httpclient: update to 938f2107d6e2 of httpplus

This enhances proxy support in httpclient a little bit, though I don't
know that we used that functionality at all. It also switches httpplus
to using absolute_import.

mercurial/httpclient/__init__.py

@@ -1,731 +1,768
 # Copyright 2010, Google Inc.
 # All rights reserved.
 #
 # Redistribution and use in source and binary forms, with or without
 # modification, are permitted provided that the following conditions are
 # met:
 #
 #     * Redistributions of source code must retain the above copyright
 # notice, this list of conditions and the following disclaimer.
 #     * Redistributions in binary form must reproduce the above
 # copyright notice, this list of conditions and the following disclaimer
 # in the documentation and/or other materials provided with the
 # distribution.
 #     * Neither the name of Google Inc. nor the names of its
 # contributors may be used to endorse or promote products derived from
 # this software without specific prior written permission.
 
 # THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 # "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
 # LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
 # A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
 # OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
 # SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
 # LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
 # DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
 # THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 # (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 """Improved HTTP/1.1 client library
 
 This library contains an HTTPConnection which is similar to the one in
 httplib, but has several additional features:
 
   * supports keepalives natively
   * uses select() to block for incoming data
   * notices when the server responds early to a request
   * implements ssl inline instead of in a different class
 """
+from __future__ import absolute_import
 
 # Many functions in this file have too many arguments.
 # pylint: disable=R0913
 
 import cStringIO
 import errno
 import httplib
 import logging
 import rfc822
 import select
 import socket
 
-import _readers
-import socketutil
+from . import (
+    _readers,
+    socketutil,
+    )
 
 logger = logging.getLogger(__name__)
 
 __all__ = ['HTTPConnection', 'HTTPResponse']
 
 HTTP_VER_1_0 = 'HTTP/1.0'
 HTTP_VER_1_1 = 'HTTP/1.1'
 
 OUTGOING_BUFFER_SIZE = 1 << 15
 INCOMING_BUFFER_SIZE = 1 << 20
 
 HDR_ACCEPT_ENCODING = 'accept-encoding'
 HDR_CONNECTION_CTRL = 'connection'
 HDR_CONTENT_LENGTH = 'content-length'
 HDR_XFER_ENCODING = 'transfer-encoding'
 
 XFER_ENCODING_CHUNKED = 'chunked'
 
 CONNECTION_CLOSE = 'close'
 
 EOL = '\r\n'
 _END_HEADERS = EOL * 2
 
 # Based on some searching around, 1 second seems like a reasonable
 # default here.
 TIMEOUT_ASSUME_CONTINUE = 1
 TIMEOUT_DEFAULT = None
 
 
 class HTTPResponse(object):
     """Response from an HTTP server.
 
     The response will continue to load as available. If you need the
     complete response before continuing, check the .complete() method.
     """
     def __init__(self, sock, timeout, method):
         self.sock = sock
         self.method = method
         self.raw_response = ''
         self._headers_len = 0
         self.headers = None
         self.will_close = False
         self.status_line = ''
         self.status = None
         self.continued = False
         self.http_version = None
         self.reason = None
         self._reader = None
 
         self._read_location = 0
         self._eol = EOL
 
         self._timeout = timeout
 
     @property
     def _end_headers(self):
         return self._eol * 2
 
     def complete(self):
         """Returns true if this response is completely loaded.
 
         Note that if this is a connection where complete means the
         socket is closed, this will nearly always return False, even
         in cases where all the data has actually been loaded.
         """
         if self._reader:
             return self._reader.done()
 
     def _close(self):
         if self._reader is not None:
             # We're a friend of the reader class here.
             # pylint: disable=W0212
             self._reader._close()
 
+    def getheader(self, header, default=None):
+        return self.headers.getheader(header, default=default)
+
+    def getheaders(self):
+        return self.headers.items()
+
     def readline(self):
         """Read a single line from the response body.
 
         This may block until either a line ending is found or the
         response is complete.
         """
         blocks = []
         while True:
             self._reader.readto('\n', blocks)
 
             if blocks and blocks[-1][-1] == '\n' or self.complete():
                 break
 
             self._select()
 
         return ''.join(blocks)
 
     def read(self, length=None):
         """Read data from the response body."""
         # if length is None, unbounded read
         while (not self.complete()  # never select on a finished read
                and (not length  # unbounded, so we wait for complete()
                     or length > self._reader.available_data)):
             self._select()
         if not length:
             length = self._reader.available_data
         r = self._reader.read(length)
         if self.complete() and self.will_close:
             self.sock.close()
         return r
 
     def _select(self):
         r, unused_write, unused_err = select.select(
             [self.sock], [], [], self._timeout)
         if not r:
             # socket was not readable. If the response is not
             # complete, raise a timeout.
             if not self.complete():
                 logger.info('timed out with timeout of %s', self._timeout)
                 raise HTTPTimeoutException('timeout reading data')
         try:
             data = self.sock.recv(INCOMING_BUFFER_SIZE)
         except socket.sslerror as e:
             if e.args[0] != socket.SSL_ERROR_WANT_READ:
                 raise
             logger.debug('SSL_ERROR_WANT_READ in _select, should retry later')
             return True
         logger.debug('response read %d data during _select', len(data))
         # If the socket was readable and no data was read, that means
         # the socket was closed. Inform the reader (if any) so it can
         # raise an exception if this is an invalid situation.
         if not data:
             if self._reader:
                 # We're a friend of the reader class here.
                 # pylint: disable=W0212
                 self._reader._close()
             return False
         else:
             self._load_response(data)
             return True
 
