git-bz

#!/usr/bin/python
#
# git-bz - git subcommand to integrate with bugzilla
#
# Copyright (C) 2008  Owen Taylor
#
# 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; either version 2
# of the License, or (at your option) any later version.
#
# 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, If not, see
# http://www.gnu.org/licenses/.
#
# Patches for git-bz
# ==================
# Send to Owen Taylor <otaylor@fishsoup.net>
#
# Installation
# ============
# Copy or symlink somewhere in your path.
#
# Documentation
# =============
# See http://git.fishsoup.net/man/git-bz.html
# (generated from git-bz.txt in this directory.)
#
DEFAULT_CONFIG = \
"""
default-assigned-to =
default-op-sys = All
default-platform = All
default-version = unspecified
"""

CONFIG = {}

CONFIG['bugs.freedesktop.org'] = \
"""
https = true
default-priority = medium
"""

CONFIG['bugzilla.gnome.org'] = \
"""
https = true
default-priority = Normal
"""

CONFIG['bugzilla.mozilla.org'] = \
"""
https = true
default-priority = --
"""

################################################################################

import base64
import cPickle as pickle
from ConfigParser import RawConfigParser, NoOptionError
import httplib
from optparse import OptionParser
import os
try:
    from sqlite3 import dbapi2 as sqlite
except ImportError:
    from pysqlite2 import dbapi2 as sqlite
import re
from StringIO import StringIO
from subprocess import Popen, CalledProcessError, PIPE
import shutil
import sys
import tempfile
import time
import traceback
import xmlrpclib
import urlparse
from xml.etree.cElementTree import ElementTree
import base64
import platform

# Globals
# =======

# options dictionary from optparse
global_options = None

# Utility functions for git
# =========================

# Run a git command
#    Non-keyword arguments are passed verbatim as command line arguments
#    Keyword arguments are turned into command line options
#       <name>=True => --<name>
#       <name>='<str>' => --<name>=<str>
#    Special keyword arguments:
#       _quiet: Discard all output even if an error occurs
#       _interactive: Don't capture stdout and stderr
#       _input=<str>: Feed <str> to stdinin of the command
#       _return_error: Return tuple of captured (stdout,stderr)
#
def git_run(command, *args, **kwargs):
    to_run = ['git', command.replace("_", "-")]

    interactive = False
    quiet = False
    input = None
    return_stderr = False
    for (k,v) in kwargs.iteritems():
        if k == '_quiet':
            quiet = True
        elif k == '_interactive':
            interactive = True
        elif k == '_return_stderr':
            return_stderr = True
        elif k == '_input':
            input = v
        elif v is True:
            if len(k) == 1:
                to_run.append("-" + k)
            else:
                to_run.append("--" + k.replace("_", "-"))
        else:
            to_run.append("--" + k.replace("_", "-") + "=" + v)

    to_run.extend(args)

    process = Popen(to_run,
                    stdout=(None if interactive else PIPE),
                    stderr=(None if interactive else PIPE),
                    stdin=(PIPE if (input != None) else None))
    output, error = process.communicate(input)
    if process.returncode != 0:
        if not quiet and not interactive:
            # Using print here could result in Python adding a stray space
            # before the next print
            sys.stderr.write(error)
            sys.stdout.write(output)
        raise CalledProcessError(process.returncode, " ".join(to_run))

    if interactive:
        return None
    elif return_stderr:
        return output.strip(), error.strip()
    else:
        return output.strip()

# Wrapper to allow us to do git.<command>(...) instead of git_run()
class Git:
    def __getattr__(self, command):
        def f(*args, **kwargs):
            return git_run(command, *args, **kwargs)
        return f

git = Git()

class GitCommit:
    def __init__(self, id, subject):
        self.id = id
        self.subject = subject

def rev_list_commits(*args, **kwargs):
    kwargs_copy = dict(kwargs)
    kwargs_copy['pretty'] = 'format:%s'
    output = git.rev_list(*args, **kwargs_copy)
    if output == "":
        lines = []
    else:
        lines = output.split("\n")
    if (len(lines) % 2 != 0):
        raise RuntimeException("git rev-list didn't return an even number of lines")

    result = []
    for i in xrange(0, len(lines), 2):
        m = re.match("commit\s+([A-Fa-f0-9]+)", lines[i])
        if not m:
            raise RuntimeException("Can't parse commit it '%s'", lines[i])
        commit_id = m.group(1)
        subject = lines[i + 1]
        result.append(GitCommit(commit_id, subject))

    return result

def get_commits(commit_or_revision_range):
    # We take specifying a single revision to mean everything since that
    # revision, while git-rev-list lists that revision and all ancestors
    try:
        # See if the argument identifies a single revision
        rev = git.rev_parse(commit_or_revision_range, verify=True, _quiet=True)
        commits = rev_list_commits(rev, max_count='1')
    except CalledProcessError:
        # If not, assume the argument is a range
        commits = rev_list_commits(commit_or_revision_range)

    if len(commits) == 0:
        die("'%s' does not name any commits. Use HEAD to specify just the last commit" %
            commit_or_revision_range)

    return commits

def get_patch(commit):
    # We could pass through -M as an option, but I think you basically always
    # want it; showing renames as renames rather than removes/adds greatly
    # improves readability.
    return git.diff(commit.id + "^.." + commit.id, M=True, binary=True, unified="8") + "\n"

def get_body(commit):
    return git.log(commit.id + "^.." + commit.id, pretty="format:%b")

def commit_is_merge(commit):
    contents = git.cat_file("commit", commit.id)
    parent_count = 0
    for line in contents.split("\n"):
        if line == "":
            break
        if line.startswith("parent "):
            parent_count += 1

    return parent_count > 1

# Global configuration variables
# ==============================

def get_browser():
    try:
        return git.config('bz.browser', get=True)
    except CalledProcessError:
        return 'firefox3'

def get_tracker():
    if global_options.bugzilla != None:
        return global_options.bugzilla

    try:
        return git.config('bz.default-tracker', get=True)
    except CalledProcessError:
        return 'bugzilla.gnome.org'

def get_default_product():
    try:
        return git.config('bz.default-product', get=True)
    except CalledProcessError:
        return None

def get_default_component():
    try:
        return git.config('bz.default-component', get=True)
    except CalledProcessError:
        return None

def get_add_url():
    try:
        return git.config('bz.add-url', get=True) == 'true'
    except CalledProcessError:
        return True

def get_add_url_method():
    try:
        return git.config('bz.add-url-method', get=True)
    except CalledProcessError:
        return "body-append:%u"

def get_firefox_profile_pref():
    try:
        return git.config('bz.firefox-profile', get=True)
    except CalledProcessError:
        return ""

# Per-tracker configuration variables
# ===================================

def resolve_host_alias(alias):
    try:
        return git.config('bz-tracker.' + alias + '.host', get=True)
    except CalledProcessError:
        return alias

def split_local_config(config_text):
    result = {}

    for line in config_text.split("\n"):
        line = re.sub("#.*", "", line)
        line = line.strip()
        if line == "":
            continue
        m = re.match("([a-zA-Z0-9-]+)\s*=\s*(.*)", line)
        if not m:
            die("Bad config line '%s'" % line)

        param = m.group(1)
        value = m.group(2)

        result[param] = value

    return result

def get_git_config(name):
    try:
        name = name.replace(".", r"\.")
        config_options = git.config(r'bz-tracker\.' + name + r'\..*', get_regexp=True)
    except CalledProcessError:
        return {}

    result = {}
    for line in config_options.split("\n"):
        line = line.strip()
        m = re.match("(\S+)\s+(.*)", line)
        key = m.group(1)
        value = m.group(2)

        m = re.match(r'bz-tracker\.' + name + r'\.(.*)', key)
        param = m.group(1)

        result[param] = value

    return result

# We only ever should be the config for one tracker in the course of a single run
cached_config = None
cached_config_tracker = None

def get_config(tracker):
    global cached_config
    global cached_config_tracker

    if cached_config == None:
        cached_config_tracker = tracker
        host = resolve_host_alias(tracker)
        cached_config = split_local_config(DEFAULT_CONFIG)
        if host in CONFIG:
            cached_config.update(split_local_config(CONFIG[host]))
        cached_config.update(get_git_config(host))
        if tracker != host:
            cached_config.update(get_git_config(tracker))

    assert cached_config_tracker == tracker

    return cached_config

def tracker_uses_https(tracker):
    config = get_config(tracker)
    return 'https' in config and config['https'] == 'true'

def tracker_get_path(tracker):
    config = get_config(tracker)
    if 'path' in config:
        return config['path']
    return None

def tracker_get_auth_user(tracker):
    config = get_config(tracker)
    if 'path' in config:
        return config['auth-user']
    return None

def tracker_get_auth_password(tracker):
    config = get_config(tracker)
    if 'path' in config:
        return config['auth-password']
    return None

def get_default_fields(tracker):
    config = get_config(tracker)

    default_fields = {}

    for key, value in config.iteritems():
        if key.startswith("default-"):
            param = key[8:].replace("-", "_")
            default_fields[param] = value

    return default_fields

# Utility functions for bugzilla
# ==============================

class BugParseError(Exception):
    pass

# A BugHandle is the parsed form of a bug reference string; it
# uniquely identifies a bug on a server, though until we try
# to load it (and create a Bug) we don't know if it actually exists.
class BugHandle:
    def __init__(self, host, path, https, id, auth_user=None, auth_password=None):
        self.host = host
        self.path = path
        self.https = https
        self.id = id
        self.auth_user = auth_user
        self.auth_password = auth_password

