# -*- coding: utf-8 -*-
# vim: autoindent shiftwidth=4 expandtab textwidth=120 tabstop=4 softtabstop=4
###############################################################################
# OpenLP - Open Source Lyrics Projection #
# --------------------------------------------------------------------------- #
# Copyright (c) 2008-2017 OpenLP Developers #
# --------------------------------------------------------------------------- #
# This program is free software; you can redistribute it and/or modify it #
# under the terms of the GNU General Public License as published by the Free #
# Software Foundation; version 2 of the License. #
# #
# This program is distributed in the hope that it will be useful, but WITHOUT #
# ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or #
# FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for #
# more details. #
# #
# You should have received a copy of the GNU General Public License along #
# with this program; if not, write to the Free Software Foundation, Inc., 59 #
# Temple Place, Suite 330, Boston, MA 02111-1307 USA #
###############################################################################
"""
The :mod:`songpro` module provides the functionality for importing SongPro
songs into the OpenLP database.
"""
import re
from openlp.plugins.songs.lib import strip_rtf
from openlp.plugins.songs.lib.importers.songimport import SongImport
[docs]class SongProImport(SongImport):
"""
The :class:`SongProImport` class provides the ability to import song files
from SongPro export files.
**SongPro Song File Format:**
| SongPro has the option to export under its File menu
| This produces files containing single or multiple songs
| The file is text with lines tagged with # followed by an identifier.
| This is documented here: http://creationsoftware.com/ImportIdentifiers.php
| An example here: http://creationsoftware.com/ExampleImportingManySongs.txt
|
| #A - next line is the Song Author
| #B - the lines following until next tagged line are the "Bridge" words
| (can be in rtf or plain text) which we map as B1
| #C - the lines following until next tagged line are the chorus words
| (can be in rtf or plain text)
| which we map as C1
| #D - the lines following until next tagged line are the "Ending" words
| (can be in rtf or plain text) which we map as E1
| #E - this song ends here, so we process the song -
| and start again at the next line
| #G - next line is the Group
| #M - next line is the Song Number
| #N - next line are Notes
| #R - next line is the SongCopyright
| #O - next line is the Verse Sequence
| #T - next line is the Song Title
| #1 - #7 the lines following until next tagged line are the verse x words
| (can be in rtf or plain text)
"""
def __init__(self, manager, **kwargs):
"""
Initialise the SongPro importer.
"""
super(SongProImport, self).__init__(manager, **kwargs)
[docs] def do_import(self):
"""
Receive a single file or a list of files to import.
"""
self.encoding = None
with open(self.import_source, 'rt', errors='ignore') as songs_file:
self.import_wizard.progress_bar.setMaximum(0)
tag = ''
text = ''
for file_line in songs_file:
if self.stop_import_flag:
break
file_text = file_line.rstrip()
if file_text and file_text[0] == '#':
self.process_section(tag, text.rstrip())
tag = file_text[1:]
text = ''
else:
text += file_line
self.finish()
[docs] def process_section(self, tag, text):
"""
Process a section of the song, i.e. title, verse etc.
"""
if tag == 'T':
self.set_defaults()
if text:
self.title = text
return
elif tag == 'E':
self.finish()
return
if 'rtf1' in text:
result = strip_rtf(text, self.encoding)
if result is None:
return
text, self.encoding = result
text = text.rstrip()
if not text:
return
if tag == 'A':
self.parse_author(text)
elif tag in ['B', 'C']:
self.add_verse(text, tag)
elif tag == 'D':
self.add_verse(text, 'E')
elif tag == 'G':
self.topics.append(text)
elif tag == 'M':
matches = re.findall(r'\d+', text)
if matches:
self.song_number = matches[-1]
self.song_book_name = text[:text.rfind(self.song_number)]
elif tag == 'N':
self.comments = text
elif tag == 'O':
for char in text:
if char == 'C':
self.verse_order_list.append('C1')
elif char == 'B':
self.verse_order_list.append('B1')
elif char == 'D':
self.verse_order_list.append('E1')
elif '1' <= char <= '7':
self.verse_order_list.append('V' + char)
elif tag == 'R':
self.add_copyright(text)
elif '1' <= tag <= '7':
self.add_verse(text, 'V' + tag[1:])