     # This method gets replaced by _load later, which confuses pylint.
     def _load_response(self, data): # pylint: disable=E0202
         # Being here implies we're not at the end of the headers yet,
         # since at the end of this method if headers were completely
         # loaded we replace this method with the load() method of the
         # reader we created.
         self.raw_response += data
         # This is a bogus server with bad line endings
         if self._eol not in self.raw_response:
             for bad_eol in ('\n', '\r'):
                 if (bad_eol in self.raw_response
                     # verify that bad_eol is not the end of the incoming data
                     # as this could be a response line that just got
                     # split between \r and \n.
                     and (self.raw_response.index(bad_eol) <
                          (len(self.raw_response) - 1))):
                     logger.info('bogus line endings detected, '
                                 'using %r for EOL', bad_eol)
                     self._eol = bad_eol
                     break
         # exit early if not at end of headers
         if self._end_headers not in self.raw_response or self.headers:
             return
 
         # handle 100-continue response
         hdrs, body = self.raw_response.split(self._end_headers, 1)
         unused_http_ver, status = hdrs.split(' ', 1)
         if status.startswith('100'):
             self.raw_response = body
             self.continued = True
             logger.debug('continue seen, setting body to %r', body)
             return
 
         # arriving here means we should parse response headers
         # as all headers have arrived completely
         hdrs, body = self.raw_response.split(self._end_headers, 1)
         del self.raw_response
         if self._eol in hdrs:
             self.status_line, hdrs = hdrs.split(self._eol, 1)
         else:
             self.status_line = hdrs
             hdrs = ''
         # TODO HTTP < 1.0 support
         (self.http_version, self.status,
          self.reason) = self.status_line.split(' ', 2)
         self.status = int(self.status)
         if self._eol != EOL:
             hdrs = hdrs.replace(self._eol, '\r\n')
         headers = rfc822.Message(cStringIO.StringIO(hdrs))
         content_len = None
         if HDR_CONTENT_LENGTH in headers:
             content_len = int(headers[HDR_CONTENT_LENGTH])
         if self.http_version == HTTP_VER_1_0:
             self.will_close = True
         elif HDR_CONNECTION_CTRL in headers:
             self.will_close = (
                 headers[HDR_CONNECTION_CTRL].lower() == CONNECTION_CLOSE)
         if (HDR_XFER_ENCODING in headers
             and headers[HDR_XFER_ENCODING].lower() == XFER_ENCODING_CHUNKED):
             self._reader = _readers.ChunkedReader(self._eol)
             logger.debug('using a chunked reader')
         else:
             # HEAD responses are forbidden from returning a body, and
             # it's implausible for a CONNECT response to use
             # close-is-end logic for an OK response.
             if (self.method == 'HEAD' or
                 (self.method == 'CONNECT' and content_len is None)):
                 content_len = 0
             if content_len is not None:
                 logger.debug('using a content-length reader with length %d',
                              content_len)
                 self._reader = _readers.ContentLengthReader(content_len)
             else:
                 # Response body had no length specified and is not
                 # chunked, so the end of the body will only be
                 # identifiable by the termination of the socket by the
                 # server. My interpretation of the spec means that we
                 # are correct in hitting this case if
                 # transfer-encoding, content-length, and
                 # connection-control were left unspecified.
                 self._reader = _readers.CloseIsEndReader()
                 logger.debug('using a close-is-end reader')
                 self.will_close = True
 
         if body:
             # We're a friend of the reader class here.
             # pylint: disable=W0212
             self._reader._load(body)
         logger.debug('headers complete')
         self.headers = headers
         # We're a friend of the reader class here.
         # pylint: disable=W0212
         self._load_response = self._reader._load
 
+def _foldheaders(headers):
+    """Given some headers, rework them so we can safely overwrite values.
+
+    >>> _foldheaders({'Accept-Encoding': 'wat'})
+    {'accept-encoding': ('Accept-Encoding', 'wat')}
+    """
+    return dict((k.lower(), (k, v)) for k, v in headers.iteritems())
+
 
 class HTTPConnection(object):
     """Connection to a single http server.
 
     Supports 100-continue and keepalives natively. Uses select() for
     non-blocking socket operations.
     """
     http_version = HTTP_VER_1_1
     response_class = HTTPResponse
 
     def __init__(self, host, port=None, use_ssl=None, ssl_validator=None,
                  timeout=TIMEOUT_DEFAULT,
                  continue_timeout=TIMEOUT_ASSUME_CONTINUE,
-                 proxy_hostport=None, ssl_wrap_socket=None, **ssl_opts):
+                 proxy_hostport=None, proxy_headers=None,
+                 ssl_wrap_socket=None, **ssl_opts):
         """Create a new HTTPConnection.
 
         Args:
           host: The host to which we'll connect.
           port: Optional. The port over which we'll connect. Default 80 for
                 non-ssl, 443 for ssl.
           use_ssl: Optional. Whether to use ssl. Defaults to False if port is
                    not 443, true if port is 443.
           ssl_validator: a function(socket) to validate the ssl cert
           timeout: Optional. Connection timeout, default is TIMEOUT_DEFAULT.
           continue_timeout: Optional. Timeout for waiting on an expected
                    "100 Continue" response. Default is TIMEOUT_ASSUME_CONTINUE.
           proxy_hostport: Optional. Tuple of (host, port) to use as an http
                        proxy for the connection. Default is to not use a proxy.
+          proxy_headers: Optional dict of header keys and values to send to
+                         a proxy when using CONNECT. For compatibility with
+                         httplib, the Proxy-Authorization header may be
+                         specified in headers for request(), which will clobber
+                         any such header specified here if specified. Providing
+                         this option and not proxy_hostport will raise an
+                         ValueError.
           ssl_wrap_socket: Optional function to use for wrapping
             sockets. If unspecified, the one from the ssl module will
             be used if available, or something that's compatible with
             it if on a Python older than 2.6.
 