        # ensure that the path to the bugzilla installation is an absolute path
        # so that it will still work even if their config option specifies
        # something like:
        #   path = bugzilla
        # instead of the proper form:
        #   path = /bugzilla
        if self.path and self.path[0] != '/':
            self.path = '/' + self.path

    def get_url(self):
        return "%s://%s/show_bug.cgi?id=%s" % ("https" if self.https else "http",
                                               self.host,
                                               self.id)

    def needs_auth(self):
        return self.auth_user and self.auth_password

    @staticmethod
    def parse(bug_reference):
        parseresult = urlparse.urlsplit (bug_reference)

        if parseresult.scheme in ('http', 'https'):
            # Catch http://www.gnome.org and the oddball http:relative/path and http:/path
            if len(parseresult.path) == 0 or parseresult.path[0] != '/' or parseresult.hostname is None:
                raise BugParseError("Invalid bug reference '%s'" % bug_reference)

            user = parseresult.username
            password = parseresult.password
            # if the url did not specify http auth credentials in the form
            # https://user:password@host.com, check to see whether the config file
            # specifies any auth credentials for this host
            if not user:
                user = tracker_get_auth_user(parseresult.hostname)
            if not password:
                password = tracker_get_auth_password(parseresult.hostname)

            # strip off everything after the last '/', so '/bugzilla/show_bug.cgi'
            # will simply become '/bugzilla'
            base_path = parseresult.path[:parseresult.path.rfind('/')]
            m = re.match("id=([^&]+)", parseresult.query)

            if m:
                return BugHandle(host=parseresult.hostname,
                                 path=base_path,
                                 https=parseresult.scheme=="https",
                                 id=m.group(1),
                                 auth_user=user,
                                 auth_password=password)

        colon = bug_reference.find(":")
        if colon > 0:
            tracker = bug_reference[0:colon]
            id = bug_reference[colon + 1:]
        else:
            tracker = get_tracker()
            id = bug_reference

        if not id.isdigit():
            raise BugParseError("Invalid bug reference '%s'" % bug_reference)

        host = resolve_host_alias(tracker)
        https = tracker_uses_https(tracker)
        path = tracker_get_path(tracker)
        auth_user = tracker_get_auth_user(tracker)
        auth_password = tracker_get_auth_password(tracker)

        if not re.match(r"^.*\.[a-zA-Z]{2,}$", host):
            raise BugParseError("'%s' doesn't look like a valid bugzilla host or alias" % host)

        return BugHandle(host=host, path=path, https=https, id=id, auth_user=auth_user, auth_password=auth_password)

    @staticmethod
    def parse_or_die(str):
        try:
            return BugHandle.parse(str)
        except BugParseError, e:
            die(e.message)

    def __hash__(self):
        return hash((self.host, self.https, self.id))

    def __eq__(self, other):
        return ((self.host, self.https, self.id) ==
                (other.host, other.https, other.id))

class CookieError(Exception):
    pass

def do_get_cookies_from_sqlite(host, cookies_sqlite, browser, query, chromium_time):
    result = {}
    # We use a timeout of 0 since we expect to hit the browser holding
    # the lock often and we need to fall back to making a copy without a delay
    connection = sqlite.connect(cookies_sqlite, timeout=0)

    try:
        cursor = connection.cursor()
        cursor.execute(query, { 'host': host })

        now = time.time()
        for name,value,path,expiry in cursor.fetchall():
            # Excessive caution: toss out values that need to be quoted in a cookie header
            expiry = float(expiry)
            if chromium_time:
                # Time stored in microseconds since epoch
                expiry /= 1000000.
                # Old chromium versions used to use the Unix epoch, but newer versions
                # use the Windows epoch of January 1, 1601. Convert the latter to Unix epoch
                if expiry > 11644473600:
                    expiry -= 11644473600
            if float(expiry) > now and not re.search(r'[()<>@,;:\\"/\[\]?={} \t]', value):
                result[name] = value

        return result
    # Let the user know what might be going wrong in the case of a cryptic database error.
    except sqlite.DatabaseError:
        print ("Python threw a database error. A likely cause is that your version of python "
               "doesn't have a recent enough sqlite module to support this cookie database. "
               "Try upgrading your python sqlite module.")
        raise
    finally:
        connection.close()

# Firefox 3.5 keeps the cookies database permamently locked; as a workaround
# hack, we make a copy, read from that, then delete the copy. Of course,
# we may hit an inconsistent state of the database
def get_cookies_from_sqlite_with_copy(host, cookies_sqlite, browser, *args, **kwargs):
    db_copy = cookies_sqlite + ".git-bz-temp"
    shutil.copyfile(cookies_sqlite, db_copy)
    try:
        return do_get_cookies_from_sqlite(host, db_copy, browser, *args, **kwargs)
    except sqlite.OperationalError, e:
        raise CookieError("Cookie database was locked; temporary copy didn't work")
    finally:
        os.remove(db_copy)

def get_cookies_from_sqlite(host, cookies_sqlite, browser, query, chromium_time=False):
    try:
        result = do_get_cookies_from_sqlite(host, cookies_sqlite, browser, query,
                                            chromium_time=chromium_time)
    except sqlite.OperationalError, e:
        if "database is locked" in str(e):
            # Try making a temporary copy
            result = get_cookies_from_sqlite_with_copy(host, cookies_sqlite, browser, query,
                                                       chromium_time=chromium_time)
        else:
            raise

    if not ('Bugzilla_login' in result and 'Bugzilla_logincookie' in result):
        raise CookieError("You don't appear to be signed into %s; please log in with %s" % (host,
                                                                                            browser))

    return result

def get_cookies_from_sqlite_xulrunner(host, cookies_sqlite, name):
    return get_cookies_from_sqlite(host, cookies_sqlite, name,
                                   "select name,value,path,expiry from moz_cookies where host = :host")

def get_bugzilla_cookies_ff3(host):
    if (platform.system() == "Darwin"):
        profiles_dir = os.path.expanduser('~/Library/Application Support/Firefox')
    else:
        profiles_dir = os.path.expanduser('~/.mozilla/firefox')

    profile_path = None
    specific_profile = get_firefox_profile_pref()

    cp = RawConfigParser()
    cp.read(os.path.join(profiles_dir, "profiles.ini"))
    for section in cp.sections():
        if not cp.has_option(section, "Path"):
            continue

        # If we're looking for a specific profile, just check for that
        if specific_profile != "":
            if cp.has_option(section, "Name") and cp.get(section, "Name").strip() == specific_profile:
                profile_path = os.path.join(profiles_dir, cp.get(section, "Path").strip())

        # Otherwise use the profile tagged as default
        elif (not profile_path or
            (cp.has_option(section, "Default") and cp.get(section, "Default").strip() == "1")):
            profile_path = os.path.join(profiles_dir, cp.get(section, "Path").strip())

    if not profile_path:
        raise CookieError("Cannot find default Firefox profile")

    cookies_sqlite = os.path.join(profile_path, "cookies.sqlite")
    if not os.path.exists(cookies_sqlite):
        raise CookieError("%s doesn't exist." % cookies_sqlite)

    return get_cookies_from_sqlite_xulrunner(host, cookies_sqlite, "Firefox")

def get_bugzilla_cookies_epy(host):
    # epiphany-webkit migrated the cookie db to a different location, but the
    # format is the same
    profile_dir = os.path.expanduser('~/.gnome2/epiphany')
    cookies_sqlite = os.path.join(profile_dir, "cookies.sqlite")
    if not os.path.exists(cookies_sqlite):
        # try the old location
        cookies_sqlite = os.path.join(profile_dir, "mozilla/epiphany/cookies.sqlite")

    if not os.path.exists(cookies_sqlite):
        raise CookieError("%s doesn't exist" % cookies_sqlite)

    return get_cookies_from_sqlite_xulrunner(host, cookies_sqlite, "Epiphany")

