songbook [from: Songs data for LaTeX songbooks http:///www.patacrep.com]
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
 

277 lines
7.6 KiB

#!/usr/bin/python
# -*- coding: utf-8 -*-
import glob
import getopt, sys
import re, os
import logging
import locale
re.LOCALE
# the dictionary has target_word:replacement_word pairs
word_dic = {
##: oe inclusion
"coeur": "cœur",
"choeur": "chœur",
"boeuf": "bœuf",
"oeuvre": "œuvre",
"soeur": "sœur",
"noeud": "nœud",
"oeil": "œil",
"voeu": "vœu",
"oeuf": "œuf",
"oe{}": "œ",
"\\œ": "œ",
##: Contractions
"ptit": "p'tit",
"Y a": "Y'a",
"ptê": "p't'ê",
"p'": "p't'ê",
"p't-ê": "p't'ê",
##: Punctuation
"": "'",
"": "'",
"´": "'",
"Ca ": "Ça ",
"...": "{\\dots}",
"": "{\\dots}",
"say: ``":"say, ``",
"says: ``":"says, ``",
"said: ``":"said, ``",
#replace tabs with two spaces
" ": " ",
##: Typo
"New-York": "New York",
" i ": " I ",
"avant hier": "avant-hier",
##: Conversion to anglo-saxon chords
"Lam ": "Am ",
"La7": "A7",
"Lasus2": "Asus2",
"Sim ": "Bm ",
"Sim}": "Bm}",
"Sim]": "Bm]",
"Si7": "B7",
"Dom ": "Cm ",
"Do7": "C7",
"Do9": "C9",
"": "D ",
"Rém ": "Dm ",
"Rém]": "Dm]",
"Ré7": "D7",
"Ré#": "D#",
"Mim ": "Em ",
"Mim]": "Em]",
"Mim7": "Em7",
"Mim}": "Em}",
"Mi7": "E7",
"Mi7sus4": "E7sus4",
"Fa ": "F ",
"Fa}": "F}",
"[Fa\\": "[F\\",
"Fam ": "Fm ",
"Fa7": "F7",
"Sol ": "G ",
"Sol]": "G]",
"Solm ": "Gm ",
"Solm]": "Gm]",
"Sol7": "G7",
"/La": "/A",
"/Si": "/B",
"/Do": "/C",
"/Ré": "/D",
"/Mi": "/E",
"/Fa": "/F",
"/Sol": "/G",
"tab{La": "tab{A",
"tab{Si": "tab{B",
"tab{Do": "tab{C",
"tab{": "tab{D",
"tab{Mi": "tab{E",
"tab{Fa": "tab{F",
"tab{Sol": "tab{G",
"\\[La": "\\[A",
"\\[Si": "\\[B",
"\\[Do": "\\[C",
"\\[Ré": "\\[D",
"\\[Mi": "\\[E",
"\\[Fa": "\\[F",
"\\[Sol": "\\[G",
"\\[Re": "\\[D",
"b]": "&]",
"b7]": "&7]",
#C
"032010": "X32010",
#A
"002220": "X02220",
"002020": "X02020",
"002210": "X02210",
#D
"000232": "XX0232",
"X00232": "XX0232",
"000212": "XX0212",
"000231": "XX0231",
"X00231": "XX0231",
#B
"021202": "X21202",
## LaTeX
"beginchorus": "begin{chorus}",
"endchorus": "end{chorus}",
"beginverse*": "begin{verse*}",
"beginverse": "begin{verse}",
"endverse": "end{verse}",
"}[by=": "}\n [by=",
### end of rules
}
def usage():
print('''
Usage: rules.py [OPTION]
OPTIONS
-h, --help
display this help and exit
-f, --files=FILES
apply the set of rules on FILES
default is songs/*/*.sg
-l, --log=LEVEL
set the logging level to LEVEL
outputs logging information to rules.log
possible LEVEL values are : debug, info, warning, error and critical
''')
def replace_words(string):
'''
Search the data string for words defined in the dictionary and
replace them. This method avoids usual spelling and typos mistakes
when writing a song.
'''
logging.info("replace_words: search and replace words from dictionary into song data")
for search, replace in word_dic.items():
string = string.replace(search, replace)
return string
#language based typographical rules
def language_rules(string):
'''
Search the data string for common typographical mistakes.
