/upload-diffs.py

Large files files are truncated, but you can click here to view the full file

#!/usr/bin/env python
#
# Copyright 2007 Google Inc.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

"""Tool for uploading diffs from a version control system to the codereview app.

Usage summary: upload.py [options] [-- diff_options] [path...]

Diff options are passed to the diff command of the underlying system.

Supported version control systems:
  Git
  Mercurial
  Subversion

It is important for Git/Mercurial users to specify a tree/node/branch to diff
against by using the '--rev' option.
"""
# Taken from rietveld trunk (http://code.google.com/p/rietveld), r579

# This code is derived from appcfg.py in the App Engine SDK (open source),
# and from ASPN recipe #146306.

import ConfigParser
import cookielib
import fnmatch
import getpass
import logging
import mimetypes
import optparse
import os
import re
import socket
import subprocess
import sys
import urllib
import urllib2
import urlparse

# The md5 module was deprecated in Python 2.5.
try:
  from hashlib import md5
except ImportError:
  from md5 import md5

try:
  import readline
except ImportError:
  pass

try:
  import keyring
except ImportError:
  keyring = None

# Constants for GoogleCL.
DEFAULT_REVIEWER = 'tom.h.miller@gmail.com'
DEFAULT_CC = 'googlecl-dev@googlegroups.com'

# The logging verbosity:
#  0: Errors only.
#  1: Status messages.
#  2: Info logs.
#  3: Debug logs.
verbosity = 1

# The account type used for authentication.
# This line could be changed by the review server (see handler for
# upload.py).
AUTH_ACCOUNT_TYPE = "GOOGLE"

# URL of the default review server. As for AUTH_ACCOUNT_TYPE, this line could be
# changed by the review server (see handler for upload.py).
DEFAULT_REVIEW_SERVER = "codereview.appspot.com"

# Max size of patch or base file.
MAX_UPLOAD_SIZE = 900 * 1024

# Constants for version control names.  Used by GuessVCSName.
VCS_GIT = "Git"
VCS_MERCURIAL = "Mercurial"
VCS_SUBVERSION = "Subversion"
VCS_UNKNOWN = "Unknown"

# whitelist for non-binary filetypes which do not start with "text/"
# .mm (Objective-C) shows up as application/x-freemind on my Linux box.
TEXT_MIMETYPES = ['application/javascript', 'application/x-javascript',
                  'application/xml', 'application/x-freemind', 
                  'application/x-sh']

VCS_ABBREVIATIONS = {
  VCS_MERCURIAL.lower(): VCS_MERCURIAL,
  "hg": VCS_MERCURIAL,
  VCS_SUBVERSION.lower(): VCS_SUBVERSION,
  "svn": VCS_SUBVERSION,
  VCS_GIT.lower(): VCS_GIT,
}

# The result of parsing Subversion's [auto-props] setting.
svn_auto_props_map = None

def GetEmail(prompt):
  """Prompts the user for their email address and returns it.

  The last used email address is saved to a file and offered up as a suggestion
  to the user. If the user presses enter without typing in anything the last
  used email address is used. If the user enters a new address, it is saved
  for next time we prompt.

  """
  last_email_file_name = os.path.expanduser("~/.last_codereview_email_address")
  last_email = ""
  if os.path.exists(last_email_file_name):
    try:
      last_email_file = open(last_email_file_name, "r")
      last_email = last_email_file.readline().strip("\n")
      last_email_file.close()
      prompt += " [%s]" % last_email
    except IOError, e:
      pass
  email = raw_input(prompt + ": ").strip()
  if email:
    try:
      last_email_file = open(last_email_file_name, "w")
      last_email_file.write(email)
      last_email_file.close()
    except IOError, e:
      pass
  else:
    email = last_email
  return email


def StatusUpdate(msg):
  """Print a status message to stdout.

  If 'verbosity' is greater than 0, print the message.

  Args:
    msg: The string to print.
  """
  if verbosity > 0:
    print msg


def ErrorExit(msg):
  """Print an error message to stderr and exit."""
  print >>sys.stderr, msg
  sys.exit(1)


class ClientLoginError(urllib2.HTTPError):
  """Raised to indicate there was an error authenticating with ClientLogin."""

  def __init__(self, url, code, msg, headers, args):
    urllib2.HTTPError.__init__(self, url, code, msg, headers, None)
    self.args = args
    self.reason = args["Error"]


class AbstractRpcServer(object):
  """Provides a common interface for a simple RPC server."""

  def __init__(self, host, auth_function, host_override=None, extra_headers={},
               save_cookies=False, account_type=AUTH_ACCOUNT_TYPE):
    """Creates a new HttpRpcServer.

    Args:
      host: The host to send requests to.
      auth_function: A function that takes no arguments and returns an
        (email, password) tuple when called. Will be called if authentication
        is required.
      host_override: The host header to send to the server (defaults to host).
      extra_headers: A dict of extra headers to append to every request.
      save_cookies: If True, save the authentication cookies to local disk.
        If False, use an in-memory cookiejar instead.  Subclasses must
        implement this functionality.  Defaults to False.
      account_type: Account type used for authentication. Defaults to
        AUTH_ACCOUNT_TYPE.
    """
    self.host = host
    if (not self.host.startswith("http://") and
        not self.host.startswith("https://")):
      self.host = "http://" + self.host
    self.host_override = host_override
    self.auth_function = auth_function
    self.authenticated = False
    self.extra_headers = extra_headers
    self.save_cookies = save_cookies
    self.account_type = account_type
    self.opener = self._GetOpener()
    if self.host_override:
      logging.info("Server: %s; Host: %s", self.host, self.host_override)
    else:
      logging.info("Server: %s", self.host)

  def _GetOpener(self):
    """Returns an OpenerDirector for making HTTP requests.