# Shared for Chromium and Google Chrome
def get_bugzilla_cookies_chr(host, browser, config_dir):
    config_dir = os.path.expanduser(config_dir)
    cookies_sqlite = os.path.join(config_dir, "Cookies")
    if not os.path.exists(cookies_sqlite):
        raise CookieError("%s doesn't exist" % cookies_sqlite)
    return get_cookies_from_sqlite(host, cookies_sqlite, browser,
                                   "select name,value,path,expires_utc from cookies where host_key = :host",
                                   chromium_time=True)

def get_bugzilla_cookies_chromium(host):
    return get_bugzilla_cookies_chr(host,
                                    "Chromium",
                                    '~/.config/chromium/Default')

def get_bugzilla_cookies_google_chrome(host):
    return get_bugzilla_cookies_chr(host,
                                    "Google Chrome",
                                    '~/.config/google-chrome/Default')

browsers = { 'firefox3'     : get_bugzilla_cookies_ff3,
             'epiphany'     : get_bugzilla_cookies_epy,
             'chromium'     : get_bugzilla_cookies_chromium,
             'google-chrome': get_bugzilla_cookies_google_chrome }

def browser_list():
    return ", ".join(sorted(browsers.keys()))

def get_bugzilla_cookies(host):
    browser = get_browser()
    if browser in browsers:
        do_get_cookies = browsers[browser]
    else:
        die('Unsupported browser %s (we only support %s)' % (browser, browser_list()))

    try:
        return do_get_cookies(host)
    except CookieError, e:
        die("""Error getting login cookie from browser:
   %s

Configured browser: %s (change with 'git config --global bz.browser <value>')
Possible browsers: %s""" %
            (str(e), browser, browser_list()))

# Based on http://code.activestate.com/recipes/146306/ - Wade Leftwich
def encode_multipart_formdata(fields, files=None):
    """
    fields is a dictionary of { name : value } for regular form fields. if value is a list,
      one form field is added for each item in the list
    files is a dictionary of { name : ( filename, content_type, value) } for data to be uploaded as files
    Return (content_type, body) ready for httplib.HTTPContent instance
    """
    BOUNDARY = '----------ThIs_Is_tHe_bouNdaRY_$'
    CRLF = '\r\n'
    L = []
    for key in sorted(fields.keys()):
        value = fields[key]
        if isinstance(value, list):
            for v in value:
                L.append('--' + BOUNDARY)
                L.append('Content-Disposition: form-data; name="%s"' % key)
                L.append('')
                L.append(v)
        else:
            L.append('--' + BOUNDARY)
            L.append('Content-Disposition: form-data; name="%s"' % key)
            L.append('')
            L.append(value)
    if files:
        for key in sorted(files.keys()):
            (filename, content_type, value) = files[key]
            L.append('--' + BOUNDARY)
            L.append('Content-Disposition: form-data; name="%s"; filename="%s"' % (key, filename))
            L.append('Content-Type: %s' % content_type)
            L.append('')
            L.append(value)
    L.append('--' + BOUNDARY + '--')
    L.append('')
    body = CRLF.join(L)
    content_type = 'multipart/form-data; boundary=%s' % BOUNDARY
    return content_type, body

# Cache of constant-responses per bugzilla server
# ===============================================

CACHE_EXPIRY_TIME = 3600 * 24 # one day

class Cache(object):
    def __init__(self):
        self.cfp = None

    def __ensure(self, host):
        if self.cfp == None:
            self.cfp = RawConfigParser()
            self.cfp.read(os.path.expanduser("~/.git-bz-cache"))

        if self.cfp.has_section(host):
            if time.time() > self.cfp.getfloat(host, "expires"):
                self.cfp.remove_section(host)

        if not self.cfp.has_section(host):
            self.cfp.add_section(host)
            self.cfp.set(host, "expires", time.time() + CACHE_EXPIRY_TIME)

    def get(self, host, key):
        self.__ensure(host)
        try:
            return pickle.loads(self.cfp.get(host, key))
        except NoOptionError:
            raise IndexError()

    def set(self, host, key, value):
        self.__ensure(host)
        self.cfp.set(host, key, pickle.dumps(value))
        f = open(os.path.expanduser("~/.git-bz-cache"), "w")
        self.cfp.write(f)
        f.close()

cache = Cache()

# General Utility Functions
# =========================

def make_filename(description):
    filename = re.sub(r"\s+", "-", description)
    filename = re.sub(r"[^A-Za-z0-9-]+", "", filename)
    filename = filename[0:50]

    return filename

def edit_file(filename):
    editor = None
    if 'GIT_EDITOR' in os.environ:
        editor = os.environ['GIT_EDITOR']
    if editor == None:
        try:
            editor = git.config('core.editor', get=True)
        except CalledProcessError:
            pass
    if editor == None and 'EDITOR' in os.environ:
        editor = os.environ['EDITOR']
    if editor == None:
        editor = "vi"

    process = Popen(editor + " " + filename, shell=True)
    process.wait()
    if process.returncode != 0:
        die("Editor exited with non-zero return code")

def edit_template(template):
    # Prompts the user to edit the text 'template' and returns list of
    # lines with comments stripped

    handle, filename = tempfile.mkstemp(".txt", "git-bz-")
    f = os.fdopen(handle, "w")
    f.write(template)
    f.close()

    edit_file(filename)

    f = open(filename, "r")
    lines = filter(lambda x: not x.startswith("#"), f.readlines())
    f.close

    return lines

def split_subject_body(lines):
    # Splits the first line (subject) from the subsequent lines (body)

    i = 0
    subject = ""
    while i < len(lines):
        subject = lines[i].strip()
        if subject != "":
            break
        i += 1

    return subject, "".join(lines[i + 1:]).strip()

def _shortest_unique_abbreviation(full, l):
    for i in xrange(1, len(full) + 1):
        abbrev = full[0:i]
        if not any((x != full and x.startswith(abbrev) for x in l)):
            return abbrev
    # Duplicate items or one item is a prefix of another
    raise ValueError("%s has no unique abbreviation in %s" % (full, l))

def _abbreviation_item_help(full, l):
    abbrev = _shortest_unique_abbreviation(full, l)
    return '[%s]%s' % (abbrev, full[len(abbrev):])

# Return '[a]pple, [pe]ar, [po]tato'
def abbreviation_help_string(l):
    return ", ".join((_abbreviation_item_help(full, l) for full in l))

# Find the unique element in l that starts with abbrev
def expand_abbreviation(abbrev, l):
    for full in l:
        if full.startswith(abbrev) and len(abbrev) >= len(_shortest_unique_abbreviation(full, l)):
            return full
    raise ValueError("No unique abbreviation expansion")

def prompt(message):
    while True:
        # Using print here could result in Python adding a stray space
        # before the next print
        sys.stdout.write(message + " [yn] ")
        line = sys.stdin.readline().strip()
        if line == 'y' or line == 'Y':
            return True
        elif line == 'n' or line == 'N':
            return False

def die(message):
    print >>sys.stderr, message
    sys.exit(1)

def http_auth_header(user, password):
    return 'Basic ' + base64.encodestring("%s:%s" % (user, password)).strip()

# Classes for bug handling
# ========================

class BugPatch(object):
    def __init__(self, attach_id):
        self.attach_id = attach_id

class NoXmlRpcError(Exception):
    pass

connections = {}

def get_connection(host, https):
    identifier = (host, https)
    if not identifier in connections:
        if https:
            connection = httplib.HTTPSConnection(host, 443)
        else:
            connection = httplib.HTTPConnection(host, 80)

        connections[identifier] = connection

    return connections[identifier]

def kill_connection(host, https):
    identifier = (host, https)
    if identifier in connections:
        del connections[identifier]


class BugServer(object):
    def __init__(self, host, path, https, auth_user=None, auth_password=None):
        self.host = host
        self.path = path
        self.https = https
        self.auth_user = auth_user
        self.auth_password = auth_password

        self.cookies = get_bugzilla_cookies(host)

        self._xmlrpc_proxy = None

    def get_cookie_string(self):
        return ("Bugzilla_login=%s; Bugzilla_logincookie=%s" %
                (self.cookies['Bugzilla_login'], self.cookies['Bugzilla_logincookie']))

    def send_request(self, method, url, data=None, headers={}):
        headers = dict(headers)
        headers['Cookie'] = self.get_cookie_string()
        headers['User-Agent'] = "git-bz"
        if self.auth_user and self.auth_password:
            headers['Authorization'] = http_auth_header(self.auth_user, self.auth_password)
        if self.path:
            url = self.path + url

        seen_urls = []
        retries = 0
        connection = get_connection(self.host, self.https)
        while True:

            # BMO seems to generate some "bad status line" exception intermittently.
            try:
              connection.request(method, url, data, headers)
              response = connection.getresponse()
            except httplib.BadStatusLine:
              retries = retries + 1
              print "Got bad status line - retrying..."
              connection.close()
              kill_connection(self.host, self.https)
              time.sleep(2)
              connection = get_connection(self.host, self.https)
              continue
            retries = 0

            seen_urls.append(url)