         Any extra keyword arguments to this function will be provided
         to the ssl_wrap_socket method. If no ssl
         """
         if port is None and host.count(':') == 1 or ']:' in host:
             host, port = host.rsplit(':', 1)
             port = int(port)
             if '[' in host:
                 host = host[1:-1]
         if ssl_wrap_socket is not None:
             self._ssl_wrap_socket = ssl_wrap_socket
         else:
             self._ssl_wrap_socket = socketutil.wrap_socket
         if use_ssl is None and port is None:
             use_ssl = False
             port = 80
         elif use_ssl is None:
             use_ssl = (port == 443)
         elif port is None:
-            if use_ssl:
-                port = 443
-            else:
-                port = 80
+            port = (use_ssl and 443 or 80)
         self.port = port
         if use_ssl and not socketutil.have_ssl:
             raise Exception('ssl requested but unavailable on this Python')
         self.ssl = use_ssl
         self.ssl_opts = ssl_opts
         self._ssl_validator = ssl_validator
         self.host = host
         self.sock = None
         self._current_response = None
         self._current_response_taken = False
         if proxy_hostport is None:
             self._proxy_host = self._proxy_port = None
+            if proxy_headers:
+                raise ValueError(
+                    'proxy_headers may not be specified unless '
+                    'proxy_hostport is also specified.')
+            else:
+                self._proxy_headers = {}
         else:
             self._proxy_host, self._proxy_port = proxy_hostport
+            self._proxy_headers = _foldheaders(proxy_headers or {})
 
         self.timeout = timeout
         self.continue_timeout = continue_timeout
 
-    def _connect(self):
+    def _connect(self, proxy_headers):
         """Connect to the host and port specified in __init__."""
         if self.sock:
             return
         if self._proxy_host is not None:
             logger.info('Connecting to http proxy %s:%s',
                         self._proxy_host, self._proxy_port)
             sock = socketutil.create_connection((self._proxy_host,
                                                  self._proxy_port))
             if self.ssl:
-                # TODO proxy header support
                 data = self._buildheaders('CONNECT', '%s:%d' % (self.host,
                                                                 self.port),
-                                          {}, HTTP_VER_1_0)
+                                          proxy_headers, HTTP_VER_1_0)
                 sock.send(data)
                 sock.setblocking(0)
                 r = self.response_class(sock, self.timeout, 'CONNECT')
                 timeout_exc = HTTPTimeoutException(
                     'Timed out waiting for CONNECT response from proxy')
                 while not r.complete():
                     try:
                         # We're a friend of the response class, so let
                         # us use the private attribute.
                         # pylint: disable=W0212
                         if not r._select():
                             if not r.complete():
                                 raise timeout_exc
                     except HTTPTimeoutException:
                         # This raise/except pattern looks goofy, but
                         # _select can raise the timeout as well as the
                         # loop body. I wish it wasn't this convoluted,
                         # but I don't have a better solution
                         # immediately handy.
                         raise timeout_exc
                 if r.status != 200:
                     raise HTTPProxyConnectFailedException(
                         'Proxy connection failed: %d %s' % (r.status,
                                                             r.read()))
                 logger.info('CONNECT (for SSL) to %s:%s via proxy succeeded.',
                             self.host, self.port)
         else:
             sock = socketutil.create_connection((self.host, self.port))
         if self.ssl:
             # This is the default, but in the case of proxied SSL
             # requests the proxy logic above will have cleared
             # blocking mode, so re-enable it just to be safe.
             sock.setblocking(1)
             logger.debug('wrapping socket for ssl with options %r',
                          self.ssl_opts)
             sock = self._ssl_wrap_socket(sock, **self.ssl_opts)
             if self._ssl_validator:
                 self._ssl_validator(sock)
         sock.setblocking(0)
         self.sock = sock
 
     def _buildheaders(self, method, path, headers, http_ver):
         if self.ssl and self.port == 443 or self.port == 80:
             # default port for protocol, so leave it out
             hdrhost = self.host
         else:
             # include nonstandard port in header
             if ':' in self.host:  # must be IPv6
                 hdrhost = '[%s]:%d' % (self.host, self.port)
             else:
                 hdrhost = '%s:%d' % (self.host, self.port)
         if self._proxy_host and not self.ssl:
             # When talking to a regular http proxy we must send the
             # full URI, but in all other cases we must not (although
             # technically RFC 2616 says servers must accept our
             # request if we screw up, experimentally few do that
             # correctly.)
             assert path[0] == '/', 'path must start with a /'
             path = 'http://%s%s' % (hdrhost, path)
         outgoing = ['%s %s %s%s' % (method, path, http_ver, EOL)]
         headers['host'] = ('Host', hdrhost)
         headers[HDR_ACCEPT_ENCODING] = (HDR_ACCEPT_ENCODING, 'identity')
         for hdr, val in headers.itervalues():
             outgoing.append('%s: %s%s' % (hdr, val, EOL))
         outgoing.append(EOL)
         return ''.join(outgoing)
 
     def close(self):
         """Close the connection to the server.
 
