-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 every_nth(text, n, fillvalue=''):
- """Returns n strings, each of which consists of every nth character,
- starting with the 0th, 1st, 2nd, ... (n-1)th character
-
- >>> every_nth(string.ascii_lowercase, 5)
- ['afkpuz', 'bglqv', 'chmrw', 'dinsx', 'ejoty']
- >>> every_nth(string.ascii_lowercase, 1)
- ['abcdefghijklmnopqrstuvwxyz']
- >>> every_nth(string.ascii_lowercase, 26) # doctest: +NORMALIZE_WHITESPACE
- ['a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j', 'k', 'l', 'm', 'n',
- 'o', 'p', 'q', 'r', 's', 't', 'u', 'v', 'w', 'x', 'y', 'z']
- >>> every_nth(string.ascii_lowercase, 5, fillvalue='!')
- ['afkpuz', 'bglqv!', 'chmrw!', 'dinsx!', 'ejoty!']
- """
- split_text = chunks(text, n, fillvalue)
- return [''.join(l) for l in zip_longest(*split_text, fillvalue=fillvalue)]
-
-def combine_every_nth(split_text):
- """Reforms a text split into every_nth strings
-
- >>> combine_every_nth(every_nth(string.ascii_lowercase, 5))
- 'abcdefghijklmnopqrstuvwxyz'
- >>> combine_every_nth(every_nth(string.ascii_lowercase, 1))
- 'abcdefghijklmnopqrstuvwxyz'
- >>> combine_every_nth(every_nth(string.ascii_lowercase, 26))
- 'abcdefghijklmnopqrstuvwxyz'
- """
- return ''.join([''.join(l)
- for l in zip_longest(*split_text, fillvalue='')])
-
-def chunks(text, n, fillvalue=None):
- """Split a text into chunks of n characters
-
- >>> chunks('abcdefghi', 3)
- ['abc', 'def', 'ghi']
- >>> chunks('abcdefghi', 4)
- ['abcd', 'efgh', 'i']
- >>> chunks('abcdefghi', 4, fillvalue='!')
- ['abcd', 'efgh', 'i!!!']
- """
- if fillvalue:
- padding = fillvalue[0] * (n - len(text) % n)
- else:
- padding = ''
- return [(text+padding)[i:i+n] for i in range(0, len(text), n)]
-
-def deduplicate(text):
- """If a string contains duplicate letters, remove all but the first. Retain
- the order of the letters.
-
- >>> deduplicate('cat')
- ['c', 'a', 't']
- >>> deduplicate('happy')
- ['h', 'a', 'p', 'y']
- >>> deduplicate('cattca')
- ['c', 'a', 't']
- """
- return list(collections.OrderedDict.fromkeys(text))
-
-