    Returns:
      A urllib2.OpenerDirector object.
    """
    raise NotImplementedError()

  def _CreateRequest(self, url, data=None):
    """Creates a new urllib request."""
    logging.debug("Creating request for: '%s' with payload:\n%s", url, data)
    req = urllib2.Request(url, data=data)
    if self.host_override:
      req.add_header("Host", self.host_override)
    for key, value in self.extra_headers.iteritems():
      req.add_header(key, value)
    return req

  def _GetAuthToken(self, email, password):
    """Uses ClientLogin to authenticate the user, returning an auth token.

    Args:
      email:    The user's email address
      password: The user's password

    Raises:
      ClientLoginError: If there was an error authenticating with ClientLogin.
      HTTPError: If there was some other form of HTTP error.

    Returns:
      The authentication token returned by ClientLogin.
    """
    account_type = self.account_type
    if self.host.endswith(".google.com"):
      # Needed for use inside Google.
      account_type = "HOSTED"
    req = self._CreateRequest(
        url="https://www.google.com/accounts/ClientLogin",
        data=urllib.urlencode({
            "Email": email,
            "Passwd": password,
            "service": "ah",
            "source": "rietveld-codereview-upload",
            "accountType": account_type,
        }),
    )
    try:
      response = self.opener.open(req)
      response_body = response.read()
      response_dict = dict(x.split("=")
                           for x in response_body.split("\n") if x)
      return response_dict["Auth"]
    except urllib2.HTTPError, e:
      if e.code == 403:
        body = e.read()
        response_dict = dict(x.split("=", 1) for x in body.split("\n") if x)
        raise ClientLoginError(req.get_full_url(), e.code, e.msg,
                               e.headers, response_dict)
      else:
        raise

  def _GetAuthCookie(self, auth_token):
    """Fetches authentication cookies for an authentication token.

    Args:
      auth_token: The authentication token returned by ClientLogin.

    Raises:
      HTTPError: If there was an error fetching the authentication cookies.
    """
    # This is a dummy value to allow us to identify when we're successful.
    continue_location = "http://localhost/"
    args = {"continue": continue_location, "auth": auth_token}
    req = self._CreateRequest("%s/_ah/login?%s" %
                              (self.host, urllib.urlencode(args)))
    try:
      response = self.opener.open(req)
    except urllib2.HTTPError, e:
      response = e
    if (response.code != 302 or
        response.info()["location"] != continue_location):
      raise urllib2.HTTPError(req.get_full_url(), response.code, response.msg,
                              response.headers, response.fp)
    self.authenticated = True

  def _Authenticate(self):
    """Authenticates the user.

    The authentication process works as follows:
     1) We get a username and password from the user
     2) We use ClientLogin to obtain an AUTH token for the user
        (see http://code.google.com/apis/accounts/AuthForInstalledApps.html).
     3) We pass the auth token to /_ah/login on the server to obtain an
        authentication cookie. If login was successful, it tries to redirect
        us to the URL we provided.

    If we attempt to access the upload API without first obtaining an
    authentication cookie, it returns a 401 response (or a 302) and
    directs us to authenticate ourselves with ClientLogin.
    """
    for i in range(3):
      credentials = self.auth_function()
      try:
        auth_token = self._GetAuthToken(credentials[0], credentials[1])
      except ClientLoginError, e:
        if e.reason == "BadAuthentication":
          print >>sys.stderr, "Invalid username or password."
          continue
        if e.reason == "CaptchaRequired":
          print >>sys.stderr, (
              "Please go to\n"
              "https://www.google.com/accounts/DisplayUnlockCaptcha\n"
              "and verify you are a human.  Then try again.\n"
              "If you are using a Google Apps account the URL is:\n"
              "https://www.google.com/a/yourdomain.com/UnlockCaptcha")
          break
        if e.reason == "NotVerified":
          print >>sys.stderr, "Account not verified."
          break
        if e.reason == "TermsNotAgreed":
          print >>sys.stderr, "User has not agreed to TOS."
          break
        if e.reason == "AccountDeleted":
          print >>sys.stderr, "The user account has been deleted."
          break
        if e.reason == "AccountDisabled":
          print >>sys.stderr, "The user account has been disabled."
          break
        if e.reason == "ServiceDisabled":
          print >>sys.stderr, ("The user's access to the service has been "
                               "disabled.")
          break
        if e.reason == "ServiceUnavailable":
          print >>sys.stderr, "The service is not available; try again later."
          break
        raise
      self._GetAuthCookie(auth_token)
      return

  def Send(self, request_path, payload=None,
           content_type="application/octet-stream",
           timeout=None,
           extra_headers=None,
           **kwargs):
    """Sends an RPC and returns the response.

    Args:
      request_path: The path to send the request to, eg /api/appversion/create.
      payload: The body of the request, or None to send an empty request.
      content_type: The Content-Type header to use.
      timeout: timeout in seconds; default None i.e. no timeout.
        (Note: for large requests on OS X, the timeout doesn't work right.)
      extra_headers: Dict containing additional HTTP headers that should be
        included in the request (string header names mapped to their values),
        or None to not include any additional headers.
      kwargs: Any keyword arguments are converted into query string parameters.

