillstyle=AmscoFillStyle.continuous) + 'hecsoisttererteipexhomen' + >>> amsco_transposition_encipher('hereissometexttoencipher', 'cipher', fillpattern=(2, 1)) + 'heecisoosttrrtepeixhemen' + >>> amsco_transposition_encipher('hereissometexttoencipher', 'cipher', fillpattern=(1, 3, 2)) + 'hxtomephescieretoeisnter' + >>> amsco_transposition_encipher('hereissometexttoencipher', 'cipher', fillpattern=(1, 3, 2), fillstyle=AmscoFillStyle.continuous) + 'hxomeiphscerettoisenteer' + """ + grid = amsco_transposition_positions(message, keyword, + fillpattern=fillpattern, fillstyle=fillstyle) + ct_as_grid = [[message[s.start:s.end] for s in r] for r in grid] + return combine_every_nth(ct_as_grid) + + +def amsco_transposition_decipher(message, keyword, + fillpattern=(1,2), fillstyle=AmscoFillStyle.reverse_each_row): + """AMSCO transposition decipher + + >>> amsco_transposition_decipher('hoteelhler', 'abc', fillpattern=(1, 2)) + 'hellothere' + >>> amsco_transposition_decipher('hetelhelor', 'abc', fillpattern=(2, 1)) + 'hellothere' + >>> amsco_transposition_decipher('hotelerelh', 'acb', fillpattern=(1, 2)) + 'hellothere' + >>> amsco_transposition_decipher('hetelorlhe', 'acb', fillpattern=(2, 1)) + 'hellothere' + >>> amsco_transposition_decipher('etecstthhomoerereenisxip', 'encode') + 'hereissometexttoencipher' + >>> amsco_transposition_decipher('hetcsoeisterereipexthomn', 'cipher', fillpattern=(1, 2)) + 'hereissometexttoencipher' + >>> amsco_transposition_decipher('hecsoisttererteipexhomen', 'cipher', fillpattern=(1, 2), fillstyle=AmscoFillStyle.continuous) + 'hereissometexttoencipher' + >>> amsco_transposition_decipher('heecisoosttrrtepeixhemen', 'cipher', fillpattern=(2, 1)) + 'hereissometexttoencipher' + >>> amsco_transposition_decipher('hxtomephescieretoeisnter', 'cipher', fillpattern=(1, 3, 2)) + 'hereissometexttoencipher' + >>> amsco_transposition_decipher('hxomeiphscerettoisenteer', 'cipher', fillpattern=(1, 3, 2), fillstyle=AmscoFillStyle.continuous) + 'hereissometexttoencipher' + """ + + grid = amsco_transposition_positions(message, keyword, + fillpattern=fillpattern, fillstyle=fillstyle) + transposed_sections = [s for c in [l for l in zip(*grid)] for s in c] + plaintext_list = [''] * len(transposed_sections) + current_pos = 0 + for slice in transposed_sections: + plaintext_list[slice.index] = message[current_pos:current_pos-slice.start+slice.end][:len(message[slice.start:slice.end])] + current_pos += len(message[slice.start:slice.end]) + return cat(plaintext_list) + + +def bifid_grid(keyword, wrap_alphabet, letter_mapping): + """Create the grids for a Bifid cipher + """ + cipher_alphabet = keyword_cipher_alphabet_of(keyword, wrap_alphabet) + if letter_mapping is None: + letter_mapping = {'j': 'i'} + translation = ''.maketrans(letter_mapping) + cipher_alphabet = cat(collections.OrderedDict.fromkeys(cipher_alphabet.translate(translation))) + f_grid = {k: ((i // 5) + 1, (i % 5) + 1) + for i, k in enumerate(cipher_alphabet)} + r_grid = {((i // 5) + 1, (i % 5) + 1): k + for i, k in enumerate(cipher_alphabet)} + return translation, f_grid, r_grid + +def bifid_encipher(message, keyword, wrap_alphabet=KeywordWrapAlphabet.from_a, + letter_mapping=None, period=None, fillvalue=None): + """Bifid cipher + + >>> bifid_encipher("indiajelly", 'iguana') + 'ibidonhprm' + >>> bifid_encipher("indiacurry", 'iguana', period=4) + 'ibnhgaqltm' + >>> bifid_encipher("indiacurry", 'iguana', period=4, fillvalue='x') + 'ibnhgaqltzml' + """ + translation, f_grid, r_grid = bifid_grid(keyword, wrap_alphabet, letter_mapping) + + t_message = message.translate(translation) + pairs0 = [f_grid[l] for l in sanitise(t_message)] + if period: + chunked_pairs = [pairs0[i:i+period] for i in range(0, len(pairs0), period)] + if len(chunked_pairs[-1]) < period and fillvalue: + chunked_pairs[-1] += [f_grid[fillvalue]] * (period - len(chunked_pairs[-1])) + else: + chunked_pairs = [pairs0] + + pairs1 = [] + for c in chunked_pairs: + items = sum(list(list(i) for i in zip(*c)), []) + p = [(items[i], items[i+1]) for i in range(0, len(items), 2)] + pairs1 += p + + return cat(r_grid[p] for p in pairs1) + + +def bifid_decipher(message, keyword, wrap_alphabet=KeywordWrapAlphabet.from_a, + letter_mapping=None, period=None, fillvalue=None): + """Decipher with bifid cipher + + >>> bifid_decipher('ibidonhprm', 'iguana') + 'indiaielly' + >>> bifid_decipher("ibnhgaqltm", 'iguana', period=4) + 'indiacurry' + >>> bifid_decipher("ibnhgaqltzml", 'iguana', period=4) + 'indiacurryxx' + """ + translation, f_grid, r_grid = bifid_grid(keyword, wrap_alphabet, letter_mapping) + + t_message = message.translate(translation) + pairs0 = [f_grid[l] for l in sanitise(t_message)] + if period: + chunked_pairs = [pairs0[i:i+period] for i in range(0, len(pairs0), period)] + if len(chunked_pairs[-1]) < period and fillvalue: + chunked_pairs[-1] += [f_grid[fillvalue]] * (period - len(chunked_pairs[-1])) + else: + chunked_pairs = [pairs0] + + pairs1 = [] + for c in chunked_pairs: + items = [j for i in c for j in i] + gap = len(c) + p = [(items[i], items[i+gap]) for i in range(gap)] + pairs1 += p + + return cat(r_grid[p] for p in pairs1) + + +def autokey_encipher(message, keyword): + """Encipher with the autokey cipher + + >>> autokey_encipher('meetatthefountain', 'kilt') + 'wmpmmxxaeyhbryoca' + """ + shifts = [pos(l) for l in keyword + message] + pairs = zip(message, shifts) + return cat([caesar_encipher_letter(l, k) for l, k in pairs]) + +def autokey_decipher(ciphertext, keyword): + """Decipher with the autokey cipher + + >>> autokey_decipher('wmpmmxxaeyhbryoca', 'kilt') + 'meetatthefountain' + """ + plaintext = [] + keys = list(keyword) + for c in ciphertext: + plaintext_letter = caesar_decipher_letter(c, pos(keys[0])) + plaintext += [plaintext_letter] + keys = keys[1:] + [plaintext_letter] + return cat(plaintext) + + +class PocketEnigma(object): + """A pocket enigma machine + The wheel is internally represented as a 26-element list self.wheel_map, + where wheel_map[i] == j shows that the position i places on from the arrow + maps to the position j places on. + """ + def __init__(self, wheel=1, position='a'): + """initialise the pocket enigma, including which wheel to use and the + starting position of the wheel. + + The wheel is either 1 or 2 (the predefined wheels) or a list of letter + pairs. + + The position is the letter pointed to by the arrow on the wheel. + + >>> pe.wheel_map + [25, 4, 23, 10, 1, 7, 9, 5, 12, 6, 3, 17, 8, 14, 13, 21, 19, 11, 20, 16, 18, 15, 24, 2, 22, 0] + >>> pe.position + 0 + """ + self.wheel1 = [('a', 'z'), ('b', 'e'), ('c', 'x'), ('d', 'k'), + ('f', 'h'), ('g', 'j'), ('i', 'm'), ('l', 'r'), ('n', 'o'), + ('p', 'v'), ('q', 't'), ('s', 'u'), ('w', 'y')] + self.wheel2 = [('a', 'c'), ('b', 'd'), ('e', 'w'), ('f', 'i'), + ('g', 'p'), ('h', 'm'), ('j', 'k'), ('l', 'n'), ('o', 'q'), + ('r', 'z'), ('s', 'u'), ('t', 'v'), ('x', 'y')] + if wheel == 1: + self.make_wheel_map(self.wheel1) + elif wheel == 2: + self.make_wheel_map(self.wheel2) + else: + self.validate_wheel_spec(wheel) + self.make_wheel_map(wheel) + if position in string.ascii_lowercase: + self.position = pos(position) + else: + self.position = position + + def make_wheel_map(self, wheel_spec): + """Expands a wheel specification from a list of letter-letter pairs + into a full wheel_map. + + >>> pe.make_wheel_map(pe.wheel2) + [2, 3, 0, 1, 22, 8, 15, 12, 5, 10, 9, 13, 7, 11, 16, 6, 14, 25, 20, 21, 18, 19, 4, 24, 23, 17] + """ + self.validate_wheel_spec(wheel_spec) + self.wheel_map = [0] * 26 + for p in wheel_spec: + self.wheel_map[pos(p[0])] = pos(p[1]) + self.wheel_map[pos(p[1])] = pos(p[0]) + return self.wheel_map + + def validate_wheel_spec(self, wheel_spec): + """Validates that a wheel specificaiton will turn into a valid wheel + map. + + >>> pe.validate_wheel_spec([]) + Traceback (most recent call last): + ... + ValueError: Wheel specification has 0 pairs, requires 13 + >>> pe.validate_wheel_spec([('a', 'b', 'c')]*13) + Traceback (most recent call last): + ... + ValueError: Not all mappings in wheel specificationhave two elements + >>> pe.validate_wheel_spec([('a', 'b')]*13) + Traceback (most recent call last): + ... + ValueError: Wheel specification does not contain 26 letters + """ + if len(wheel_spec) != 13: + raise ValueError("Wheel specification has {} pairs, requires 13". + format(len(wheel_spec))) + for p in wheel_spec: + if len(p) != 2: + raise ValueError("Not all mappings in wheel specification" + "have two elements") + if len(set([p[0] for p in wheel_spec] + + [p[1] for p in wheel_spec])) != 26: + raise ValueError("Wheel specification does not contain 26 letters") + + def encipher_letter(self, letter): + """Enciphers a single letter, by advancing the wheel before looking up + the letter on the wheel. + + >>> pe.set_position('f') + 5 + >>> pe.encipher_letter('k') + 'h' + """ + self.advance() + return self.lookup(letter) + decipher_letter = encipher_letter + + def lookup(self, letter): + """Look up what a letter enciphers to, without turning the wheel. + + >>> pe.set_position('f') + 5 + >>> cat([pe.lookup(l) for l in string.ascii_lowercase]) + 'udhbfejcpgmokrliwntsayqzvx' + >>> pe.lookup('A') + '' + """ + if letter in string.ascii_lowercase: + return unpos( + (self.wheel_map[(pos(letter) - self.position) % 26] + + self.position)) + else: + return '' + + def advance(self): + """Advances the wheel one position. + + >>> pe.set_position('f') + 5 + >>> pe.advance() + 6 + """ + self.position = (self.position + 1) % 26 + return self.position + + def encipher(self, message, starting_position=None): + """Enciphers a whole message. + + >>> pe.set_position('f') + 5 + >>> pe.encipher('helloworld') + 'kjsglcjoqc' + >>> pe.set_position('f') + 5 + >>> pe.encipher('kjsglcjoqc') + 'helloworld' + >>> pe.encipher('helloworld', starting_position = 'x') + 'egrekthnnf' + """ + if starting_position: + self.set_position(starting_position) + transformed = '' + for l in message: + transformed += self.encipher_letter(l) + return transformed + decipher = encipher + + def set_position(self, position): + """Sets the position of the wheel, by specifying the letter the arrow + points to. -beaufort_encipher=vigenere_decipher -beaufort_decipher=vigenere_encipher + >>> pe.set_position('a') + 0 + >>> pe.set_position('m') + 12 + >>> pe.set_position('z') + 25 + """ + self.position = pos(position) + return self.position if __name__ == "__main__": import doctest - doctest.testmod() + doctest.testmod(extraglobs={'pe': PocketEnigma(1, 'a')})