            # Redirect status codes:
            #
            # 301 (Moved Permanently): Redo with the new URL,
            #   save the new location.
            # 303 (See Other): Redo with the method changed to GET/HEAD.
            # 307 (Temporary Redirect): Redo with the new URL, don't
            #   save the new location.
            #
            # [ For 301/307, you are supposed to ask the user if the
            #   method isn't GET/HEAD, but we're automating anyways... ]
            #
            # 302 (Found): The confusing one, and the one that
            # Bugzilla uses, both to redirect to http to https and to
            # redirect attachment.cgi&action=view to a different base URL
            # for security. Specified like 307, traditionally treated as 301.
            #
            # See http://en.wikipedia.org/wiki/HTTP_302

            if response.status in (301, 302, 303, 307):
                new_url = response.getheader("location")
                if new_url is None:
                    die("Redirect received without a location to redirect to")
                if new_url in seen_urls or len(seen_urls) >= 10:
                    die("Circular redirect or too many redirects")

                old_split = urlparse.urlsplit(url)
                new_split = urlparse.urlsplit(new_url)

                new_https = new_split.scheme == 'https'

                if new_split.hostname != self.host or new_https != self.https:
                    connection = get_connection(new_split.hostname, new_https != self.https)

                    # This is a bit of a hack to avoid keeping on redirecting for every
                    # request. If the server redirected show_bug.cgi we assume it's
                    # really saying "hey, the bugzilla instance is really over here".
                    #
                    # We can't do this for old.split.path == new_split.path because of
                    # attachment.cgi, though we alternatively could just exclude
                    # attachment.cgi here.
                    if (response.status in (301, 302) and
                        method == 'GET' and
                        old_split.path == '/show_bug.cgi' and new_split.path == '/show_bug.cgi'):

                        self.host = new_split.hostname
                        self.https = new_https

                # We can't treat 302 like 303 because of the use of 302 for http
                # to https, though the hack above will hopefully get us on https
                # before we try to POST.
                if response.status == 303:
                    if method not in ('GET', 'HEAD'):
                        method = 'GET'

                # Get the relative component of the new URL
                url = urlparse.urlunsplit((None, None, new_split.path, new_split.query, new_split.fragment))
            else:
                return response

    def send_post(self, url, fields, files=None):
        content_type, body = encode_multipart_formdata(fields, files)
        return self.send_request("POST", url, data=body, headers={ 'Content-Type': content_type })

    def get_xmlrpc_proxy(self):
        if self._xmlrpc_proxy is None:
            uri = "%s://%s/xmlrpc.cgi" % ("https" if self.https else "http",
                                          self.host)
            if self.https:
                transport = SafeBugTransport(self)
            else:
                transport = BugTransport(self)
            self._xmlrpc_proxy = xmlrpclib.ServerProxy(uri, transport)

        return self._xmlrpc_proxy

    # Query the server for the legal values of the given field; returns an
    # array, or None if the query failed
    def _legal_values(self, field):
        try:
            response = self.get_xmlrpc_proxy().Bug.legal_values({ 'field': field })
            cache.set(self.host, 'legal_' + field, response['values'])
            return response['values']
        except xmlrpclib.Fault, e:
            if e.faultCode == -32000: # https://bugzilla.mozilla.org/show_bug.cgi?id=513511
                return None
            raise
        except xmlrpclib.ProtocolError, e:
            if e.errcode == 500: # older bugzilla versions die this way
                return None
            elif e.errcode == 404: # really old bugzilla, no XML-RPC
                return None
            raise

    def legal_values(self, field):
        try:
            return cache.get(self.host, 'legal_' + field)
        except IndexError:
            values = self._legal_values(field)
            cache.set(self.host, 'legal_' + field, values)
            return values

# mixin for xmlrpclib.Transport classes to add cookies
class CookieTransportMixin(object):
    def send_request(self, connection, *args):
        xmlrpclib.Transport.send_request(self, connection, *args)
        connection.putheader("Cookie", self.server.get_cookie_string())
        connection.putheader("Authorization", http_auth_header(self.server.auth_user, self.server.auth_password))

class BugTransport(CookieTransportMixin, xmlrpclib.Transport):
    def __init__(self, server):
        xmlrpclib.Transport.__init__(self)
        self.server = server

class SafeBugTransport(CookieTransportMixin, xmlrpclib.SafeTransport):
    def __init__(self, server):
        xmlrpclib.SafeTransport.__init__(self)
        self.server = server

servers = {}

# Note that if we detect that we are redirecting, we may rewrite the
# host/https of the server to avoid doing too many redirections, and
# so the host,https we connect to may be different than what we use
# to look up the server.
def get_bug_server(host, path, https, auth_user, auth_password):
    identifier = (host, path, https)
    if not identifier in servers:
        servers[identifier] = BugServer(host, path, https, auth_user, auth_password)

    return servers[identifier]

# Unfortunately, Bugzilla doesn't set a useful status code for
# form posts.  Because it's very confusing to claim we succeeded
# but not, we look for text in the response indicating success,
# and not text indicating failure.
#
# We generally look for specific <title> tags - these have been
# quite stable across versions, though translations will throw
# us off.
#
# *args are regular expressions to search for in response_data
# that indicate success. Returns the matched regular expression
# on success, None otherwise
def check_for_success(response, response_data, *args):

    if response.status != 200:
        return False

    for pattern in args:
        m = re.search(pattern, response_data)
        if m:
            return m

    return None

class Bug(object):
    def __init__(self, server):
        self.server = server
        self.id = None
        self.product = None
        self.component = None
        self.short_desc = None
        self.patches = []

    def _load(self, id, attachmentdata=False):
        url = "/show_bug.cgi?id=" + id + "&ctype=xml"
        if not attachmentdata:
            url += "&excludefield=attachmentdata"

        response = self.server.send_request("GET", url)
        if response.status != 200:
            die ("Failed to retrieve bug information: %d" % response.status)

        etree = ElementTree()
        etree.parse(response)

        bug = etree.find("bug")
        error = bug.get("error")
        if error != None:
            die ("Failed to retrieve bug information: %s" % error)

        self.id = int(bug.find("bug_id").text)
        self.short_desc = bug.find("short_desc").text
        self.bug_status = bug.find("bug_status").text
        if self.bug_status == "RESOLVED":
            self.resolution = bug.find("resolution").text
        token = bug.find("token")
        self.token = None if token is None else token.text

        for attachment in bug.findall("attachment"):
            if attachment.get("ispatch") == "1" and not attachment.get("isobsolete") == "1" :
                attach_id = int(attachment.find("attachid").text)
                patch = BugPatch(attach_id)
                # We have to save fields we might not otherwise care about
                # (like isprivate) so that we can pass them back when updating
                # the attachment
                patch.description = attachment.find("desc").text
                patch.date = attachment.find("date").text
                status = attachment.find("status")
                patch.status = None if status is None else status.text
                patch.filename = attachment.find("filename").text
                patch.isprivate = attachment.get("isprivate") == "1"
                token = attachment.find("token")
                patch.token = None if token is None else token.text

                if attachmentdata:
                    data = attachment.find("data").text
                    patch.data = base64.b64decode(data)
                else:
                    patch.data = None

                self.patches.append(patch)

    def _create_via_xmlrpc(self, product, component, short_desc, comment, default_fields):
        params = dict()
        params['product'] = product
        params['component'] = component
        params['summary'] = short_desc
        params['description'] = comment
        for (field, value) in default_fields.iteritems():
            params[field] = value

        try:
            response = self.server.get_xmlrpc_proxy().Bug.create(params)
            self.id = response['id']
        except xmlrpclib.Fault, e:
            die(e.faultString)
        except xmlrpclib.ProtocolError, e:
            if e.errcode == 404:
                raise NoXmlRpcError(e.errmsg)
            else:
                print >>sys.stderr, "Problem filing bug via XML-RPC: %s (%d)\n" % (e.errmsg, e.errcode)
                print >>sys.stderr, "falling back to form post\n"
                raise NoXmlRpcError("Communication error")

    def _create_with_form(self, product, component, short_desc, comment, default_fields):
        fields = dict()
        fields['product'] = product
        fields['component'] = component
        fields['short_desc'] = short_desc
        fields['comment'] = comment

        # post_bug.cgi wants some names that are less congenial than the names
        # expected in XML-RPC.
        for (field, value) in default_fields.iteritems():
            if field == 'severity':
                field = 'bug_severity'
            elif field == 'platform':
                field = 'rep_platform'
            fields[field] = value