    Returns:
      The response body, as a string.
    """
    # TODO: Don't require authentication.  Let the server say
    # whether it is necessary.
    if not self.authenticated:
      self._Authenticate()

    old_timeout = socket.getdefaulttimeout()
    socket.setdefaulttimeout(timeout)
    try:
      tries = 0
      while True:
        tries += 1
        args = dict(kwargs)
        url = "%s%s" % (self.host, request_path)
        if args:
          url += "?" + urllib.urlencode(args)
        req = self._CreateRequest(url=url, data=payload)
        req.add_header("Content-Type", content_type)
        if extra_headers:
          for header, value in extra_headers.items():
            req.add_header(header, value)
        try:
          f = self.opener.open(req)
          response = f.read()
          f.close()
          return response
        except urllib2.HTTPError, e:
          if tries > 3:
            raise
          elif e.code == 401 or e.code == 302:
            self._Authenticate()
##           elif e.code >= 500 and e.code < 600:
##             # Server Error - try again.
##             continue
          else:
            raise
    finally:
      socket.setdefaulttimeout(old_timeout)


class HttpRpcServer(AbstractRpcServer):
  """Provides a simplified RPC-style interface for HTTP requests."""

  def _Authenticate(self):
    """Save the cookie jar after authentication."""
    super(HttpRpcServer, self)._Authenticate()
    if self.save_cookies:
      StatusUpdate("Saving authentication cookies to %s" % self.cookie_file)
      self.cookie_jar.save()

  def _GetOpener(self):
    """Returns an OpenerDirector that supports cookies and ignores redirects.

    Returns:
      A urllib2.OpenerDirector object.
    """
    opener = urllib2.OpenerDirector()
    opener.add_handler(urllib2.ProxyHandler())
    opener.add_handler(urllib2.UnknownHandler())
    opener.add_handler(urllib2.HTTPHandler())
    opener.add_handler(urllib2.HTTPDefaultErrorHandler())
    opener.add_handler(urllib2.HTTPSHandler())
    opener.add_handler(urllib2.HTTPErrorProcessor())
    if self.save_cookies:
      self.cookie_file = os.path.expanduser("~/.codereview_upload_cookies")
      self.cookie_jar = cookielib.MozillaCookieJar(self.cookie_file)
      if os.path.exists(self.cookie_file):
        try:
          self.cookie_jar.load()
          self.authenticated = True
          StatusUpdate("Loaded authentication cookies from %s" %
                       self.cookie_file)
        except (cookielib.LoadError, IOError):
          # Failed to load cookies - just ignore them.
          pass
      else:
        # Create an empty cookie file with mode 600
        fd = os.open(self.cookie_file, os.O_CREAT, 0600)
        os.close(fd)
      # Always chmod the cookie file
      os.chmod(self.cookie_file, 0600)
    else:
      # Don't save cookies across runs of update.py.
      self.cookie_jar = cookielib.CookieJar()
    opener.add_handler(urllib2.HTTPCookieProcessor(self.cookie_jar))
    return opener


parser = optparse.OptionParser(
    usage="%prog [options] [-- diff_options] [path...]")
parser.add_option("-y", "--assume_yes", action="store_true",
                  dest="assume_yes", default=False,
                  help="Assume that the answer to yes/no questions is 'yes'.")
# Logging
group = parser.add_option_group("Logging options")
group.add_option("-q", "--quiet", action="store_const", const=0,
                 dest="verbose", help="Print errors only.")
group.add_option("-v", "--verbose", action="store_const", const=2,
                 dest="verbose", default=1,
                 help="Print info level logs.")
group.add_option("--noisy", action="store_const", const=3,
                 dest="verbose", help="Print all logs.")
# Review server
group = parser.add_option_group("Review server options")
group.add_option("-s", "--server", action="store", dest="server",
                 default=DEFAULT_REVIEW_SERVER,
                 metavar="SERVER",
                 help=("The server to upload to. The format is host[:port]. "
                       "Defaults to '%default'."))
group.add_option("-e", "--email", action="store", dest="email",
                 metavar="EMAIL", default=None,
                 help="The username to use. Will prompt if omitted.")
group.add_option("-H", "--host", action="store", dest="host",
                 metavar="HOST", default=None,
                 help="Overrides the Host header sent with all RPCs.")
group.add_option("--no_cookies", action="store_false",
                 dest="save_cookies", default=True,
                 help="Do not save authentication cookies to local disk.")
group.add_option("--account_type", action="store", dest="account_type",
                 metavar="TYPE", default=AUTH_ACCOUNT_TYPE,
                 choices=["GOOGLE", "HOSTED"],
                 help=("Override the default account type "
                       "(defaults to '%default', "
                       "valid choices are 'GOOGLE' and 'HOSTED')."))
# Issue
group = parser.add_option_group("Issue options")
group.add_option("-d", "--description", action="store", dest="description",
                 metavar="DESCRIPTION", default=None,
                 help="Optional description when creating an issue.")
group.add_option("-f", "--description_file", action="store",
                 dest="description_file", metavar="DESCRIPTION_FILE",
                 default=None,
                 help="Optional path of a file that contains "
                      "the description when creating an issue.")
group.add_option("-r", "--reviewers", action="store", dest="reviewers",
                 metavar="REVIEWERS", default=DEFAULT_REVIEWER,
                 help="Add reviewers (comma separated email addresses).")
group.add_option("--cc", action="store", dest="cc",
                 metavar="CC", default=DEFAULT_CC,
                 help="Add CC (comma separated email addresses).")
group.add_option("--private", action="store_true", dest="private",
                 default=False,
                 help="Make the issue restricted to reviewers and those CCed")
# Upload options
group = parser.add_option_group("Patch options")
group.add_option("-m", "--message", action="store", dest="message",
                 metavar="MESSAGE", default=None,
                 help="A message to identify the patch. "
                      "Will prompt if omitted.")
group.add_option("-i", "--issue", type="int", action="store",
                 metavar="ISSUE", default=None,
                 help="Issue number to which to add. Defaults to new issue.")
group.add_option("--base_url", action="store", dest="base_url", default=None,
                 help="Base repository URL (listed as \"Base URL\" when "
                 "viewing issue).  If omitted, will be guessed automatically "
                 "for SVN repos and left blank for others.")
group.add_option("--download_base", action="store_true",
                 dest="download_base", default=False,
                 help="Base files will be downloaded by the server "
                 "(side-by-side diffs may not work on files with CRs).")
group.add_option("--rev", action="store", dest="revision",
                 metavar="REV", default=None,
                 help="Base revision/branch/tree to diff against. Use "
                      "rev1:rev2 range to review already committed changeset.")
group.add_option("--send_mail", action="store_true",
                 dest="send_mail", default=False,
                 help="Send notification email to reviewers.")
group.add_option("--vcs", action="store", dest="vcs",
                 metavar="VCS", default=None,
                 help=("Version control system (optional, usually upload.py "
                       "already guesses the right VCS)."))
group.add_option("--emulate_svn_auto_props", action="store_true",
                 dest="emulate_svn_auto_props", default=False,
                 help=("Emulate Subversion's auto properties feature."))


