summaryrefslogtreecommitdiffstats
path: root/Demo/pdist/cvslib.py
diff options
context:
space:
mode:
authorGuido van Rossum <guido@python.org>1995-04-26 22:57:11 (GMT)
committerGuido van Rossum <guido@python.org>1995-04-26 22:57:11 (GMT)
commit5f07b84796c207a0068f6ac5a0d6981d14801ff6 (patch)
treed82516c437ec3872d6fd344bd6556407bfd940e1 /Demo/pdist/cvslib.py
parent107c747009b3717840f4d5fae8f0d5987ae27f89 (diff)
downloadcpython-5f07b84796c207a0068f6ac5a0d6981d14801ff6.zip
cpython-5f07b84796c207a0068f6ac5a0d6981d14801ff6.tar.gz
cpython-5f07b84796c207a0068f6ac5a0d6981d14801ff6.tar.bz2
Initial revision
Diffstat (limited to 'Demo/pdist/cvslib.py')
-rwxr-xr-xDemo/pdist/cvslib.py186
1 files changed, 186 insertions, 0 deletions
diff --git a/Demo/pdist/cvslib.py b/Demo/pdist/cvslib.py
new file mode 100755
index 0000000..c98f0ba
--- /dev/null
+++ b/Demo/pdist/cvslib.py
@@ -0,0 +1,186 @@
+"""Utilities to read and write CVS admin files (esp. CVS/Entries)"""
+
+import string
+import os
+import time
+
+
+class Entry:
+
+ """Class representing one (parsed) line from CVS/Entries"""
+
+ def __init__(self, line):
+ words = string.splitfields(line, '/')
+ self.file = words[1]
+ self.rev = words[2]
+ dates = words[3] # ctime, mtime
+ if dates[:7] == 'Initial':
+ self.ctime = None
+ self.mtime = None
+ self.new = 1
+ else:
+ self.ctime = unctime(dates[:24])
+ self.mtime = unctime(dates[25:])
+ self.new = 0
+ self.extra = words[4]
+ self.sum = None
+
+ def unparse(self):
+ if self.new:
+ dates = "Initial %s" % self.file
+ else:
+ dates = gmctime(self.ctime) + ' ' + gmctime(self.mtime)
+ return "/%s/%s/%s/%s/\n" % (
+ self.file,
+ self.rev,
+ dates,
+ self.extra)
+
+ def setsum(self, sum):
+ self.sum = sum
+
+ def getsum(self):
+ return self.sum
+
+ def sethexsum(self, hexsum):
+ self.setsum(unhexify(hexsum))
+
+ def gethexsum(self):
+ if self.sum:
+ return hexify(self.sum)
+ else:
+ return None
+
+
+class CVS:
+
+ """Class representing the contents of CVS/Entries (and CVS/Sums)"""
+
+ def __init__(self):
+ self.readentries()
+
+ def readentries(self):
+ self.entries = {}
+ f = self.cvsopen("Entries")
+ while 1:
+ line = f.readline()
+ if not line: break
+ e = Entry(line)
+ self.entries[e.file] = e
+ f.close()
+
+ def readsums(self):
+ try:
+ f = self.cvsopen("Sums")
+ except IOError:
+ return
+ while 1:
+ line = f.readline()
+ if not line: break
+ words = string.split(line)
+ [file, rev, hexsum] = words
+ e = self.entries[file]
+ if e.rev == rev:
+ e.sethexsum(hexsum)
+ f.close()
+
+ def writeentries(self):
+ f = self.cvsopen("Entries", 'w')
+ for file in self.keys():
+ f.write(self.entries[file].unparse())
+ f.close()
+
+ def writesums(self):
+ if self.cvsexists("Sums"):
+ f = self.cvsopen("Sums", 'w')
+ else:
+ f = None
+ for file in self.keys():
+ e = self.entries[file]
+ hexsum = e.gethexsum()
+ if hexsum:
+ if not f:
+ f = self.cvsopen("Sums", 'w')
+ f.write("%s %s %s\n" % (file, e.rev, hexsum))
+ if f:
+ f.close()
+
+ def keys(self):
+ keys = self.entries.keys()
+ keys.sort()
+ return keys
+
+ def cvsexists(self, file):
+ file = os.path.join("CVS", file)
+ return os.path.exists(file)
+
+ def cvsopen(self, file, mode = 'r'):
+ file = os.path.join("CVS", file)
+ if 'r' not in mode:
+ self.backup(file)
+ return open(file, mode)
+
+ def backup(self, file):
+ if os.path.isfile(file):
+ bfile = file + '~'
+ os.rename(file, bfile)
+
+
+hexify_format = '%02x' * 16
+def hexify(sum):
+ "Return a hex representation of a 16-byte string (e.g. an MD5 digest)"
+ return hexify_format % tuple(map(ord, sum))
+
+def unhexify(hexsum):
+ "Return the original from a hexified string"
+ sum = ''
+ for i in range(0, len(hexsum), 2):
+ sum = sum + chr(string.atoi(hexsum[i:i+2], 16))
+ return sum
+
+
+unctime_monthmap = {}
+def unctime(date):
+ if not unctime_monthmap:
+ months = ['Jan', 'Feb', 'Mar', 'Apr', 'May', 'Jun',
+ 'Jul', 'Aug', 'Sep', 'Oct', 'Nov', 'Dec']
+ i = 0
+ for m in months:
+ i = i+1
+ unctime_monthmap[m] = i
+ words = string.split(date) # Day Mon DD HH:MM:SS YEAR
+ year = string.atoi(words[4])
+ month = unctime_monthmap[words[1]]
+ day = string.atoi(words[2])
+ [hh, mm, ss] = map(string.atoi, string.splitfields(words[3], ':'))
+ ss = ss - time.timezone
+ return time.mktime((year, month, day, hh, mm, ss, 0, 0, 0))
+
+def gmctime(t):
+ return time.asctime(time.gmtime(t))
+
+def test_unctime():
+ now = int(time.time())
+ t = time.gmtime(now)
+ at = time.asctime(t)
+ print 'GMT', now, at
+ print 'timezone', time.timezone
+ print 'local', time.ctime(now)
+ u = unctime(at)
+ print 'unctime()', u
+ gu = time.gmtime(u)
+ print '->', gu
+ print time.asctime(gu)
+
+def test():
+ x = CVS()
+ keys = x.entries.keys()
+ keys.sort()
+ for file in keys:
+ e = x.entries[file]
+ print file, e.rev, gmctime(e.ctime), gmctime(e.mtime), e.extra,
+ print e.gethexsum()
+
+
+if __name__ == "__main__":
+ test()