cpython/Lib/plat-irix5/cddb.py

207 lines
5.4 KiB
Python
Raw Normal View History

# This file implements a class which forms an interface to the .cddb
# directory that is maintained by SGI's cdman program.
#
# Usage is as follows:
#
# import readcd
# r = readcd.Readcd()
# c = Cddb(r.gettrackinfo())
#
# Now you can use c.artist, c.title and c.track[trackno] (where trackno
# starts at 1). When the CD is not recognized, all values will be the empty
# string.
# It is also possible to set the above mentioned variables to new values.
# You can then use c.write() to write out the changed values to the
# .cdplayerrc file.
1994-09-06 13:19:03 -03:00
import string, posix, os
_cddbrc = '.cddb'
_DB_ID_NTRACKS = 5
_dbid_map = '0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ@_=+abcdefghijklmnopqrstuvwxyz'
def _dbid(v):
if v >= len(_dbid_map):
return string.zfill(v, 2)
else:
return _dbid_map[v]
def tochash(toc):
if type(toc) == type(''):
tracklist = []
for i in range(2, len(toc), 4):
tracklist.append((None,
2001-02-09 12:45:10 -04:00
(int(toc[i:i+2]),
int(toc[i+2:i+4]))))
else:
tracklist = toc
ntracks = len(tracklist)
hash = _dbid((ntracks >> 4) & 0xF) + _dbid(ntracks & 0xF)
if ntracks <= _DB_ID_NTRACKS:
nidtracks = ntracks
else:
nidtracks = _DB_ID_NTRACKS - 1
min = 0
sec = 0
for track in tracklist:
start, length = track
min = min + length[0]
sec = sec + length[1]
min = min + sec / 60
sec = sec % 60
hash = hash + _dbid(min) + _dbid(sec)
for i in range(nidtracks):
start, length = tracklist[i]
hash = hash + _dbid(length[0]) + _dbid(length[1])
return hash
class Cddb:
def __init__(self, tracklist):
1994-09-06 13:19:03 -03:00
if os.environ.has_key('CDDB_PATH'):
path = os.environ['CDDB_PATH']
2001-02-09 12:45:10 -04:00
cddb_path = path.split(',')
else:
1994-09-06 13:19:03 -03:00
home = os.environ['HOME']
cddb_path = [home + '/' + _cddbrc]
1994-09-06 13:19:03 -03:00
self._get_id(tracklist)
for dir in cddb_path:
file = dir + '/' + self.id + '.rdb'
try:
f = open(file, 'r')
self.file = file
break
except IOError:
pass
2001-02-09 12:45:10 -04:00
ntracks = int(self.id[:2], 16)
1994-09-06 13:19:03 -03:00
self.artist = ''
self.title = ''
self.track = [None] + [''] * ntracks
self.trackartist = [None] + [''] * ntracks
self.notes = []
if not hasattr(self, 'file'):
return
import re
reg = re.compile(r'^([^.]*)\.([^:]*):[\t ]+(.*)')
while 1:
line = f.readline()
if not line:
break
match = reg.match(line)
if not match:
print 'syntax error in ' + file
continue
name1, name2, value = match.group(1, 2, 3)
if name1 == 'album':
if name2 == 'artist':
self.artist = value
elif name2 == 'title':
self.title = value
elif name2 == 'toc':
1994-09-06 13:19:03 -03:00
if not self.toc:
self.toc = value
if self.toc != value:
print 'toc\'s don\'t match'
elif name2 == 'notes':
self.notes.append(value)
elif name1[:5] == 'track':
1993-03-29 06:49:59 -04:00
try:
2001-02-09 12:45:10 -04:00
trackno = int(name1[5:])
1993-03-29 06:49:59 -04:00
except strings.atoi_error:
print 'syntax error in ' + file
continue
if trackno > ntracks:
print 'track number ' + `trackno` + \
1993-03-29 06:49:59 -04:00
' in file ' + file + \
' out of range'
continue
if name2 == 'title':
self.track[trackno] = value
elif name2 == 'artist':
self.trackartist[trackno] = value
f.close()
for i in range(2, len(self.track)):
track = self.track[i]
# if track title starts with `,', use initial part
# of previous track's title
1994-09-06 13:19:03 -03:00
if track and track[0] == ',':
try:
2001-02-09 12:45:10 -04:00
off = self.track[i - 1].index(',')
except ValueError:
pass
else:
self.track[i] = self.track[i-1][:off] \
+ track
1994-09-06 13:19:03 -03:00
def _get_id(self, tracklist):
# fill in self.id and self.toc.
# if the argument is a string ending in .rdb, the part
# upto the suffix is taken as the id.
if type(tracklist) == type(''):
if tracklist[-4:] == '.rdb':
self.id = tracklist[:-4]
self.toc = ''
return
t = []
for i in range(2, len(tracklist), 4):
t.append((None, \
2001-02-09 12:45:10 -04:00
(int(tracklist[i:i+2]), \
int(tracklist[i+2:i+4]))))
1994-09-06 13:19:03 -03:00
tracklist = t
ntracks = len(tracklist)
self.id = _dbid((ntracks >> 4) & 0xF) + _dbid(ntracks & 0xF)
if ntracks <= _DB_ID_NTRACKS:
nidtracks = ntracks
else:
nidtracks = _DB_ID_NTRACKS - 1
min = 0
sec = 0
for track in tracklist:
start, length = track
min = min + length[0]
sec = sec + length[1]
min = min + sec / 60
sec = sec % 60
self.id = self.id + _dbid(min) + _dbid(sec)
for i in range(nidtracks):
start, length = tracklist[i]
self.id = self.id + _dbid(length[0]) + _dbid(length[1])
self.toc = string.zfill(ntracks, 2)
for track in tracklist:
start, length = track
self.toc = self.toc + string.zfill(length[0], 2) + \
string.zfill(length[1], 2)
def write(self):
import posixpath
1994-09-06 13:19:03 -03:00
if os.environ.has_key('CDDB_WRITE_DIR'):
dir = os.environ['CDDB_WRITE_DIR']
else:
1994-09-06 13:19:03 -03:00
dir = os.environ['HOME'] + '/' + _cddbrc
file = dir + '/' + self.id + '.rdb'
if posixpath.exists(file):
# make backup copy
posix.rename(file, file + '~')
f = open(file, 'w')
f.write('album.title:\t' + self.title + '\n')
f.write('album.artist:\t' + self.artist + '\n')
f.write('album.toc:\t' + self.toc + '\n')
for note in self.notes:
f.write('album.notes:\t' + note + '\n')
prevpref = None
for i in range(1, len(self.track)):
1995-03-01 06:03:08 -04:00
if self.trackartist[i]:
f.write('track'+`i`+'.artist:\t'+self.trackartist[i]+'\n')
track = self.track[i]
try:
2001-02-09 12:45:10 -04:00
off = track.index(',')
except ValuError:
prevpref = None
else:
if prevpref and track[:off] == prevpref:
track = track[off:]
else:
prevpref = track[:off]
f.write('track' + `i` + '.title:\t' + track + '\n')
f.close()