def GetRpcServer(server, email=None, host_override=None, save_cookies=True,
                 account_type=AUTH_ACCOUNT_TYPE):
  """Returns an instance of an AbstractRpcServer.

  Args:
    server: String containing the review server URL.
    email: String containing user's email address.
    host_override: If not None, string containing an alternate hostname to use
      in the host header.
    save_cookies: Whether authentication cookies should be saved to disk.
    account_type: Account type for authentication, either 'GOOGLE'
      or 'HOSTED'. Defaults to AUTH_ACCOUNT_TYPE.

  Returns:
    A new AbstractRpcServer, on which RPC calls can be made.
  """

  rpc_server_class = HttpRpcServer

  # If this is the dev_appserver, use fake authentication.
  host = (host_override or server).lower()
  if host == "localhost" or host.startswith("localhost:"):
    if email is None:
      email = "test@example.com"
      logging.info("Using debug user %s.  Override with --email" % email)
    server = rpc_server_class(
        server,
        lambda: (email, "password"),
        host_override=host_override,
        extra_headers={"Cookie":
                       'dev_appserver_login="%s:False"' % email},
        save_cookies=save_cookies,
        account_type=account_type)
    # Don't try to talk to ClientLogin.
    server.authenticated = True
    return server

  def GetUserCredentials():
    """Prompts the user for a username and password."""
    # Create a local alias to the email variable to avoid Python's crazy
    # scoping rules.
    local_email = email
    if local_email is None:
      local_email = GetEmail("Email (login for uploading to %s)" % server)
    password = None
    if keyring:
      password = keyring.get_password(host, local_email)
    if password is not None:
      print "Using password from system keyring."
    else:
      password = getpass.getpass("Password for %s: " % local_email)
      if keyring:
        answer = raw_input("Store password in system keyring?(y/N) ").strip()
        if answer == "y":
          keyring.set_password(host, local_email, password)
    return (local_email, password)

  return rpc_server_class(server,
                          GetUserCredentials,
                          host_override=host_override,
                          save_cookies=save_cookies)


def EncodeMultipartFormData(fields, files):
  """Encode form fields for multipart/form-data.

  Args:
    fields: A sequence of (name, value) elements for regular form fields.
    files: A sequence of (name, filename, value) elements for data to be
           uploaded as files.
  Returns:
    (content_type, body) ready for httplib.HTTP instance.

  Source:
    http://aspn.activestate.com/ASPN/Cookbook/Python/Recipe/146306
  """
  BOUNDARY = '-M-A-G-I-C---B-O-U-N-D-A-R-Y-'
  CRLF = '\r\n'
  lines = []
  for (key, value) in fields:
    lines.append('--' + BOUNDARY)
    lines.append('Content-Disposition: form-data; name="%s"' % key)
    lines.append('')
    if isinstance(value, unicode):
      value = value.encode('utf-8')
    lines.append(value)
  for (key, filename, value) in files:
    lines.append('--' + BOUNDARY)
    lines.append('Content-Disposition: form-data; name="%s"; filename="%s"' %
             (key, filename))
    lines.append('Content-Type: %s' % GetContentType(filename))
    lines.append('')
    if isinstance(value, unicode):
      value = value.encode('utf-8')
    lines.append(value)
  lines.append('--' + BOUNDARY + '--')
  lines.append('')
  body = CRLF.join(lines)
  content_type = 'multipart/form-data; boundary=%s' % BOUNDARY
  return content_type, body


def GetContentType(filename):
  """Helper to guess the content-type from the filename."""
  return mimetypes.guess_type(filename)[0] or 'application/octet-stream'


# Use a shell for subcommands on Windows to get a PATH search.
use_shell = sys.platform.startswith("win")

def RunShellWithReturnCode(command, print_output=False,
                           universal_newlines=True,
                           env=os.environ):
  """Executes a command and returns the output from stdout and the return code.

  Args:
    command: Command to execute.
    print_output: If True, the output is printed to stdout.
                  If False, both stdout and stderr are ignored.
    universal_newlines: Use universal_newlines flag (default: True).

  Returns:
    Tuple (output, return code)
  """
  logging.info("Running %s", command)
  p = subprocess.Popen(command, stdout=subprocess.PIPE, stderr=subprocess.PIPE,
                       shell=use_shell, universal_newlines=universal_newlines,
                       env=env)
  if print_output:
    output_array = []
    while True:
      line = p.stdout.readline()
      if not line:
        break
      print line.strip("\n")
      output_array.append(line)
    output = "".join(output_array)
  else:
    output = p.stdout.read()
  p.wait()
  errout = p.stderr.read()
  if print_output and errout:
    print >>sys.stderr, errout
  p.stdout.close()
  p.stderr.close()
  return output, p.returncode


def RunShell(command, silent_ok=False, universal_newlines=True,
             print_output=False, env=os.environ):
  data, retcode = RunShellWithReturnCode(command, print_output,
                                         universal_newlines, env)
  if retcode:
    ErrorExit("Got error status from %s:\n%s" % (command, data))
  if not silent_ok and not data:
    ErrorExit("No output from %s" % command)
  return data


class VersionControlSystem(object):
  """Abstract base class providing an interface to the VCS."""

  def __init__(self, options):
    """Constructor.

