mirror of
https://git.lyx.org/repos/lyx.git
synced 2024-12-23 05:25:26 +00:00
8c8e19bb1c
gnuhtml2latex does not handle encodings at all. Therefore the result is not imported correctly by tex2lyx if the HTML file is encoded in anything else than ascii or latin1 (the default of tex2lyx). The simple wrapper script loads inputenc if needed. It may not be possible to compile the result with LaTeX, (e.gif utf8 is used), but for running tex2lyx it will work just fine.
128 lines
4.5 KiB
Python
128 lines
4.5 KiB
Python
#! /usr/bin/env python
|
|
# -*- coding: utf-8 -*-
|
|
|
|
# file html2latexwrapper.py
|
|
# This file is part of LyX, the document processor.
|
|
# Licence details can be found in the file COPYING.
|
|
|
|
# author Georg Baum
|
|
|
|
# Full author contact details are available in file CREDITS
|
|
|
|
# Usage:
|
|
# html2latexwrapper.py <converter> <from file> <to file>
|
|
|
|
# This script will call <converter> -s <from file> > <to file>
|
|
# and add a \usepackage{inputenc} line if needed.
|
|
|
|
|
|
import os, string, sys, re
|
|
|
|
from lyxpreview_tools import error, run_command
|
|
|
|
|
|
def usage(prog_name):
|
|
return "Usage: %s <converter> <from file> <to file>" % prog_name
|
|
|
|
|
|
def get_encoding(from_file_name):
|
|
'''Read the encoding from a HTML or XHTML file'''
|
|
try:
|
|
from_file = open(from_file_name, 'rt')
|
|
regexpxml = re.compile(r'^\s?<\?xml\s+.*?encoding\s*=\s*"([^"]+)"', re.IGNORECASE)
|
|
regexptype = re.compile(r'^\s?<meta\s+.*?charset\s*=\s*"([^"]+)"', re.IGNORECASE)
|
|
for line in from_file.readlines():
|
|
m = regexpxml.match(line)
|
|
if not m:
|
|
m = regexptype.match(line)
|
|
if m:
|
|
from_file.close()
|
|
return m.group(1).lower()
|
|
from_file.close()
|
|
except:
|
|
pass
|
|
return ''
|
|
|
|
|
|
def main(argv):
|
|
# Parse and manipulate the command line arguments.
|
|
if len(argv) != 4:
|
|
error(usage(argv[0]))
|
|
|
|
converter = argv[1]
|
|
from_file_name = argv[2]
|
|
to_file_name = argv[3]
|
|
|
|
# Run gnuhtml2latex
|
|
cmd = '%s -s %s' % (converter, from_file_name)
|
|
(ret, output) = run_command(cmd, False)
|
|
|
|
# Determine encoding of HTML file
|
|
enc = get_encoding(from_file_name).replace('iso_8859', 'iso-8859')
|
|
# The HTML encodings were taken from http://www.iana.org/assignments/character-sets/character-sets.xml.
|
|
# Only those with inputenc support were added, and only thge most important aliases.
|
|
# List of encodings that have the same name in HTML (may be as an alias) and inputenc
|
|
same_enc = ['cp437', 'cp850', 'cp852', 'cp855', 'cp858', 'cp862', 'cp865', 'cp866', \
|
|
'cp1250', 'cp1251', 'cp1252', 'cp1255', 'cp1256', 'cp1257', \
|
|
'koi8-r', 'koi8-u', 'pt154', 'pt254', \
|
|
'latin1', 'latin2', 'latin3', 'latin4', 'latin5', 'latin9', 'latin10']
|
|
# Translation table from HTML encoding names to inputenc encoding names
|
|
encodings = {'utf-8' : 'utf8', 'csutf8' : 'utf8', \
|
|
'iso-8859-1' : 'latin1', 'cp819' : 'latin1', \
|
|
'iso-8859-2' : 'latin2', \
|
|
'iso-8859-3' : 'latin3', \
|
|
'iso-8859-4' : 'latin4', \
|
|
'iso-8859-5' : 'iso88595', 'cyrillic' : 'iso88595', \
|
|
'iso-8859-6' : '8859-6', 'arabic' : '8859-6', \
|
|
'iso-8859-7' : 'iso-8859-7', 'greek' : 'iso-8859-7', \
|
|
'iso-8859-8' : '8859-8', 'hebrew' : '8859-8', \
|
|
'iso-8859-9' : 'latin5', \
|
|
'iso-8859-13' : 'l7xenc', \
|
|
'iso-8859-15' : 'latin9', \
|
|
'iso-8859-16' : 'latin10', \
|
|
'ibm437' : 'cp437', \
|
|
'ibm850' : 'cp850', \
|
|
'ibm852' : 'cp852', \
|
|
'ibm855' : 'cp855', \
|
|
'ibm858' : 'cp858', \
|
|
'ibm862' : 'cp862', \
|
|
'ibm865' : 'cp865', \
|
|
'ibm866' : 'cp866', \
|
|
'ibm1250' : 'cp1250', \
|
|
'ibm1251' : 'cp1251', \
|
|
'ibm1255' : 'cp1255', \
|
|
'ibm1256' : 'cp1256', \
|
|
'ibm1257' : 'cp1257', \
|
|
'macintosh' : 'applemac', 'mac' : 'applemac', 'csmacintosh' : 'applemac'}
|
|
if enc != '':
|
|
if enc in encodings.keys():
|
|
enc = encodings[enc]
|
|
elif enc not in same_enc:
|
|
enc = ''
|
|
|
|
# Read conversion result
|
|
lines = output.split('\n')
|
|
|
|
# Do not add the inputenc call if inputenc or CJK is already loaded
|
|
add_inputenc = (enc != '')
|
|
if add_inputenc:
|
|
regexp = re.compile(r'^\s?\\usepackage\s?(\[[^]+]\])?\s?{(inputenc)|(CJK)|(CJKutf8)}')
|
|
for line in lines:
|
|
if regexp.match(line):
|
|
add_inputenc = False
|
|
break
|
|
|
|
# Write output file and insert inputenc call if needed
|
|
to_file = open(to_file_name, 'wt')
|
|
for line in lines:
|
|
to_file.write(line + '\n')
|
|
if add_inputenc and line.find('\\documentclass') == 0:
|
|
to_file.write('\\usepackage[%s]{inputenc}\n' % enc)
|
|
to_file.close()
|
|
|
|
return ret
|
|
|
|
|
|
if __name__ == "__main__":
|
|
main(sys.argv)
|