sedでmerge sortをしてみる

入力は空白区切りの小文字のみで構成される文字列のリスト．出力はソート結果．あなごるのWord frequency count (FIXED) 用に作ったのだが…　問題の入力に出てくる単語のソートだけで手元のマシンでも6秒弱かかる．

# input: word1 word2 ... wordn
# should be lower-case letters. no spaces except for the separator
 
 
s/ /\n/g
s/$/ /gm
 
# each line is a sorted list
 
:a
 
/\n/!q
 
s/^\(.*\)\n\(.*\)$/@\1: %\2/gm
 
# each line: @word11 word12 ... word1n : %word21 word22 ... word2m
# ':' is a separator of two lists
# '@' is the head of non-merged part of the first list
# '%' is the head of non-merged part of the second list
 
 
:b
 
# each line: sortedpart @word1i ... word1n : %word2j ... word2m
 
s/^[^%@:\n]*$/~&/gm                  # already done 
s/^\(.*\)@\(.*\): % *$/~\1\2/gm      # done (2nd list is consumed completely)
s/^\(.*\)@: \(.*\)%\(.*\)/~\1\2\3/gm # done (1st list is consumed completely)
 
/@/!be
h
 
s/^~.*$/~/gm                          # ignore merged lists
s/^.*@\(\w*\).*%\(\w*\).*$/\1 \2/gm  # prepare to comparation
 
 
#------------------- parallel comparation -----
 
# input: word11 word12\nword21 word22\n...\nwordn1 wordn2
# should be lower-case letters. no spaces except for the separator
# the entry '~' is ignored
 
# algorithm:
#   make pairs of corresponding characters in two words
#   compare all pairs at the same time
#   take the first non-equal result of he comparisons
 
 
s/^\| /&#/gm
s/^#~ *$/~/gm
 
:5
s/#\(\w\)\(.*\)#\(\w\)/\1\3 #\2#/mg
s/#\w.*#$/>/mg
s/# #..*/</mg
s/# #$/=/mg
t5
 
:1
 
:2
/a/b9
y/bcdefghijklmnopqrstuvwxyz/abcdefghijklmnopqrstuvwxy/
b2
:9
 
s/a[^a] /</g
s/[^a]a />/g
s/aa /=/g
 
/[^\n=<>~]/b1
s/^=*\(.\)/\1/gm
s/^\(.\).*/\1/gm
 
#----------- parallel comparation done ------
 
s/\n//g
s/$/#/
G
 
# [<>=]*#\nmergedpart @rest_of_list1 %rest_of_list2\n... 
 
# according to the results of comparations held in the head,
# move a head of either the rest of list1 or list2 to the tail of merged_part.
 
 
:m
s/^[<=]\(.*\)\n\(.*\)@\(\w*\) \(.*\)%\(\w*\)/\1-\2\3 @\4%\5/m
s/^>\(.*\)\n\(.*\)@\(\w*\)\(.*\)%\(\w*\) /\1-\2\5 @\3\4%/m
s/^~\([^\n]*\)\n~/\1-/
tm
s/-/\n/g
s/#\n//
bb
 
:e
s/~//g
ba

一文字目でバケットソートしてマージソートを各バケットに適用とかしないとダメっぽい．とにかく操作文字列長を小さくしないと速度が…