         This is a no-op if the connection is already closed. The
         connection may automatically close if requested by the server
         or required by the nature of a response.
         """
         if self.sock is None:
             return
         self.sock.close()
         self.sock = None
         logger.info('closed connection to %s on %s', self.host, self.port)
 
     def busy(self):
         """Returns True if this connection object is currently in use.
 
         If a response is still pending, this will return True, even if
         the request has finished sending. In the future,
         HTTPConnection may transparently juggle multiple connections
         to the server, in which case this will be useful to detect if
         any of those connections is ready for use.
         """
         cr = self._current_response
         if cr is not None:
             if self._current_response_taken:
                 if cr.will_close:
                     self.sock = None
                     self._current_response = None
                     return False
                 elif cr.complete():
                     self._current_response = None
                     return False
             return True
         return False
 
-    def _reconnect(self, where):
+    def _reconnect(self, where, pheaders):
         logger.info('reconnecting during %s', where)
         self.close()
-        self._connect()
+        self._connect(pheaders)
 
     def request(self, method, path, body=None, headers={},
                 expect_continue=False):
         """Send a request to the server.
 
         For increased flexibility, this does not return the response
         object. Future versions of HTTPConnection that juggle multiple
         sockets will be able to send (for example) 5 requests all at
         once, and then let the requests arrive as data is
         available. Use the `getresponse()` method to retrieve the
         response.
         """
         if self.busy():
             raise httplib.CannotSendRequest(
                 'Can not send another request before '
                 'current response is read!')
         self._current_response_taken = False
 
         logger.info('sending %s request for %s to %s on port %s',
                     method, path, self.host, self.port)
-        hdrs = dict((k.lower(), (k, v)) for k, v in headers.iteritems())
+        hdrs = _foldheaders(headers)
         if hdrs.get('expect', ('', ''))[1].lower() == '100-continue':
             expect_continue = True
         elif expect_continue:
             hdrs['expect'] = ('Expect', '100-Continue')
+        # httplib compatibility: if the user specified a
+        # proxy-authorization header, that's actually intended for a
+        # proxy CONNECT action, not the real request, but only if
+        # we're going to use a proxy.
+        pheaders = dict(self._proxy_headers)
+        if self._proxy_host and self.ssl:
+            pa = hdrs.pop('proxy-authorization', None)
+            if pa is not None:
+                pheaders['proxy-authorization'] = pa
 
         chunked = False
         if body and HDR_CONTENT_LENGTH not in hdrs:
             if getattr(body, '__len__', False):
                 hdrs[HDR_CONTENT_LENGTH] = (HDR_CONTENT_LENGTH, len(body))
             elif getattr(body, 'read', False):
                 hdrs[HDR_XFER_ENCODING] = (HDR_XFER_ENCODING,
                                            XFER_ENCODING_CHUNKED)
                 chunked = True
             else:
                 raise BadRequestData('body has no __len__() nor read()')
 
         # If we're reusing the underlying socket, there are some
         # conditions where we'll want to retry, so make a note of the
         # state of self.sock
         fresh_socket = self.sock is None
-        self._connect()
+        self._connect(pheaders)
         outgoing_headers = self._buildheaders(
             method, path, hdrs, self.http_version)
         response = None
         first = True
 
         while ((outgoing_headers or body)
                and not (response and response.complete())):
             select_timeout = self.timeout
             out = outgoing_headers or body
             blocking_on_continue = False
             if expect_continue and not outgoing_headers and not (
                 response and (response.headers or response.continued)):
                 logger.info(
                     'waiting up to %s seconds for'
                     ' continue response from server',
                     self.continue_timeout)
                 select_timeout = self.continue_timeout
                 blocking_on_continue = True
                 out = False
             if out:
                 w = [self.sock]
             else:
                 w = []
             r, w, x = select.select([self.sock], w, [], select_timeout)
             # if we were expecting a 100 continue and it's been long
             # enough, just go ahead and assume it's ok. This is the
             # recommended behavior from the RFC.
             if r == w == x == []:
                 if blocking_on_continue:
                     expect_continue = False
                     logger.info('no response to continue expectation from '
                                 'server, optimistically sending request body')
                 else:
                     raise HTTPTimeoutException('timeout sending data')
             was_first = first
 
             # incoming data
             if r:
                 try:
                     try:
                         data = r[0].recv(INCOMING_BUFFER_SIZE)
                     except socket.sslerror as e:
                         if e.args[0] != socket.SSL_ERROR_WANT_READ:
                             raise
                         logger.debug('SSL_ERROR_WANT_READ while sending '
                                      'data, retrying...')
                         continue
                     if not data:
                         logger.info('socket appears closed in read')
                         self.sock = None
                         self._current_response = None
                         if response is not None:
                             # We're a friend of the response class, so let
                             # us use the private attribute.
                             # pylint: disable=W0212
                             response._close()
                         # This if/elif ladder is a bit subtle,
                         # comments in each branch should help.
                         if response is not None and response.complete():
                             # Server responded completely and then
                             # closed the socket. We should just shut
                             # things down and let the caller get their
                             # response.
                             logger.info('Got an early response, '
                                         'aborting remaining request.')
                             break
                         elif was_first and response is None:
                             # Most likely a keepalive that got killed
                             # on the server's end. Commonly happens
                             # after getting a really large response
                             # from the server.
                             logger.info(
                                 'Connection appeared closed in read on first'
                                 ' request loop iteration, will retry.')
-                            self._reconnect('read')
+                            self._reconnect('read', pheaders)
                             continue
                         else:
                             # We didn't just send the first data hunk,
                             # and either have a partial response or no
                             # response at all. There's really nothing
                             # meaningful we can do here.
                             raise HTTPStateError(
                                 'Connection appears closed after '
                                 'some request data was written, but the '
                                 'response was missing or incomplete!')
                     logger.debug('read %d bytes in request()', len(data))
                     if response is None:
                         response = self.response_class(
                             r[0], self.timeout, method)
                     # We're a friend of the response class, so let us
                     # use the private attribute.
                     # pylint: disable=W0212
                     response._load_response(data)
                     # Jump to the next select() call so we load more
                     # data if the server is still sending us content.
                     continue
                 except socket.error as e:
                     if e[0] != errno.EPIPE and not was_first:
                         raise
 