        # Priority values vary wildly between different servers because the stock
        # Bugzilla uses the awkward P1/../P5. It will be defaulted on the XML-RPC
        # code path, but we need to take a wild guess here.
        if not 'priority' in fields:
            fields['priority'] = 'P5'
        # Legal severity values are much more standardized, but not specifying it
        # in the XML-RPC code path allows the server default to win. We need to
        # specify something here.
        if not 'severity' in fields:
            fields['bug_severity'] = 'normal'
        # Required, but a configured default doesn't make any sense
        if not 'bug_file_loc' in fields:
            fields['bug_file_loc'] = ''

        response = self.server.send_post("/post_bug.cgi", fields)
        response_data = response.read()
        m = check_for_success(response, response_data,
                              r"<title>\s*Bug\s+([0-9]+)")
        if not m:
            print response_data
            die("Failed to create bug, status=%d" % response.status)

        self.id = int(m.group(1))

    def _create(self, product, component, short_desc, comment, default_fields):
        try:
            self._create_via_xmlrpc(product, component, short_desc, comment, default_fields)
        except NoXmlRpcError:
            self._create_with_form(product, component, short_desc, comment, default_fields)

        print "Successfully created"
        print "Bug %d - %s" % (self.id, short_desc)
        print self.get_url()

    def create_patch(self, description, comment, filename, data, obsoletes=[], status='none', review=[], superreview=[]):
        fields = {}
        fields['bugid'] = str(self.id)
        fields['action'] = 'insert'
        fields['ispatch'] = '1'
        fields['attachments.status'] = status
        fields['description'] = description
        if comment:
            fields['comment'] = comment
        if obsoletes:
            # this will produce multiple parts in the encoded data with the
            # name 'obsolete' for each item in the list
            fields['obsolete'] = map(str, obsoletes)

        # Support for flagging for review/superreview. I have no idea if this works on
        # installations other than bugzilla.mozilla.org.
        if review:
            if review[0] == ':me+':
                fields['flag_type-4'] = '+'
            else:
                fields['flag_type-4'] = '?'
                fields['requestee_type-4'] = ",".join(review)
        if superreview:
            if superreview[0] == ':me+':
                fields['flag_type-5'] = '+'
            else:
                fields['flag_type-5'] = '?'
                fields['requestee_type-5'] = ",".join(superreview)

        files = { 'data': (filename, 'text/plain', data) }

        response = self.server.send_post("/attachment.cgi", fields, files)
        response_data = response.read()
        if not check_for_success(response, response_data,
                                 # Older bugzilla's used this for successful attachments
                                 r"<title>\s*Changes\s+Submitted",
                                 # Newer bugzilla's, use, instead:
                                 r"<title>\s*Attachment\s+\d+\s+added"):
            print response_data
            die ("Failed to attach patch to bug %d, status=%d" % (self.id, response.status))

        print "Attached %s" % filename

    # Update specified fields of a bug; keyword arguments are interpreted
    # as field_name=value
    def update(self, **changes):
        changes['id'] = str(self.id)
        if self.token:
            changes['token'] = self.token
        # Since we don't send delta_ts we'll never get a mid-air collision
        # This is probably a good thing

        response = self.server.send_post("/process_bug.cgi", changes)
        response_data = response.read()
        if not check_for_success(response, response_data,
                                 r"<title>\s*Bug[\S\s]*processed\s*</title>"):

            # Mid-air collisions would be indicated by
            # "<title>Mid-air collision!</title>"

            print response_data
            die ("Failed to update bug %d, status=%d" % (self.id, response.status))

    # Update specified fields of an attachment; keyword arguments are
    # interpreted as field_name=value
    def update_patch(self, patch, **changes):
        # Unlike /process_bug.cgi, the attachment editing interface doesn't
        # support defaulting missing fields to their existing values, so we
        # have to pass everything back.
        fields = {
            'action': 'update',
            'id': str(patch.attach_id),
            'description': patch.description,
            'filename': patch.filename,
            'ispatch': "1",
            'isobsolete': "0",
            'isprivate': "1" if patch.isprivate else "0",
        };

        if patch.token:
            fields['token'] = patch.token
        if patch.status is not None:
            fields['attachments.status'] = patch.status

        for (field, value) in changes.iteritems():
            if field == 'status': # encapsulate oddball form field name
                field = 'attachments.status'
            fields[field] = value

        response = self.server.send_post("/attachment.cgi", fields)
        response_data = response.read()
        if not check_for_success(response, response_data,
                                 r"<title>\s*Changes\s+Submitted"):
            print response_data
            die ("Failed to update attachment %d to bug %d, status=%d" % (patch.attach_id,
                                                                          self.id,
                                                                          response.status))

    def get_url(self):
        return "%s://%s/show_bug.cgi?id=%d" % ("https" if self.server.https else "http",
                                               self.server.host,
                                               self.id)

    @staticmethod
    def load(bug_reference, attachmentdata=False):
        server = get_bug_server(bug_reference.host, bug_reference.path, bug_reference.https, bug_reference.auth_user, bug_reference.auth_password)
        bug = Bug(server)
        bug._load(bug_reference.id, attachmentdata)

        return bug

    @staticmethod
    def create(tracker, product, component, short_desc, comment):
        host = resolve_host_alias(tracker)
        https = tracker_uses_https(tracker)
        path = tracker_get_path(tracker)
        auth_user = tracker_get_auth_user(tracker)
        auth_password = tracker_get_auth_password(tracker)
        default_fields = get_default_fields(tracker)

        server = get_bug_server(host, path, https, auth_user, auth_password)
        bug = Bug(server)
        bug._create(product, component, short_desc, comment, default_fields)

        return bug

# The Commands
# =============

def commit_needs_url(commit, bug_id):
        pat = re.compile(r"(?<!\d)%d(?!\d)" % bug_id)
        return (pat.search(commit.subject) is None and
                pat.search(get_body(commit)) is None)

def check_add_url(commits, bug_id=None, is_add_url=False):
    if bug_id != None:
        # We only need to check the commits that we'll add the URL to
        commits = [commit for commit in commits if commit_needs_url(commit, bug_id)]

    if len(commits) == 0: # Nothing to do
        return

    try:
        git.diff(exit_code=True, ignore_submodules=True, _quiet=True)
        git.diff(exit_code=True, ignore_submodules=True, cached=True,  _quiet=True)
    except CalledProcessError:
        die("Cannot add bug reference to commit message(s); You must commit (or stash) all changes first")

    for commit in commits:
        # check that the commit is an ancestor of the current revision
        base = git.merge_base("HEAD", commit.id)
        if base != commit.id:
            die("%s %s\nNot an ancestor of HEAD, can't add bug URL to it" % (commit.id[0:7], commit.subject))

        # see if the commit is present in any remote branches
        remote_branches = git.branch(contains=commit.id, r=True)
        if remote_branches != "":
            print commit.id[0:7], commit.subject
            print "Commit is already in remote branch(es):", " ".join(remote_branches.split())
            if not prompt("Rewrite the commit add the bug URL anyways?"):
                if is_add_url:
                    print "Aborting."
                else:
                    print "Aborting. You can use -n/--no-add-url to turn off adding the URL"
                sys.exit(0)

    # Check for merge commits
    oldest_commit = commits[-1]
    all_commits = rev_list_commits(commits[-1].id + "^..HEAD")
    for commit in all_commits:
        if commit_is_merge(commit):
            print "Found merge commit:"
            print commit.id[0:7], commit.subject
            print "Can't rewrite this commit or an ancestor commit to add bug URL"
            sys.exit(1)

def bad_url_method(add_url_method):
    die("""add-url-method '%s' is invalid
Should be [subject-prepend|subject-append|body-prepend|body-append]:<format>""" %
        add_url_method)

def add_url_to_subject_body(subject, body, bug):
    add_url_method = get_add_url_method()
    if not ':' in add_url_method:
        bad_url_method(add_url_method)

    method, format = add_url_method.split(':', 1)

    def sub_percent(m):
        if m.group(1) == 'u':
            return bug.get_url()
        elif m.group(1) == 'd':
            return str(bug.id)
        elif m.group(1) == 'n':
            return "\n"
        elif m.group(1) == '%':
            return "%"
        else:
            die("Bad add-url-method escape %%%s" % m.group(1))

    formatted = re.sub("%(.)", sub_percent, format)

    if method == 'subject-prepend':
        subject = formatted + " " + subject
    elif method == 'subject-append':
        subject = subject + " " + formatted
    elif method == 'body-prepend':
        body = formatted + "\n\n" + body
    elif method == 'body-append':
        body = body + "\n\n" + formatted
    else:
        bad_url_method(add_url_method)

    return subject, body

def validate_add_url_method(bug):
    # Dry run
    add_url_to_subject_body("", "", bug)

def add_url_to_head_commit(commit, bug):
    subject = commit.subject
    body = get_body(commit)

    subject, body = add_url_to_subject_body(subject, body, bug)

    input = subject + "\n\n" + body
    git.commit(file="-", amend=True, _input=input)

def add_url(bug, commits):
    commit_map = {}
    oldest_commit = None
    for commit in commits:
        commit_map[commit.id] = commit
        if commit_needs_url(commit, bug.id):
            oldest_commit = commit

    if not oldest_commit:
        return

    # Check that the add-url method is valid before starting the rebase
    validate_add_url_method(bug)

    all_commits = rev_list_commits(oldest_commit.id + "^..HEAD")
    orig_head = all_commits[0].id

    try:
        branch_name = git.symbolic_ref("HEAD", q=True)
    except CalledProcessError:
        branch_name = None
    try:
        # Detach HEAD from the branch; this gives a cleaner reflog for the branch
        print "Moving to starting point"
        git.checkout(oldest_commit.id + "^", q=True)

        for commit in reversed(all_commits):
            # Map back to the original commit object so we can update it
            if commit.id in commit_map:
                commit = commit_map[commit.id]

            if commit.id in commit_map and commit_needs_url(commit, bug.id):
                print "Adding bug reference  ", commit.id[0:7], commit.subject
                git.cherry_pick(commit.id)
                add_url_to_head_commit(commit, bug)
            else:
                if commit.id in commit_map:
                    print "Recommitting", commit.id[0:7], commit.subject, "(already has bug #)"
                else:
                    print "Recommitting", commit.id[0:7], commit.subject
                git.cherry_pick(commit.id)