Implemented rules depend on the current song language that is
defined by babel for every .sg file through the macro
\selectlanguage{<lang>}
'''
logging.info("language_rules: looking for common typographical mistakes")
if (re.compile("selectlanguage{french}").search(string)):
logging.info(" song language is set to : french")
#ensure non-breaking spaces before symbols ? ! ; :
string = re.sub("(?P<last_char>\S)(?P<symbol>[!?;:])","\g<last_char> \g<symbol>", string)
#... except for gtabs macros with capos
string = re.sub("(?P<gtab>tab.?{.*)\s:","\g<gtab>:", string)
#... and for urls
string = re.sub("http\s:","http:", string)
#and apply a second time for cases like \gtab{Gm}{10:X02210:}
string = re.sub("(?P<gtab>tab.?{.*)\s:","\g<gtab>:", string)
#ensure no spaces after symbols (
string = re.sub("(?P<symbol>[\(])\s(?P<next_char>\S)","\g<symbol>\g<next_char>", string)
#convert inverted commas
string = re.sub("``","{\\og}", string)
string = re.sub("''","{\\\\fg}", string)
elif (re.compile("selectlanguage{english}").search(string)):
logging.info(" song language is set to : english")
#ensure no spaces before symbols ? ! ; : )
string = re.sub("(?P<last_char>\S)\s(?P<symbol>[!?;:\)])","\g<last_char>\g<symbol>", string)
#ensure no spaces after symbols (
string = re.sub("(?P<symbol>[\(])\s(?P<next_char>\S)","\g<symbol>\g<next_char>", string)
elif (re.compile("selectlanguage{spanish}").search(string)):
logging.info(" song language is set to : spanish")
#ensure no spaces before symbols ? ! ; : )
string = re.sub("(?P<last_char>\S)\s(?P<symbol>[!?;:\)])","\g<last_char>\g<symbol>", string)
#ensure no spaces after symbols ¿ ¡ (
string = re.sub("(?P<symbol>[¿¡\(])\s(?P<next_char>\S)","\g<symbol>\g<next_char>", string)
elif (re.compile("selectlanguage{portuguese}").search(string)):
logging.info(" song language is set to : portuguese")
#convert inverted commas
string = re.sub("``","{\\og}", string)
string = re.sub("''","{\\\\fg}", string)
else :
raise NotImplementedError('No supported language found')
return string
def process_lines(lines):
'''
Removes trailing punctuation and multi-spaces from lines. Note
that it preserves whitespaces at the beginning of lines that
correspond to indentation.
'''
logging.info("process_lines: handling song data line by line")
star = False
for index, line in enumerate(lines):
if re.compile("begin{verse\*}").search(line):
star = True
if re.compile("end{verse\*}").search(line):
star = False
if star == True and re.compile("end{verse}").search(line):
line = line.replace("verse", "verse*")
star = False
#remove trailing spaces and punctuation
line = line.rstrip().rstrip(',.;').rstrip()
#remove multi-spaces within lines
line = re.sub("(?P<last_char>\S)\s{2,}","\g<last_char> ", line)
lines[index] = line
return lines
def main():
locale.setlocale(locale.LC_ALL, '')
try:
opts, args = getopt.getopt(sys.argv[1:],
"hf:l:",
["help", "files=", "log="])
except getopt.GetoptError:
usage()
sys.exit(2)
library = './'
loglevel = "warning"
for option, arg in opts:
if option in ("-h", "--help"):
usage()
sys.exit()
elif option in ("-f", "--files"):
songfiles = glob.glob(arg)
elif option in ("-l", "--log"):
loglevel = arg
numeric_level = getattr(logging, loglevel.upper(), None)
if not isinstance(numeric_level, int):
raise ValueError('Invalid log level: %s' % loglevel)
logging.basicConfig(level=numeric_level, filename='rules.log', filemode='w')
else:
assert False, "unhandled option"
songfiles = glob.iglob(os.path.join(library, 'songs', '**', '*.sg'), recursive=True)
for filename in songfiles:
with open(filename, 'r+') as songfile:
logging.info("checking file: "+filename)
data = songfile.read()
#no dots for acronyms
#data = re.sub("(?P<capital_letter>[A-Z])\.","\g<capital_letter>", data)
data = replace_words(data)
try:
data = language_rules(data)
except NotImplementedError as e:
logging.info(filename + ": " + str(e))
lines = process_lines(data.split('\n'))
data = "\n".join(lines)
songfile.seek(0)
songfile.write(data)
songfile.truncate()
if __name__ == '__main__':
main()