             # outgoing data
             if w and out:
                 try:
                     if getattr(out, 'read', False):
                         # pylint guesses the type of out incorrectly here
                         # pylint: disable=E1103
                         data = out.read(OUTGOING_BUFFER_SIZE)
                         if not data:
                             continue
                         if len(data) < OUTGOING_BUFFER_SIZE:
                             if chunked:
                                 body = '0' + EOL + EOL
                             else:
                                 body = None
                         if chunked:
                             out = hex(len(data))[2:] + EOL + data + EOL
                         else:
                             out = data
                     amt = w[0].send(out)
                 except socket.error as e:
                     if e[0] == socket.SSL_ERROR_WANT_WRITE and self.ssl:
                         # This means that SSL hasn't flushed its buffer into
                         # the socket yet.
                         # TODO: find a way to block on ssl flushing its buffer
                         # similar to selecting on a raw socket.
                         continue
                     if e[0] == errno.EWOULDBLOCK or e[0] == errno.EAGAIN:
                         continue
                     elif (e[0] not in (errno.ECONNRESET, errno.EPIPE)
                           and not first):
                         raise
-                    self._reconnect('write')
+                    self._reconnect('write', pheaders)
                     amt = self.sock.send(out)
                 logger.debug('sent %d', amt)
                 first = False
                 if out is body:
                     body = out[amt:]
                 else:
                     outgoing_headers = out[amt:]
 
         # close if the server response said to or responded before eating
         # the whole request
         if response is None:
             response = self.response_class(self.sock, self.timeout, method)
             if not fresh_socket:
                 if not response._select():
                     # This means the response failed to get any response
                     # data at all, and in all probability the socket was
                     # closed before the server even saw our request. Try
                     # the request again on a fresh socket.
-                    logging.debug('response._select() failed during request().'
+                    logger.debug('response._select() failed during request().'
                                   ' Assuming request needs to be retried.')
                     self.sock = None
                     # Call this method explicitly to re-try the
                     # request. We don't use self.request() because
                     # some tools (notably Mercurial) expect to be able
                     # to subclass and redefine request(), and they
                     # don't have the same argspec as we do.
                     #
                     # TODO restructure sending of requests to avoid
                     # this recursion
                     return HTTPConnection.request(
                         self, method, path, body=body, headers=headers,
                         expect_continue=expect_continue)
         data_left = bool(outgoing_headers or body)
         if data_left:
             logger.info('stopped sending request early, '
                          'will close the socket to be safe.')
             response.will_close = True
         if response.will_close:
             # The socket will be closed by the response, so we disown
             # the socket
             self.sock = None
         self._current_response = response
 
     def getresponse(self):
         """Returns the response to the most recent request."""
         if self._current_response is None:
             raise httplib.ResponseNotReady()
         r = self._current_response
         while r.headers is None:
             # We're a friend of the response class, so let us use the
             # private attribute.
             # pylint: disable=W0212
             if not r._select() and not r.complete():
                 raise _readers.HTTPRemoteClosedError()
         if r.will_close:
             self.sock = None
             self._current_response = None
         elif r.complete():
             self._current_response = None
         else:
             self._current_response_taken = True
         return r
 
 
 class HTTPTimeoutException(httplib.HTTPException):
     """A timeout occurred while waiting on the server."""
 
 
 class BadRequestData(httplib.HTTPException):
     """Request body object has neither __len__ nor read."""
 
 
 class HTTPProxyConnectFailedException(httplib.HTTPException):
     """Connecting to the HTTP proxy failed."""
 
 
 class HTTPStateError(httplib.HTTPException):
     """Invalid internal state encountered."""
 
 # Forward this exception type from _readers since it needs to be part
 # of the public API.
 HTTPRemoteClosedError = _readers.HTTPRemoteClosedError
 # no-check-code

mercurial/httpclient/_readers.py

@@ -1,232 +1,234
 # Copyright 2011, Google Inc.
 # All rights reserved.
 #
 # Redistribution and use in source and binary forms, with or without
 # modification, are permitted provided that the following conditions are
 # met:
 #
 #     * Redistributions of source code must retain the above copyright
 # notice, this list of conditions and the following disclaimer.
 #     * Redistributions in binary form must reproduce the above
 # copyright notice, this list of conditions and the following disclaimer
 # in the documentation and/or other materials provided with the
 # distribution.
 #     * Neither the name of Google Inc. nor the names of its
 # contributors may be used to endorse or promote products derived from
 # this software without specific prior written permission.
 
 # THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 # "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
 # LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
 # A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
 # OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
 # SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
 # LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
 # DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
 # THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 # (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 """Reader objects to abstract out different body response types.
 