    Args:
      options: Command line options.
    """
    self.options = options

  def PostProcessDiff(self, diff):
    """Return the diff with any special post processing this VCS needs, e.g.
    to include an svn-style "Index:"."""
    return diff

  def GenerateDiff(self, args):
    """Return the current diff as a string.

    Args:
      args: Extra arguments to pass to the diff command.
    """
    raise NotImplementedError(
        "abstract method -- subclass %s must override" % self.__class__)

  def GetUnknownFiles(self):
    """Return a list of files unknown to the VCS."""
    raise NotImplementedError(
        "abstract method -- subclass %s must override" % self.__class__)

  def CheckForUnknownFiles(self):
    """Show an "are you sure?" prompt if there are unknown files."""
    unknown_files = self.GetUnknownFiles()
    if unknown_files:
      print "The following files are not added to version control:"
      for line in unknown_files:
        print line
      prompt = "Are you sure to continue?(y/N) "
      answer = raw_input(prompt).strip()
      if answer != "y":
        ErrorExit("User aborted")

  def GetBaseFile(self, filename):
    """Get the content of the upstream version of a file.

    Returns:
      A tuple (base_content, new_content, is_binary, status)
        base_content: The contents of the base file.
        new_content: For text files, this is empty.  For binary files, this is
          the contents of the new file, since the diff output won't contain
          information to reconstruct the current file.
        is_binary: True iff the file is binary.
        status: The status of the file.
    """

    raise NotImplementedError(
        "abstract method -- subclass %s must override" % self.__class__)


  def GetBaseFiles(self, diff):
    """Helper that calls GetBase file for each file in the patch.

    Returns:
      A dictionary that maps from filename to GetBaseFile's tuple.  Filenames
      are retrieved based on lines that start with "Index:" or
      "Property changes on:".
    """
    files = {}
    for line in diff.splitlines(True):
      if line.startswith('Index:') or line.startswith('Property changes on:'):
        unused, filename = line.split(':', 1)
        # On Windows if a file has property changes its filename uses '\'
        # instead of '/'.
        filename = filename.strip().replace('\\', '/')
        files[filename] = self.GetBaseFile(filename)
    return files


  def UploadBaseFiles(self, issue, rpc_server, patch_list, patchset, options,
                      files):
    """Uploads the base files (and if necessary, the current ones as well)."""

    def UploadFile(filename, file_id, content, is_binary, status, is_base):
      """Uploads a file to the server."""
      file_too_large = False
      if is_base:
        type = "base"
      else:
        type = "current"
      if len(content) > MAX_UPLOAD_SIZE:
        print ("Not uploading the %s file for %s because it's too large." %
               (type, filename))
        file_too_large = True
        content = ""
      checksum = md5(content).hexdigest()
      if options.verbose > 0 and not file_too_large:
        print "Uploading %s file for %s" % (type, filename)
      url = "/%d/upload_content/%d/%d" % (int(issue), int(patchset), file_id)
      form_fields = [("filename", filename),
                     ("status", status),
                     ("checksum", checksum),
                     ("is_binary", str(is_binary)),
                     ("is_current", str(not is_base)),
                    ]
      if file_too_large:
        form_fields.append(("file_too_large", "1"))
      if options.email:
        form_fields.append(("user", options.email))
      ctype, body = EncodeMultipartFormData(form_fields,
                                            [("data", filename, content)])
      response_body = rpc_server.Send(url, body,
                                      content_type=ctype)
      if not response_body.startswith("OK"):
        StatusUpdate("  --> %s" % response_body)
        sys.exit(1)

    patches = dict()
    [patches.setdefault(v, k) for k, v in patch_list]
    for filename in patches.keys():
      base_content, new_content, is_binary, status = files[filename]
      file_id_str = patches.get(filename)
      if file_id_str.find("nobase") != -1:
        base_content = None
        file_id_str = file_id_str[file_id_str.rfind("_") + 1:]
      file_id = int(file_id_str)
      if base_content != None:
        UploadFile(filename, file_id, base_content, is_binary, status, True)
      if new_content != None:
        UploadFile(filename, file_id, new_content, is_binary, status, False)

  def IsImage(self, filename):
    """Returns true if the filename has an image extension."""
    mimetype =  mimetypes.guess_type(filename)[0]
    if not mimetype:
      return False
    return mimetype.startswith("image/")

  def IsBinary(self, filename):
    """Returns true if the guessed mimetyped isnt't in text group."""
    mimetype = mimetypes.guess_type(filename)[0]
    if not mimetype:
      return False  # e.g. README, "real" binaries usually have an extension
    # special case for text files which don't start with text/
    if mimetype in TEXT_MIMETYPES:
      return False
    return not mimetype.startswith("text/")


class SubversionVCS(VersionControlSystem):
  """Implementation of the VersionControlSystem interface for Subversion."""

  def __init__(self, options):
    super(SubversionVCS, self).__init__(options)
    if self.options.revision:
      match = re.match(r"(\d+)(:(\d+))?", self.options.revision)
      if not match:
        ErrorExit("Invalid Subversion revision %s." % self.options.revision)
      self.rev_start = match.group(1)
      self.rev_end = match.group(3)
    else:
      self.rev_start = self.rev_end = None
    # Cache output from "svn list -r REVNO dirname".
    # Keys: dirname, Values: 2-tuple (ouput for start rev and end rev).
    self.svnls_cache = {}
    # Base URL is required to fetch files deleted in an older revision.
    # Result is cached to not guess it over and over again in GetBaseFile().
    required = self.options.download_base or self.options.revision is not None
    self.svn_base = self._GuessBase(required)

  def GuessBase(self, required):
    """Wrapper for _GuessBase."""
    return self.svn_base

  def _GuessBase(self, required):
    """Returns the SVN base URL.