            # Get the commit ID; we update the commit with the new ID, so we in the case
            # where we later format the patch, we format the patch with the added bug URL
            new_head = commit.id = git.rev_parse("HEAD")

        if branch_name is not None:
            git.update_ref("-m", "bz add-url: adding references to %s" % bug.get_url(),
                           branch_name, new_head)
            git.symbolic_ref("HEAD", branch_name)
    except:
        print "Cleaning up back to original state on error"
        git.reset(orig_head, hard=True)
        if branch_name is not None:
            git.symbolic_ref("HEAD", branch_name)
        raise

def do_add_url(bug_reference, commit_or_revision_range):
    commits = get_commits(commit_or_revision_range)

    bug = Bug.load(BugHandle.parse_or_die(bug_reference))

    check_add_url(commits, bug.id, is_add_url=True)

    print "Bug %d - %s" % (bug.id, bug.short_desc)
    print bug.get_url()
    print

    found = False
    for commit in commits:
        if commit_needs_url(commit, bug.id):
            print commit.id[0:7], commit.subject
            found = True
        else:
            print "SKIPPING", commit.id[0:7], commit.subject
    if not found:
        sys.exit(0)

    print
    if not prompt("Add bug URL to above commits?"):
        print "Aborting"
        sys.exit(0)

    print
    add_url(bug, commits)

def do_apply(bug_reference):
    bug = Bug.load(BugHandle.parse_or_die(bug_reference),
                   attachmentdata=True)

    print "Bug %d - %s" % (bug.id, bug.short_desc)
    print

    for patch in bug.patches:
        if patch.status == 'committed' or patch.status == 'rejected':
            print "Skipping, %s: %s" % (patch.status, patch.description)
            continue

        print patch.description
        if not prompt("Apply?"):
            continue

        print

        handle, filename = tempfile.mkstemp(".patch", make_filename(patch.description) + "-")
        f = os.fdopen(handle, "w")
        f.write(patch.data)
        f.close()

        try:
            process = git.am(filename, _interactive=True)
        except CalledProcessError:
            print "Patch left in %s" % filename
            break

        os.remove(filename)

        if global_options.add_url:
            # Slightly hacky, would be better to just commit right the first time
            commits = rev_list_commits("HEAD^!")
            add_url(bug, commits)

def strip_bug_url(bug, commit_body):
    # Strip off the trailing bug URLs we add with -u; we do this before
    # using commit body in as a comment; doing it by stripping right before
    # posting means that we are robust against someone running add-url first
    # and attach second.
    pattern = "\s*" + re.escape(bug.get_url()) + "\s*$"
    return re.sub(pattern, "", commit_body)

def edit_attachment_comment(bug, initial_description, initial_body):
    template = StringIO()
    template.write("# Attachment to Bug %d - %s\n\n" % (bug.id, bug.short_desc))
    template.write(initial_description)
    template.write("\n\n")
    template.write(initial_body)
    template.write("\n\n")
    template.write("Commit-Message: Bug %d - %s.\n\n" % (bug.id, bug.short_desc));
    template.write("#Review: reviewer1@example.com\n")
    template.write("#Review: reviewer2@example.com\n")
    template.write("#Review: :me+\n")
    template.write("#Super-Review: superreviewer@example.com\n\n")
    template.write("#Super-Review: :me+\n")
    if len(bug.patches) > 0:
        for patch in bug.patches:
            template.write("#Obsoletes: %d - %s\n" % (patch.attach_id, patch.description))
        template.write("\n")

    template.write("""# Please edit the description (first line) and comment (other lines). Lines
# starting with '#' will be ignored.  Delete everything to abort.
""")
    if len(bug.patches) > 0:
        template.write("# To obsolete existing patches, uncomment the appropriate lines.\n")

    lines = edit_template(template.getvalue())

    obsoletes= []
    review= []
    superreview= []
    commit_message = []

    def filter_special(line):
        m = re.match("^\s*Obsoletes\s*:\s*([\d]+)", line)

        r = re.match("^\s*Review\s*:\s*([\S]+@[\S]+\.[\w]+)\s*$", line)
        # Support :names
        if not r:
            r = re.match("^\s*Review\s*:\s*(:[\w]+\+?)\s*$", line)

        sr = re.match("^\s*Super-Review\s*:\s*([\S]+@[\S]+\.[\w]+)\s*$", line)
        # Support names
        if not sr:
            sr = re.match("^\s*Super-Review\s*:\s*(:[\w]+\+?)\s*$", line)

        # Parse commit messages
        c = re.match("^\s*Commit-Message\s*:\s*(.*)\s*$", line)

        if m:
            obsoletes.append(int(m.group(1)))
            return False
        elif r:
            review.append(r.group(1))
        elif sr:
            superreview.append(sr.group(1))
        elif c:
            commit_message.append(c.group(1))
        else:
            return True

    lines = filter(filter_special, lines)

    description, comment = split_subject_body(lines)

    if description == "":
        die("Empty description, aborting")

    # Work around inner-method variable access weirdness
    if commit_message:
        commit_message = commit_message[0]

    return description, comment, obsoletes, commit_message, review, superreview

def attach_commits(bug, commits, include_comments=True, edit_comments=False, status='none'):
    # We want to attach the patches in chronological order
    commits = list(commits)
    commits.reverse()

    for commit in commits:
        filename = make_filename(commit.subject) + ".patch"
        patch = get_patch(commit)
        if include_comments:
            body = strip_bug_url(bug, get_body(commit))
        else:
            body = None
        if edit_comments:
            description, body, obsoletes, commit_message, review, superreview = edit_attachment_comment(bug, commit.subject, body)
        else:
            description = commit.subject
            commit_message = commit.subject
            obsoletes = []
            review = []
            superreview = []
        if commit_message:
            patch = commit_message + "\n\n" + patch
        bug.create_patch(description, body, filename,  patch, obsoletes=obsoletes, status=status, review=review, superreview=superreview)

def do_attach(*args):
    if len(args) == 1:
        commit_or_revision_range = args[0]
        commits = get_commits(commit_or_revision_range)

        extracted = list(extract_and_collate_bugs(commits))
        if len(extracted) == 0:
            die("No bug references found in specified commits")
        elif len(extracted) > 1:
            # This could be sensible in the case of "attach updated patches
            # for all these commits", but for now, just make it an error
            die("Found multiple bug references specified commits:\n  " +
                "\n  ".join((handle.get_url() for handle, _ in extracted)))

        # extract_and_collate_bugs returns a list of commits that reference
        # the handle, but we ignore that - we want to attach all of the
        # specified commits, even if only some of the reference the bug
        handle, _ = extracted[0]
    else:
        bug_reference = args[0]
        commit_or_revision_range = args[1]

        commits = get_commits(commit_or_revision_range)
        handle = BugHandle.parse_or_die(bug_reference)

    bug = Bug.load(handle)

    if global_options.add_url:
        check_add_url(commits, bug.id, is_add_url=False)

    # We always want to prompt if the user has specified multiple attachments.
    # For the common case of one attachment don't prompt if we are going
    # to give them a chance to edit and abort anyways.
    if len(commits) > 1 or not global_options.edit:
        print "Bug %d - %s" % (bug.id, bug.short_desc)
        print

        for commit in reversed(commits):
            print commit.id[0:7], commit.subject

        print
        if not prompt("Attach?"):
            print "Aborting"
            sys.exit(0)

    if global_options.add_url:
        add_url(bug, commits)

    attach_commits(bug, commits, edit_comments=global_options.edit)

# Sort the patches in the bug into categories based on a set of Git
# git commits that we're considering to be newly applied. Matching
# is done on exact git subject <=> patch description matches.
def filter_patches(bug, applied_commits):
    newly_applied_patches = dict() # maps to the commit object where it was applied
    obsoleted_patches = set()
    unapplied_patches = set()

    applied_subjects = dict(((commit.subject, commit) for commit in applied_commits))
    seen_subjects = set()