 This module is package-private. It is not expected that these will
 have any clients outside of httpplus.
 """
+from __future__ import absolute_import
 
 import httplib
 import logging
 
 logger = logging.getLogger(__name__)
 
 
 class ReadNotReady(Exception):
     """Raised when read() is attempted but not enough data is loaded."""
 
 
 class HTTPRemoteClosedError(httplib.HTTPException):
     """The server closed the remote socket in the middle of a response."""
 
 
 class AbstractReader(object):
     """Abstract base class for response readers.
 
     Subclasses must implement _load, and should implement _close if
     it's not an error for the server to close their socket without
     some termination condition being detected during _load.
     """
     def __init__(self):
         self._finished = False
         self._done_chunks = []
         self.available_data = 0
 
     def _addchunk(self, data):
         self._done_chunks.append(data)
         self.available_data += len(data)
 
     def _pushchunk(self, data):
         self._done_chunks.insert(0, data)
         self.available_data += len(data)
 
     def _popchunk(self):
         b = self._done_chunks.pop(0)
         self.available_data -= len(b)
 
         return b
 
     def done(self):
         """Returns true if the response body is entirely read."""
         return self._finished
 
     def read(self, amt):
         """Read amt bytes from the response body."""
         if self.available_data < amt and not self._finished:
             raise ReadNotReady()
         blocks = []
         need = amt
         while self._done_chunks:
             b = self._popchunk()
             if len(b) > need:
                 nb = b[:need]
                 self._pushchunk(b[need:])
                 b = nb
             blocks.append(b)
             need -= len(b)
             if need == 0:
                 break
         result = ''.join(blocks)
         assert len(result) == amt or (self._finished and len(result) < amt)
 
         return result
 
     def readto(self, delimstr, blocks = None):
-        """return available data chunks up to the first one in which delimstr
-        occurs. No data will be returned after delimstr -- the chunk in which
-        it occurs will be split and the remainder pushed back onto the available
-        data queue. If blocks is supplied chunks will be added to blocks, otherwise
-        a new list will be allocated.
+        """return available data chunks up to the first one in which
+        delimstr occurs. No data will be returned after delimstr --
+        the chunk in which it occurs will be split and the remainder
+        pushed back onto the available data queue. If blocks is
+        supplied chunks will be added to blocks, otherwise a new list
+        will be allocated.
         """
         if blocks is None:
             blocks = []
 
         while self._done_chunks:
             b = self._popchunk()
             i = b.find(delimstr) + len(delimstr)
             if i:
                 if i < len(b):
                     self._pushchunk(b[i:])
                 blocks.append(b[:i])
                 break
             else:
                 blocks.append(b)
 
         return blocks
 
     def _load(self, data): # pragma: no cover
         """Subclasses must implement this.
 
         As data is available to be read out of this object, it should
         be placed into the _done_chunks list. Subclasses should not
         rely on data remaining in _done_chunks forever, as it may be
         reaped if the client is parsing data as it comes in.
         """
         raise NotImplementedError
 
     def _close(self):
         """Default implementation of close.
 
         The default implementation assumes that the reader will mark
         the response as finished on the _finished attribute once the
         entire response body has been read. In the event that this is
         not true, the subclass should override the implementation of
         close (for example, close-is-end responses have to set
         self._finished in the close handler.)
         """
         if not self._finished:
             raise HTTPRemoteClosedError(
                 'server appears to have closed the socket mid-response')
 
 
 class AbstractSimpleReader(AbstractReader):
     """Abstract base class for simple readers that require no response decoding.
 
     Examples of such responses are Connection: Close (close-is-end)
     and responses that specify a content length.
     """
     def _load(self, data):
         if data:
             assert not self._finished, (
                 'tried to add data (%r) to a closed reader!' % data)
         logger.debug('%s read an additional %d data',
                      self.name, len(data)) # pylint: disable=E1101
         self._addchunk(data)
 
 
 class CloseIsEndReader(AbstractSimpleReader):
     """Reader for responses that specify Connection: Close for length."""
     name = 'close-is-end'
 
     def _close(self):
         logger.info('Marking close-is-end reader as closed.')
         self._finished = True
 
 
 class ContentLengthReader(AbstractSimpleReader):
     """Reader for responses that specify an exact content length."""
     name = 'content-length'
 
     def __init__(self, amount):
         AbstractSimpleReader.__init__(self)
         self._amount = amount
         if amount == 0:
             self._finished = True
         self._amount_seen = 0
 
     def _load(self, data):
         AbstractSimpleReader._load(self, data)
         self._amount_seen += len(data)
         if self._amount_seen >= self._amount:
             self._finished = True
             logger.debug('content-length read complete')
 
 
 class ChunkedReader(AbstractReader):
     """Reader for chunked transfer encoding responses."""
     def __init__(self, eol):
         AbstractReader.__init__(self)
         self._eol = eol
         self._leftover_skip_amt = 0
         self._leftover_data = ''
 
     def _load(self, data):
         assert not self._finished, 'tried to add data to a closed reader!'
         logger.debug('chunked read an additional %d data', len(data))
         position = 0
         if self._leftover_data:
             logger.debug(
                 'chunked reader trying to finish block from leftover data')
             # TODO: avoid this string concatenation if possible
             data = self._leftover_data + data
             position = self._leftover_skip_amt
             self._leftover_data = ''
             self._leftover_skip_amt = 0
         datalen = len(data)
         while position < datalen:
             split = data.find(self._eol, position)
             if split == -1:
                 self._leftover_data = data
                 self._leftover_skip_amt = position
                 return
             amt = int(data[position:split], base=16)
             block_start = split + len(self._eol)
             # If the whole data chunk plus the eol trailer hasn't
             # loaded, we'll wait for the next load.
             if block_start + amt + len(self._eol) > len(data):
                 self._leftover_data = data
                 self._leftover_skip_amt = position
                 return
             if amt == 0:
                 self._finished = True
                 logger.debug('closing chunked reader due to chunk of length 0')
                 return
             self._addchunk(data[block_start:block_start + amt])
             position = block_start + amt + len(self._eol)
 # no-check-code