    Args:
      required: If true, exits if the url can't be guessed, otherwise None is
        returned.
    """
    info = RunShell(["svn", "info"])
    for line in info.splitlines():
      words = line.split()
      if len(words) == 2 and words[0] == "URL:":
        url = words[1]
        scheme, netloc, path, params, query, fragment = urlparse.urlparse(url)
        username, netloc = urllib.splituser(netloc)
        if username:
          logging.info("Removed username from base URL")
        if netloc.endswith("svn.python.org"):
          if netloc == "svn.python.org":
            if path.startswith("/projects/"):
              path = path[9:]
          elif netloc != "pythondev@svn.python.org":
            ErrorExit("Unrecognized Python URL: %s" % url)
          base = "http://svn.python.org/view/*checkout*%s/" % path
          logging.info("Guessed Python base = %s", base)
        elif netloc.endswith("svn.collab.net"):
          if path.startswith("/repos/"):
            path = path[6:]
          base = "http://svn.collab.net/viewvc/*checkout*%s/" % path
          logging.info("Guessed CollabNet base = %s", base)
        elif netloc.endswith(".googlecode.com"):
          path = path + "/"
          base = urlparse.urlunparse(("http", netloc, path, params,
                                      query, fragment))
          logging.info("Guessed Google Code base = %s", base)
        else:
          path = path + "/"
          base = urlparse.urlunparse((scheme, netloc, path, params,
                                      query, fragment))
          logging.info("Guessed base = %s", base)
        return base
    if required:
      ErrorExit("Can't find URL in output from svn info")
    return None

  def GenerateDiff(self, args):
    cmd = ["svn", "diff"]
    if self.options.revision:
      cmd += ["-r", self.options.revision]
    cmd.extend(args)
    data = RunShell(cmd)
    count = 0
    for line in data.splitlines():
      if line.startswith("Index:") or line.startswith("Property changes on:"):
        count += 1
        logging.info(line)
    if not count:
      ErrorExit("No valid patches found in output from svn diff")
    return data

  def _CollapseKeywords(self, content, keyword_str):
    """Collapses SVN keywords."""
    # svn cat translates keywords but svn diff doesn't. As a result of this
    # behavior patching.PatchChunks() fails with a chunk mismatch error.
    # This part was originally written by the Review Board development team
    # who had the same problem (http://reviews.review-board.org/r/276/).
    # Mapping of keywords to known aliases
    svn_keywords = {
      # Standard keywords
      'Date':                ['Date', 'LastChangedDate'],
      'Revision':            ['Revision', 'LastChangedRevision', 'Rev'],
      'Author':              ['Author', 'LastChangedBy'],
      'HeadURL':             ['HeadURL', 'URL'],
      'Id':                  ['Id'],

      # Aliases
      'LastChangedDate':     ['LastChangedDate', 'Date'],
      'LastChangedRevision': ['LastChangedRevision', 'Rev', 'Revision'],
      'LastChangedBy':       ['LastChangedBy', 'Author'],
      'URL':                 ['URL', 'HeadURL'],
    }

    def repl(m):
       if m.group(2):
         return "$%s::%s$" % (m.group(1), " " * len(m.group(3)))
       return "$%s$" % m.group(1)
    keywords = [keyword
                for name in keyword_str.split(" ")
                for keyword in svn_keywords.get(name, [])]
    return re.sub(r"\$(%s):(:?)([^\$]+)\$" % '|'.join(keywords), repl, content)

  def GetUnknownFiles(self):
    status = RunShell(["svn", "status", "--ignore-externals"], silent_ok=True)
    unknown_files = []
    for line in status.split("\n"):
      if line and line[0] == "?":
        unknown_files.append(line)
    return unknown_files

  def ReadFile(self, filename):
    """Returns the contents of a file."""
    file = open(filename, 'rb')
    result = ""
    try:
      result = file.read()
    finally:
      file.close()
    return result

  def GetStatus(self, filename):
    """Returns the status of a file."""
    if not self.options.revision:
      status = RunShell(["svn", "status", "--ignore-externals", filename])
      if not status:
        ErrorExit("svn status returned no output for %s" % filename)
      status_lines = status.splitlines()
      # If file is in a cl, the output will begin with
      # "\n--- Changelist 'cl_name':\n".  See
      # http://svn.collab.net/repos/svn/trunk/notes/changelist-design.txt
      if (len(status_lines) == 3 and
          not status_lines[0] and
          status_lines[1].startswith("--- Changelist")):
        status = status_lines[2]
      else:
        status = status_lines[0]
    # If we have a revision to diff against we need to run "svn list"
    # for the old and the new revision and compare the results to get
    # the correct status for a file.
    else:
      dirname, relfilename = os.path.split(filename)
      if dirname not in self.svnls_cache:
        cmd = ["svn", "list", "-r", self.rev_start, dirname or "."]
        out, returncode = RunShellWithReturnCode(cmd)
        if returncode:
          ErrorExit("Failed to get status for %s." % filename)
        old_files = out.splitlines()
        args = ["svn", "list"]
        if self.rev_end:
          args += ["-r", self.rev_end]
        cmd = args + [dirname or "."]
        out, returncode = RunShellWithReturnCode(cmd)
        if returncode:
          ErrorExit("Failed to run command %s" % cmd)
        self.svnls_cache[dirname] = (old_files, out.splitlines())
      old_files, new_files = self.svnls_cache[dirname]
      if relfilename in old_files and relfilename not in new_files:
        status = "D   "
      elif relfilename in old_files and relfilename in new_files:
        status = "M   "
      else:
        status = "A   "
    return status

  def GetBaseFile(self, filename):
    status = self.GetStatus(filename)
    base_content = None
    new_content = None