    # Work backwards so that the latest patch is considered applied, and older
    # patches with the same subject obsoleted.
    for patch in reversed(bug.patches):
        # Previously committted or rejected patches are never a match
        if patch.status == "committed" or patch.status == "rejected":
            continue

        if patch.description in seen_subjects:
            obsoleted_patches.add(patch)
        elif patch.description in applied_subjects:
            newly_applied_patches[patch] = applied_subjects[patch.description]
            seen_subjects.add(patch)
        else:
            unapplied_patches.add(patch)

    return newly_applied_patches, obsoleted_patches, unapplied_patches

def edit_bug(bug, applied_commits=None, fix_commits=None):
    if applied_commits is not None:
        newly_applied_patches, obsoleted_patches, unapplied_patches = filter_patches(bug, applied_commits)
        mark_resolved = len(unapplied_patches) == 0 and bug.bug_status != "RESOLVED"
    else:
        newly_applied_patches = obsoleted_patches = set()
        mark_resolved = fix_commits is not None

    template = StringIO()
    template.write("# Bug %d - %s - %s" % (bug.id, bug.short_desc, bug.bug_status))
    if bug.bug_status == "RESOLVED":
        template.write(" - %s" % bug.resolution)
    template.write("\n")
    template.write("# %s\n" % bug.get_url())
    template.write("# Enter comment on following lines; delete everything to abort\n\n")

    if fix_commits is not None:
        if len(fix_commits) == 1:
            template.write("The following fix has been pushed:\n")
        else:
            template.write("The following fixes have been pushed:\n")
        for commit in reversed(fix_commits):
            template.write(commit.id[0:7] + " " + commit.subject + "\n")
        template.write("\n")

    for patch in bug.patches:
        if patch in newly_applied_patches:
            commit = newly_applied_patches[patch]
            template.write("Attachment %d pushed as %s - %s\n" % (patch.attach_id, commit.id[0:7], commit.subject))

    if mark_resolved:
        template.write("# Comment to keep bug open\n")
    elif bug.bug_status == "RESOLVED":
        template.write("# Uncommment and edit to change resolution\n")
    else:
        template.write("# Uncomment to resolve bug\n")
    legal_resolutions = bug.server.legal_values('resolution')
    if legal_resolutions:
        # Require non-empty resolution. DUPLICATE, MOVED would need special support
        legal_resolutions = [x for x in legal_resolutions if x not in ('', 'DUPLICATE', 'MOVED')]
        template.write("# possible resolutions: %s\n" % abbreviation_help_string(legal_resolutions))
    if not mark_resolved:
        template.write("#")
    template.write("Resolution: FIXED\n")

    if len(bug.patches) > 0:
        patches_have_status = any((patch.status is not None for patch in bug.patches))
        if patches_have_status:
            if len(newly_applied_patches) > 0 or len(obsoleted_patches) > 0:
                template.write("\n# Lines below change patch status, unless commented out\n")
            else:
                template.write("\n# To change patch status, uncomment below, edit 'committed' as appropriate.\n")
            legal_statuses = bug.server.legal_values('attachments.status')
            if legal_statuses:
                legal_statuses.append('obsolete')
                template.write("# possible statuses: %s\n" % abbreviation_help_string(legal_statuses))

            for patch in bug.patches:
                if patch in newly_applied_patches:
                    new_status = "committed"
                elif patch in obsoleted_patches:
                    new_status = "obsolete"
                else:
                    new_status = "#committed"
                template.write("%s @%d - %s - %s\n" % (new_status, patch.attach_id, patch.description, patch.status))
        else:
            template.write("\n# To mark patches obsolete, uncomment below\n")
            for patch in bug.patches:
                template.write("#obsolete @%d - %s\n" % (patch.attach_id, patch.description))

        template.write("\n")

    lines = edit_template(template.getvalue())

    def filter_line(line):
        m = re.match("^\s*Resolution\s*:\s*(\S+)", line)
        if m:
            resolutions.append(m.group(1))
            return False
        m = re.match("^\s*(\S+)\s*@\s*(\d+)", line)
        if m:
            status = m.group(1)
            changed_attachments[int(m.group(2))] = status
            return False
        return True

    changed_attachments = {}
    resolutions = []

    lines = filter(filter_line, lines)

    comment = "".join(lines).strip()
    resolution = resolutions[0] if len(resolutions) > 0 else None

    if resolution is None and len(changed_attachments) == 0 and comment == "":
        print "No changes, not editing Bug %d - %s" % (bug.id, bug.short_desc)
        return False

    if fix_commits is not None:
        if global_options.add_url:
            # We don't want to add the URLs until the user has decided not to
            # cancel the operation. But the comment that the user edited
            # included commit IDs. If adding the URL changes the commit IDs
            # we need to replace them in the comment.
            old_ids = [(commit, commit.id[0:7]) for commit in fix_commits]
            add_url(bug, fix_commits)
            for commit, old_id in old_ids:
                new_id = commit.id[0:7]
                if new_id != old_id:
                    comment = comment.replace(old_id, new_id)

    bug_changes = {}
    if resolution is not None:
        if legal_resolutions:
            try:
                resolution = expand_abbreviation(resolution, legal_resolutions)
            except ValueError:
                die("Bad resolution: %s" % resolution)
        bug_changes['bug_status'] = 'RESOLVED'
        bug_changes['resolution'] = resolution

    if comment != "":
        if len(bug_changes) == 0 and len(changed_attachments) == 1:
            # We can add the comment when we submit the attachment change.
            # Bugzilla will add a helpful notation ad we'll only send out
            # one set of email
            pass # We'll put the comment with the attachment
        else:
            bug_changes['comment'] = comment

    # If we did the attachment updates first, we'd have to fetch a new
    # token hash for the bug, since they'll change it. But each attachment
    # has an individual token hash for just that attachment, so we can
    # do the attachment updates afterwards.
    if len(bug_changes) > 0:
        bug.update(**bug_changes)

    for (attachment_id, status) in changed_attachments.iteritems():
        patch = None
        if patches_have_status:
            if legal_statuses:
                try:
                    status = expand_abbreviation(status, legal_statuses)
                except ValueError:
                    die("Bad patch status: %s" % status)
        else:
            if status != "obsolete":
                die("Can't mark patch as '%s'; only obsolete is supported on %s" % (status,
                                                                                    bug.server.host))
        for p in bug.patches:
            if p.attach_id == attachment_id:
                patch = p
        if not patch:
            die("%d is not a valid attachment ID for Bug %d" % (attachment_id, bug.id))
        attachment_changes = {}
        if comment != "" and not 'comment' in bug_changes: # See above
            attachment_changes['comment'] = comment
        if status == 'obsolete':
            attachment_changes['isobsolete'] = "1"
        else:
            attachment_changes['status'] = status
        bug.update_patch(patch, **attachment_changes)
        if status == 'obsolete':
            print "Marked attachment as obsolete: %s - %s " % (patch.attach_id, patch.description)
        else:
            print "Changed status of attachment to %s: %s - %s" % (status, patch.attach_id, patch.description)

    if fix_commits is not None:
        attach_commits(bug, fix_commits, status='committed')

    if resolution is not None:
        print "Resolved as %s bug %d - %s" % (resolution, bug.id, bug.short_desc)
    elif len(changed_attachments) > 0:
        print "Updated bug %d - %s" % (bug.id, bug.short_desc)
    else:
        print "Added comment to bug %d - %s" % (bug.id, bug.short_desc)
    print bug.get_url()

    return True

LOG_BUG_REFERENCE = re.compile(r"""
(\b[Ss]ee\s+(?:[^\s:/]+\s+){0,2})?
(?:(https?://[^/]+/show_bug.cgi\?id=[^&\s]+)
     |
   [Bb]ug\s+\#?(\d+))
""", re.VERBOSE | re.DOTALL)

def extract_bugs_from_string(str):
    refs = []
    for m in LOG_BUG_REFERENCE.finditer(str):
        bug_reference = None

        # If something says "See http://bugzilla.gnome.org/..." or
        # "See mozilla bug http://bugzilla.mozilla.org/..." or "see
        # bug 12345" - anything like that - then it's probably talking
        # about some peripherally related bug. So, if the word see
        # occurs 0 to 2 words before the bug reference, we ignore it.
        if m.group(1) is not None:
            print "Skipping cross-reference '%s'" % m.group(0)
            continue
        if m.group(2) is not None:
            bug_reference = m.group(2)
        else:
            bug_reference = m.group(3)

        try:
            yield BugHandle.parse(bug_reference)
        except BugParseError, e:
            print "WARNING: cannot resolve bug reference '%s'" % bug_reference

def extract_bugs_from_commit(commit):
    for handle in extract_bugs_from_string(commit.subject):
        yield handle
    for handle in extract_bugs_from_string(get_body(commit)):
        yield handle