mercurial/httpclient/socketutil.py

@@ -1,138 +1,140
 # Copyright 2010, Google Inc.
 # All rights reserved.
 #
 # Redistribution and use in source and binary forms, with or without
 # modification, are permitted provided that the following conditions are
 # met:
 #
 #     * Redistributions of source code must retain the above copyright
 # notice, this list of conditions and the following disclaimer.
 #     * Redistributions in binary form must reproduce the above
 # copyright notice, this list of conditions and the following disclaimer
 # in the documentation and/or other materials provided with the
 # distribution.
 #     * Neither the name of Google Inc. nor the names of its
 # contributors may be used to endorse or promote products derived from
 # this software without specific prior written permission.
 
 # THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 # "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
 # LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
 # A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
 # OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
 # SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
 # LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
 # DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
 # THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 # (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 """Abstraction to simplify socket use for Python < 2.6
 
 This will attempt to use the ssl module and the new
 socket.create_connection method, but fall back to the old
 methods if those are unavailable.
 """
+from __future__ import absolute_import
+
 import logging
 import socket
 
 logger = logging.getLogger(__name__)
 
 try:
     import ssl
     # make demandimporters load the module
     ssl.wrap_socket # pylint: disable=W0104
     have_ssl = True
 except ImportError:
     import httplib
     import urllib2
     have_ssl = getattr(urllib2, 'HTTPSHandler', False)
     ssl = False
 
 
 try:
     create_connection = socket.create_connection
 except AttributeError:
     def create_connection(address):
         """Backport of socket.create_connection from Python 2.6."""
         host, port = address
         msg = "getaddrinfo returns an empty list"
         sock = None
         for res in socket.getaddrinfo(host, port, 0,
                                       socket.SOCK_STREAM):
             af, socktype, proto, unused_canonname, sa = res
             try:
                 sock = socket.socket(af, socktype, proto)
                 logger.info("connect: (%s, %s)", host, port)
                 sock.connect(sa)
             except socket.error as msg:
                 logger.info('connect fail: %s %s', host, port)
                 if sock:
                     sock.close()
                 sock = None
                 continue
             break
         if not sock:
             raise socket.error(msg)
         return sock
 
 if ssl:
     wrap_socket = ssl.wrap_socket
     CERT_NONE = ssl.CERT_NONE
     CERT_OPTIONAL = ssl.CERT_OPTIONAL
     CERT_REQUIRED = ssl.CERT_REQUIRED
 else:
     class FakeSocket(httplib.FakeSocket):
         """Socket wrapper that supports SSL."""
 
         # Silence lint about this goofy backport class
         # pylint: disable=W0232,E1101,R0903,R0913,C0111
 
         # backport the behavior from Python 2.6, which is to busy wait
         # on the socket instead of anything nice. Sigh.
         # See http://bugs.python.org/issue3890 for more info.
         def recv(self, buflen=1024, flags=0):
             """ssl-aware wrapper around socket.recv
             """
             if flags != 0:
                 raise ValueError(
                     "non-zero flags not allowed in calls to recv() on %s" %
                     self.__class__)
             while True:
                 try:
                     return self._ssl.read(buflen)
                 except socket.sslerror as x:
                     if x.args[0] == socket.SSL_ERROR_WANT_READ:
                         continue
                     else:
                         raise x
 
     _PROTOCOL_SSLv23 = 2
 
     CERT_NONE = 0
     CERT_OPTIONAL = 1
     CERT_REQUIRED = 2
 
     # Disable unused-argument because we're making a dumb wrapper
     # that's like an upstream method.
     #
     # pylint: disable=W0613,R0913
     def wrap_socket(sock, keyfile=None, certfile=None,
                 server_side=False, cert_reqs=CERT_NONE,
                 ssl_version=_PROTOCOL_SSLv23, ca_certs=None,
                 do_handshake_on_connect=True,
                 suppress_ragged_eofs=True):
         """Backport of ssl.wrap_socket from Python 2.6."""
         if cert_reqs != CERT_NONE and ca_certs:
             raise CertificateValidationUnsupported(
                 'SSL certificate validation requires the ssl module'
                 '(included in Python 2.6 and later.)')
         sslob = socket.ssl(sock)
         # borrow httplib's workaround for no ssl.wrap_socket
         sock = FakeSocket(sock, sslob)
         return sock
     # pylint: enable=W0613,R0913
 
 
 class CertificateValidationUnsupported(Exception):
     """Exception raised when cert validation is requested but unavailable."""
 # no-check-code

tests/test-check-py3-compat.t

@@ -1,189 +1,186
 #require test-repo
 
   $ cd "$TESTDIR"/..
 
