Server : Apache System : Linux host44.registrar-servers.com 4.18.0-513.18.1.lve.2.el8.x86_64 #1 SMP Sat Mar 30 15:36:11 UTC 2024 x86_64 User : vapecompany ( 2719) PHP Version : 7.4.33 Disable Function : NONE Directory : /lib64/python2.7/Tools/scripts/ |
Upload File : |
#! /usr/bin/python2.7 """Consolidate a bunch of CVS or RCS logs read from stdin. Input should be the output of a CVS or RCS logging command, e.g. cvs log -rrelease14: which dumps all log messages from release1.4 upwards (assuming that release 1.4 was tagged with tag 'release14'). Note the trailing colon! This collects all the revision records and outputs them sorted by date rather than by file, collapsing duplicate revision record, i.e., records with the same message for different files. The -t option causes it to truncate (discard) the last revision log entry; this is useful when using something like the above cvs log command, which shows the revisions including the given tag, while you probably want everything *since* that tag. The -r option reverses the output (oldest first; the default is oldest last). The -b tag option restricts the output to *only* checkin messages belonging to the given branch tag. The form -b HEAD restricts the output to checkin messages belonging to the CVS head (trunk). (It produces some output if tag is a non-branch tag, but this output is not very useful.) -h prints this message and exits. XXX This code was created by reverse engineering CVS 1.9 and RCS 5.7 from their output. """ import sys, errno, getopt, re sep1 = '='*77 + '\n' # file separator sep2 = '-'*28 + '\n' # revision separator def main(): """Main program""" truncate_last = 0 reverse = 0 branch = None opts, args = getopt.getopt(sys.argv[1:], "trb:h") for o, a in opts: if o == '-t': truncate_last = 1 elif o == '-r': reverse = 1 elif o == '-b': branch = a elif o == '-h': print __doc__ sys.exit(0) database = [] while 1: chunk = read_chunk(sys.stdin) if not chunk: break records = digest_chunk(chunk, branch) if truncate_last: del records[-1] database[len(database):] = records database.sort() if not reverse: database.reverse() format_output(database) def read_chunk(fp): """Read a chunk -- data for one file, ending with sep1. Split the chunk in parts separated by sep2. """ chunk = [] lines = [] while 1: line = fp.readline() if not line: break if line == sep1: if lines: chunk.append(lines) break if line == sep2: if lines: chunk.append(lines) lines = [] else: lines.append(line) return chunk def digest_chunk(chunk, branch=None): """Digest a chunk -- extract working file name and revisions""" lines = chunk[0] key = 'Working file:' keylen = len(key) for line in lines: if line[:keylen] == key: working_file = line[keylen:].strip() break else: working_file = None if branch is None: pass elif branch == "HEAD": branch = re.compile(r"^\d+\.\d+$") else: revisions = {} key = 'symbolic names:\n' found = 0 for line in lines: if line == key: found = 1 elif found: if line[0] in '\t ': tag, rev = line.split() if tag[-1] == ':': tag = tag[:-1] revisions[tag] = rev else: found = 0 rev = revisions.get(branch) branch = re.compile(r"^<>$") # <> to force a mismatch by default if rev: if rev.find('.0.') >= 0: rev = rev.replace('.0.', '.') branch = re.compile(r"^" + re.escape(rev) + r"\.\d+$") records = [] for lines in chunk[1:]: revline = lines[0] dateline = lines[1] text = lines[2:] words = dateline.split() author = None if len(words) >= 3 and words[0] == 'date:': dateword = words[1] timeword = words[2] if timeword[-1:] == ';': timeword = timeword[:-1] date = dateword + ' ' + timeword if len(words) >= 5 and words[3] == 'author:': author = words[4] if author[-1:] == ';': author = author[:-1] else: date = None text.insert(0, revline) words = revline.split() if len(words) >= 2 and words[0] == 'revision': rev = words[1] else: # No 'revision' line -- weird... rev = None text.insert(0, revline) if branch: if rev is None or not branch.match(rev): continue records.append((date, working_file, rev, author, text)) return records def format_output(database): prevtext = None prev = [] database.append((None, None, None, None, None)) # Sentinel for (date, working_file, rev, author, text) in database: if text != prevtext: if prev: print sep2, for (p_date, p_working_file, p_rev, p_author) in prev: print p_date, p_author, p_working_file, p_rev sys.stdout.writelines(prevtext) prev = [] prev.append((date, working_file, rev, author)) prevtext = text if __name__ == '__main__': try: main() except IOError, e: if e.errno != errno.EPIPE: raise