# Yields bug, [<list of commits where it is referenced>] for each bug
# referenced in the list of commits. The order of bugs is the same as the
# order of their first reference in the list of commits
def extract_and_collate_bugs(commits):
    bugs = []
    bug_to_commits = {}

    for commit in commits:
        for handle in extract_bugs_from_commit(commit):
            if not handle in bug_to_commits:
                bugs.append(handle)
                bug_to_commits[handle] = []
            bug_to_commits[handle].append(commit)

    for bug in bugs:
        yield bug, bug_to_commits[bug]

def do_edit(bug_reference_or_revision_range):
    try:
        handle = BugHandle.parse(bug_reference_or_revision_range)
        if global_options.pushed:
            die("--pushed can't be used together with a bug reference")
        if global_options.fix is not None:
            die("--fix requires commits to be specified")
        bug = Bug.load(handle)
        edit_bug(bug)
    except BugParseError, e:
        try:
            commits = get_commits(bug_reference_or_revision_range)
        except CalledProcessError:
            die("'%s' isn't a valid bug reference or revision range" % bug_reference_or_revision_range)

        if global_options.fix is not None:
            handle = BugHandle.parse_or_die(global_options.fix)
            bug = Bug.load(handle)
            edit_bug(bug, fix_commits=commits)
        else:
            # Process from oldest to newest
            commits.reverse()
            for handle, commits in extract_and_collate_bugs(commits):
                bug = Bug.load(handle)
                if global_options.pushed:
                    edit_bug(bug, applied_commits=commits)
                else:
                    edit_bug(bug)

PRODUCT_COMPONENT_HELP = """

Use:

  git config bz.default-product <product>
  git config bz.default-component <component>

to configure a default product and/or component for this module."""

def do_file(*args):
    if len(args) == 1:
        product_component, commit_or_revision_range = None, args[0]
    else:
        product_component, commit_or_revision_range = args[0], args[1]

    config = get_config(get_tracker())

    if product_component:
        m = re.match("(?:([^/]+)/)?([^/]+)", product_component)
        if not m:
            die("'%s' is not a valid [<product>/]<component>" % product_component)

        product = m.group(1)
        component = m.group(2)

        if not product:
            product = get_default_product()

        if not product:
            die("'%s' does not specify a product and no default product is configured" % product_component
                + PRODUCT_COMPONENT_HELP)
    else:
        product = get_default_product()
        component = get_default_component()

        if not product:
            die("[<product>/]<component> not specified and no default product is configured"
                + PRODUCT_COMPONENT_HELP)
        if not component:
            die("[<product>/]<component> not specified and no default component is configured"
                + PRODUCT_COMPONENT_HELP)

    commits = get_commits(commit_or_revision_range)

    if global_options.add_url:
        check_add_url(commits, is_add_url=False)

    template = StringIO()
    if len(commits) == 1:
        template.write(commits[0].subject)
        template.write("\n")
    template.write("""
# Please enter the summary (first line) and description (other lines). Lines
# starting with '#' will be ignored.  Delete everything to abort.
#
# Product: %(product)s
# Component: %(component)s
# Patches to be attached:
""" % { 'product': product, 'component': component })
    for commit in reversed(commits):
        template.write("#   " + commit.id[0:7] + " " + commit.subject + "\n")

    lines = edit_template(template.getvalue())

    summary, description = split_subject_body(lines)

    if summary == "":
        die("Empty summary, aborting")

    # If we have only one patch and no other description for the bug was
    # specified, use the body of the commit as the the description for
    # the bug rather than the descriptionfor the attachment
    include_comments=True
    if len(commits) == 1:
        if description == "":
            description = get_body(commits[0])
            include_comments = False

    bug = Bug.create(get_tracker(), product, component, summary, description)

    if global_options.add_url:
        add_url(bug, commits)

    attach_commits(bug, commits, include_comments=include_comments)

def run_push(*args, **kwargs):
    # Predicting what 'git pushes' pushes based on the command line
    # would be extraordinarily complex, but the interactive output goes
    # to stderr and is somewhat ambiguous. We do the best we can parsing
    # it. git 1.6.4 adds --porcelain to push, so we can use that eventually.
    dry = kwargs['dry'] if 'dry' in kwargs else False
    options = dict()
    if dry:
        options['dry'] = True
    if global_options.force:
        options['force'] = True
    try:
        options['_return_stderr']=True
        out, err = git.push(*args, **options)
    except CalledProcessError:
        return
    if not dry:
        # Echo the output so the user gets feedback about what happened
        print >>sys.stderr, err

    commits = []
    for line in err.strip().split("\n"):
        #
        # We only look for updates of existing branches; a much more complex
        # handling would be look for all commits that weren't pushed to a
        # remote branch. Hopefully the typical use of 'git bz push' is pushing
        # a single commit to master.
        #
        #                 e5ad33e..febe0d4             master  ->   master
        m = re.match(r"^\s*([a-f0-9]{6,}..[a-f0-9]{6,})\s+\S+\s*->\s*\S+\s*$", line)
        if m:
            branch_commits = get_commits(m.group(1))
            # Process from oldest to newest
            branch_commits.reverse()
            commits += branch_commits

    # Remove duplicate commits
    seen_commit_ids = set()
    unique_commits = []
    for commit in commits:
        if not commit.id in seen_commit_ids:
            seen_commit_ids.add(commit.id)
            unique_commits.append(commit)

    return unique_commits

def do_push(*args):
    if global_options.fix:
        handle = BugHandle.parse_or_die(global_options.fix)
        bug = Bug.load(handle)

        # We need the user to confirm before we add the URLs to the commits
        # We need to add the URLs to the commits before we push
        # We need to push in order to find out what commits we are pushing
        # So, we push --dry first
        options = { 'dry' : True }
        commits = run_push(*args, **options)
        if edit_bug(bug, fix_commits=commits):
            run_push(*args)
    else:
        unique_commits = run_push(*args)
        for handle, commits in extract_and_collate_bugs(unique_commits):
            bug = Bug.load(handle)
            edit_bug(bug, commits)

################################################################################

if len(sys.argv) > 1:
    command = sys.argv[1]
else:
    command = ''

sys.argv[1:2] = []

parser = OptionParser()
parser.add_option("-b", "--bugzilla", metavar="<host or alias>",
                  help="bug tracker to use")

def add_add_url_options():
    parser.add_option("-u", "--add-url", action="store_true",
                      help="rewrite commits to add the bug URL [default]")
    parser.add_option("-n", "--no-add-url", action="store_false", dest="add_url",
                      help="don't rewrite commits to add the bug URL")

def add_edit_option():
    parser.add_option("-e", "--edit", action="store_true",
                      help="allow editing the bugzilla comment")

def add_fix_option():
    parser.add_option("", "--fix", metavar="<bug reference>",
                      help="attach commits and close bug")

if command == 'add-url':
    parser.set_usage("git bz add-url [options] <bug reference> (<commit> | <revision range>)");
    min_args = max_args = 2
elif command == 'apply':
    parser.set_usage("git bz apply [options] <bug reference>");
    add_add_url_options()
    min_args = max_args = 1
elif command == 'attach':
    parser.set_usage("git bz attach [options] [<bug reference>] (<commit> | <revision range>)");
    add_add_url_options()
    add_edit_option()
    min_args = 1
    max_args = 2
elif command == 'edit':
    parser.set_usage("git bz edit [options] (<bug reference> | <commit> | <revision range>)");
    parser.add_option("", "--pushed", action="store_true",
                      help="pre-fill edit form treating the commits as pushed")
    add_add_url_options()
    add_fix_option()
    min_args = max_args = 1
elif command == 'file':
    parser.set_usage("git bz file [options] [[<product>]]/<component>] (<commit> | <revision range>)");
    add_add_url_options()
    min_args = 1
    max_args = 2
elif command == 'push':
    parser.set_usage("git bz push [options] [<repository> <refspec>...]");
    add_add_url_options()
    add_fix_option()
    parser.add_option("-f", "--force", action="store_true",
                      help="allow non-fast-forward commits")
    min_args = 0
    max_args = 1000 # no max
else:
    print >>sys.stderr, "Usage: git bz [add-url|apply|attach|edit|file|push] [options]"
    sys.exit(1)

global_options, args = parser.parse_args()

if hasattr(global_options, 'add_url') and global_options.add_url is None:
    global_options.add_url = get_add_url()

if len(args) < min_args or len(args) > max_args:
    parser.print_usage()
    sys.exit(1)

if command == 'add-url':
    do_add_url(*args)
elif command == 'apply':
    do_apply(*args)
elif command == 'attach':
    do_attach(*args)
elif command == 'edit':
    if global_options.pushed:
        exit
    do_edit(*args)
elif command == 'file':
    do_file(*args)
elif command == 'push':
    do_push(*args)

sys.exit(0)