    # If a file is copied its status will be "A  +", which signifies
    # "addition-with-history".  See "svn st" for more information.  We need to
    # upload the original file or else diff parsing will fail if the file was
    # edited.
    if status[0] == "A" and status[3] != "+":
      # We'll need to upload the new content if we're adding a binary file
      # since diff's output won't contain it.
      mimetype = RunShell(["svn", "propget", "svn:mime-type", filename],
                          silent_ok=True)
      base_content = ""
      is_binary = bool(mimetype) and not mimetype.startswith("text/")
      if is_binary and self.IsImage(filename):
        new_content = self.ReadFile(filename)
    elif (status[0] in ("M", "D", "R") or
          (status[0] == "A" and status[3] == "+") or  # Copied file.
          (status[0] == " " and status[1] == "M")):  # Property change.
      args = []
      if self.options.revision:
        url = "%s/%s@%s" % (self.svn_base, filename, self.rev_start)
      else:
        # Don't change filename, it's needed later.
        url = filename
        args += ["-r", "BASE"]
      cmd = ["svn"] + args + ["propget", "svn:mime-type", url]
      mimetype, returncode = RunShellWithReturnCode(cmd)
      if returncode:
        # File does not exist in the requested revision.
        # Reset mimetype, it contains an error message.
        mimetype = ""
      get_base = False
      is_binary = bool(mimetype) and not mimetype.startswith("text/")
      if status[0] == " ":
        # Empty base content just to force an upload.
        base_content = ""
      elif is_binary:
        if self.IsImage(filename):
          get_base = True
          if status[0] == "M":
            if not self.rev_end:
              new_content = self.ReadFile(filename)
            else:
              url = "%s/%s@%s" % (self.svn_base, filename, self.rev_end)
              new_content = RunShell(["svn", "cat", url],
                                     universal_newlines=True, silent_ok=True)
        else:
          base_content = ""
      else:
        get_base = True

      if get_base:
        if is_binary:
          universal_newlines = False
        else:
          universal_newlines = True
        if self.rev_start:
          # "svn cat -r REV delete_file.txt" doesn't work. cat requires
          # the full URL with "@REV" appended instead of using "-r" option.
          url = "%s/%s@%s" % (self.svn_base, filename, self.rev_start)
          base_content = RunShell(["svn", "cat", url],
                                  universal_newlines=universal_newlines,
                                  silent_ok=True)
        else:
          base_content, ret_code = RunShellWithReturnCode(
            ["svn", "cat", filename], universal_newlines=universal_newlines)
          if ret_code and status[0] == "R":
            # It's a replaced file without local history (see issue208).
            # The base file needs to be fetched from the server.
            url = "%s/%s" % (self.svn_base, filename)
            base_content = RunShell(["svn", "cat", url],
                                    universal_newlines=universal_newlines,
                                    silent_ok=True)
          elif ret_code:
            ErrorExit("Got error status from 'svn cat %s'" % filename)
        if not is_binary:
          args = []
          if self.rev_start:
            url = "%s/%s@%s" % (self.svn_base, filename, self.rev_start)
          else:
            url = filename
            args += ["-r", "BASE"]
          cmd = ["svn"] + args + ["propget", "svn:keywords", url]
          keywords, returncode = RunShellWithReturnCode(cmd)
          if keywords and not returncode:
            base_content = self._CollapseKeywords(base_content, keywords)
    else:
      StatusUpdate("svn status returned unexpected output: %s" % status)
      sys.exit(1)
    return base_content, new_content, is_binary, status[0:5]


class GitVCS(VersionControlSystem):
  """Implementation of the VersionControlSystem interface for Git."""

  def __init__(self, options):
    super(GitVCS, self).__init__(options)
    # Map of filename -> (hash before, hash after) of base file.
    # Hashes for "no such file" are represented as None.
    self.hashes = {}
    # Map of new filename -> old filename for renames.
    self.renames = {}

  def PostProcessDiff(self, gitdiff):
    """Converts the diff output to include an svn-style "Index:" line as well
    as record the hashes of the files, so we can upload them along with our
    diff."""
    # Special used by git to indicate "no such content".
    NULL_HASH = "0"*40

    def IsFileNew(filename):
      return filename in self.hashes and self.hashes[filename][0] is None

    def AddSubversionPropertyChange(filename):
      """Add svn's property change information into the patch if given file is
      new file.

