blob: b26f30b3a050a542df3e8e9e215efbba42864397 [file] [log] [blame]
#! /usr/bin/env python
"""Turn a pile of RCS log output into ChangeLog file entries.
"""
import sys
import string
import regex
import getopt
import time
def main():
args = sys.argv[1:]
opts, args = getopt.getopt(args, 'p:')
prefix = ''
for o, a in opts:
if p == '-p': prefix = a
f = sys.stdin
allrevs = []
while 1:
file = getnextfile(f)
if not file: break
revs = []
while 1:
rev = getnextrev(f, file)
if not rev:
break
revs.append(rev)
if revs:
allrevs[len(allrevs):] = revs
allrevs.sort()
allrevs.reverse()
for rev in allrevs:
formatrev(rev, prefix)
parsedateprog = regex.compile(
'^date: \([0-9]+\)/\([0-9]+\)/\([0-9]+\) ' +
'\([0-9]+\):\([0-9]+\):\([0-9]+\); author: \([^ ;]+\)')
authormap = {
'guido': 'Guido van Rossum <guido@cnri.reston.va.us>',
'jack': 'Jack Jansen <jack@cwi.nl>',
'sjoerd': 'Sjoerd Mullender <sjoerd@cwi.nl>',
}
def formatrev(rev, prefix):
dateline, file, revline, log = rev
if parsedateprog.match(dateline) >= 0:
fields = parsedateprog.group(1, 2, 3, 4, 5, 6)
author = parsedateprog.group(7)
if authormap.has_key(author): author = authormap[author]
tfields = map(string.atoi, fields) + [0, 0, 0]
tfields[5] = tfields[5] - time.timezone
t = time.mktime(tuple(tfields))
print time.ctime(t), '', author
words = string.split(log)
words[:0] = ['*', prefix + file + ':']
maxcol = 72-8
col = maxcol
for word in words:
if col > 0 and col + len(word) >= maxcol:
print
print '\t' + word,
col = -1
else:
print word,
col = col + 1 + len(word)
print
print
startprog = regex.compile("^Working file: \(.*\)$")
def getnextfile(f):
while 1:
line = f.readline()
if not line: return None
if startprog.match(line) >= 0:
file = startprog.group(1)
# Skip until first revision
while 1:
line = f.readline()
if not line: return None
if line[:10] == '='*10: return None
if line[:10] == '-'*10: break
## print "Skipped", line,
return file
## else:
## print "Ignored", line,
def getnextrev(f, file):
# This is called when we are positioned just after a '---' separator
revline = f.readline()
dateline = f.readline()
log = ''
while 1:
line = f.readline()
if not line: break
if line[:10] == '='*10:
# Ignore the *last* log entry for each file since it
# is the revision since which we are logging.
return None
if line[:10] == '-'*10: break
log = log + line
return dateline, file, revline, log
if __name__ == '__main__':
main()