      We use Subversion's auto-props setting to retrieve its property.
      See http://svnbook.red-bean.com/en/1.1/ch07.html#svn-ch-7-sect-1.3.2 for
      Subversion's [auto-props] setting.
      """
      if self.options.emulate_svn_auto_props and IsFileNew(filename):
        svnprops = GetSubversionPropertyChanges(filename)
        if svnprops:
          svndiff.append("\n" + svnprops + "\n")

    svndiff = []
    filecount = 0
    filename = None
    for line in gitdiff.splitlines():
      match = re.match(r"diff --git a/(.*) b/(.*)$", line)
      if match:
        # Add auto property here for previously seen file.
        if filename is not None:
          AddSubversionPropertyChange(filename)
        filecount += 1
        # Intentionally use the "after" filename so we can show renames.
        filename = match.group(2)
        svndiff.append("Index: %s\n" % filename)
        if match.group(1) != match.group(2):
          self.renames[match.group(2)] = match.group(1)
      else:
        # The "index" line in a git diff looks like this (long hashes elided):
        #   index 82c0d44..b2cee3f 100755
        # We want to save the left hash, as that identifies the base file.
        match = re.match(r"index (\w+)\.\.(\w+)", line)
        if match:
          before, after = (match.group(1), match.group(2))
          if before == NULL_HASH:
            before = None
          if after == NULL_HASH:
            after = None
          self.hashes[filename] = (before, after)
      svndiff.append(line + "\n")
    if not filecount:
      ErrorExit("No valid patches found in output from git diff")
    # Add auto property for the last seen file.
    assert filename is not None
    AddSubversionPropertyChange(filename)
    return "".join(svndiff)

  def GenerateDiff(self, extra_args):
    extra_args = extra_args[:]
    if self.options.revision:
      if ":" in self.options.revision:
        extra_args = self.options.revision.split(":", 1) + extra_args
      else:
        extra_args = [self.options.revision] + extra_args

    # --no-ext-diff is broken in some versions of Git, so try to work around
    # this by overriding the environment (but there is still a problem if the
    # git config key "diff.external" is used).
    env = os.environ.copy()
    if 'GIT_EXTERNAL_DIFF' in env: del env['GIT_EXTERNAL_DIFF']
    return RunShell(["git", "diff", "--no-ext-diff", "--full-index", "-M"]
                    + extra_args, env=env)

  def GetUnknownFiles(self):
    status = RunShell(["git", "ls-files", "--exclude-standard", "--others"],
                      silent_ok=True)
    return status.splitlines()

  def GetFileContent(self, file_hash, is_binary):
    """Returns the content of a file identified by its git hash."""
    data, retcode = RunShellWithReturnCode(["git", "show", file_hash],
                                            universal_newlines=not is_binary)
    if retcode:
      ErrorExit("Got error status from 'git show %s'" % file_hash)
    return data

  def GetBaseFile(self, filename):
    hash_before, hash_after = self.hashes.get(filename, (None,None))
    base_content = None
    new_content = None
    is_binary = self.IsBinary(filename)
    status = None

    if filename in self.renames:
      status = "A +"  # Match svn attribute name for renames.
      if filename not in self.hashes:
        # If a rename doesn't change the content, we never get a hash.
        base_content = RunShell(["git", "show", "HEAD:" + filename])
    elif not hash_before:
      status = "A"
      base_content = ""
    elif not hash_after:
      status = "D"
    else:
      status = "M"

    is_image = self.IsImage(filename)

    # Grab the before/after content if we need it.
    # We should include file contents if it's text or it's an image.
    if not is_binary or is_image:
      # Grab the base content if we don't have it already.
      if base_content is None and hash_before:
        base_content = self.GetFileContent(hash_before, is_binary)
      # Only include the "after" file if it's an image; otherwise it
      # it is reconstructed from the diff.
      if is_image and hash_after:
        new_content = self.GetFileContent(hash_after, is_binary)

    return (base_content, new_content, is_binary, status)


class MercurialVCS(VersionControlSystem):
  """Implementation of the VersionControlSystem interface for Mercurial."""

  def __init__(self, options, repo_dir):
    super(MercurialVCS, self).__init__(options)
    # Absolute path to repository (we can be in a subdir)
    self.repo_dir = os.path.normpath(repo_dir)
    # Compute the subdir
    cwd = os.path.normpath(os.getcwd())
    assert cwd.startswith(self.repo_dir)
    self.subdir = cwd[len(self.repo_dir):].lstrip(r"\/")
    if self.options.revision:
      self.base_rev = self.options.revision
    else:
      self.base_rev = RunShell(["hg", "parent", "-q"]).split(':')[1].strip()

  def _GetRelPath(self, filename):
    """Get relative path of a file according to the current directory,
    given its logical path in the repo."""
    assert filename.startswith(self.subdir), (filename, self.subdir)
    return filename[len(self.subdir):].lstrip(r"\/")

  def GenerateDiff(self, extra_args):
    cmd = ["hg", "diff", "--git", "-r", self.base_rev] + extra_args
    data = RunShell(cmd, silent_ok=True)
    svndiff = []
    filecount = 0
    for line in data.splitlines():
      m = re.match("diff --git a/(\S+) b/(\S+)", line)
      if m:
        # Modify line to make it look like as it comes from svn diff.
        # With this modification no changes on the server side are required
        # to make upload.py work with Mercurial repos.
        # NOTE: for proper handling of moved/copied files, we have to use
        # the second filename.
        filename = m.group(2)
        svndiff.append("Index: %s" % filename)
        svndiff.append("=" * 67)
        filecount += 1
        logging.info(line)
      else:
        svndiff.append(line)
    if not filecount:
      ErrorExit("No valid patches found in output from hg diff")
    return "\n".join(svndiff) + "\n"

  def GetUnknownFiles(self):
    """Return a list of files unknown to the VCS."""
    args = []
    status = RunShell(["hg", "status", "--rev", self.base_rev, "-u", "."],
        silent_ok=True)
    unknown_files = []
    for line in status.splitlines():
      st, fn = line.split(" ", 1)
      if st == "?":
        unknown_files.append(fn)
    return unknown_files

  def GetBaseFile(self, filename):
    # "hg status" and "hg cat" both take a path relative to the current subdir
    # rather than to the repo root, but "hg diff" has given us the full path
    # to the repo root.
    base_content = ""
    new_content = None
    is_binary = False
    oldrelpath = relpath = self._GetRelPath(filename)
    # "hg status -…