   $ hg files 'set:(**.py)' | sed 's|\\|/|g' | xargs python contrib/check-py3-compat.py
   contrib/casesmash.py not using absolute_import
   contrib/check-code.py not using absolute_import
   contrib/check-code.py requires print_function
   contrib/check-config.py not using absolute_import
   contrib/check-config.py requires print_function
   contrib/debugcmdserver.py not using absolute_import
   contrib/debugcmdserver.py requires print_function
   contrib/debugshell.py not using absolute_import
   contrib/fixpax.py not using absolute_import
   contrib/fixpax.py requires print_function
   contrib/hgclient.py not using absolute_import
   contrib/hgclient.py requires print_function
   contrib/hgfixes/fix_bytes.py not using absolute_import
   contrib/hgfixes/fix_bytesmod.py not using absolute_import
   contrib/hgfixes/fix_leftover_imports.py not using absolute_import
   contrib/import-checker.py not using absolute_import
   contrib/import-checker.py requires print_function
   contrib/memory.py not using absolute_import
   contrib/perf.py not using absolute_import
   contrib/python-hook-examples.py not using absolute_import
   contrib/revsetbenchmarks.py not using absolute_import
   contrib/revsetbenchmarks.py requires print_function
   contrib/showstack.py not using absolute_import
   contrib/synthrepo.py not using absolute_import
   contrib/win32/hgwebdir_wsgi.py not using absolute_import
   doc/check-seclevel.py not using absolute_import
   doc/gendoc.py not using absolute_import
   doc/hgmanpage.py not using absolute_import
   hgext/__init__.py not using absolute_import
   hgext/acl.py not using absolute_import
   hgext/blackbox.py not using absolute_import
   hgext/bugzilla.py not using absolute_import
   hgext/censor.py not using absolute_import
   hgext/children.py not using absolute_import
   hgext/churn.py not using absolute_import
   hgext/clonebundles.py not using absolute_import
   hgext/color.py not using absolute_import
   hgext/convert/__init__.py not using absolute_import
   hgext/convert/bzr.py not using absolute_import
   hgext/convert/common.py not using absolute_import
   hgext/convert/convcmd.py not using absolute_import
   hgext/convert/cvs.py not using absolute_import
   hgext/convert/cvsps.py not using absolute_import
   hgext/convert/darcs.py not using absolute_import
   hgext/convert/filemap.py not using absolute_import
   hgext/convert/git.py not using absolute_import
   hgext/convert/gnuarch.py not using absolute_import
   hgext/convert/hg.py not using absolute_import
   hgext/convert/monotone.py not using absolute_import
   hgext/convert/p4.py not using absolute_import
   hgext/convert/subversion.py not using absolute_import
   hgext/convert/transport.py not using absolute_import
   hgext/eol.py not using absolute_import
   hgext/extdiff.py not using absolute_import
   hgext/factotum.py not using absolute_import
   hgext/fetch.py not using absolute_import
   hgext/gpg.py not using absolute_import
   hgext/graphlog.py not using absolute_import
   hgext/hgcia.py not using absolute_import
   hgext/hgk.py not using absolute_import
   hgext/highlight/__init__.py not using absolute_import
   hgext/highlight/highlight.py not using absolute_import
   hgext/histedit.py not using absolute_import
   hgext/keyword.py not using absolute_import
   hgext/largefiles/__init__.py not using absolute_import
   hgext/largefiles/basestore.py not using absolute_import
   hgext/largefiles/lfcommands.py not using absolute_import
   hgext/largefiles/lfutil.py not using absolute_import
   hgext/largefiles/localstore.py not using absolute_import
   hgext/largefiles/overrides.py not using absolute_import
   hgext/largefiles/proto.py not using absolute_import
   hgext/largefiles/remotestore.py not using absolute_import
   hgext/largefiles/reposetup.py not using absolute_import
   hgext/largefiles/uisetup.py not using absolute_import
   hgext/largefiles/wirestore.py not using absolute_import
   hgext/mq.py not using absolute_import
   hgext/notify.py not using absolute_import
   hgext/pager.py not using absolute_import
   hgext/patchbomb.py not using absolute_import
   hgext/purge.py not using absolute_import
   hgext/rebase.py not using absolute_import
   hgext/record.py not using absolute_import
   hgext/relink.py not using absolute_import
   hgext/schemes.py not using absolute_import
   hgext/share.py not using absolute_import
   hgext/shelve.py not using absolute_import
   hgext/strip.py not using absolute_import
   hgext/transplant.py not using absolute_import
   hgext/win32mbcs.py not using absolute_import
   hgext/win32text.py not using absolute_import
   hgext/zeroconf/Zeroconf.py not using absolute_import
   hgext/zeroconf/Zeroconf.py requires print_function
   hgext/zeroconf/__init__.py not using absolute_import
   i18n/check-translation.py not using absolute_import
   i18n/polib.py not using absolute_import
   mercurial/cmdutil.py not using absolute_import
   mercurial/commands.py not using absolute_import
   mercurial/dispatch.py requires print_function
-  mercurial/httpclient/__init__.py not using absolute_import
-  mercurial/httpclient/_readers.py not using absolute_import
-  mercurial/httpclient/socketutil.py not using absolute_import
   mercurial/keepalive.py requires print_function
   mercurial/lsprof.py requires print_function
   mercurial/lsprofcalltree.py requires print_function
   mercurial/mail.py requires print_function
   setup.py not using absolute_import
   tests/filterpyflakes.py requires print_function
   tests/generate-working-copy-states.py requires print_function
   tests/get-with-headers.py requires print_function
   tests/heredoctest.py requires print_function
   tests/hypothesishelpers.py not using absolute_import
   tests/hypothesishelpers.py requires print_function
   tests/killdaemons.py not using absolute_import
   tests/md5sum.py not using absolute_import
   tests/mockblackbox.py not using absolute_import
   tests/printenv.py not using absolute_import
   tests/readlink.py not using absolute_import
   tests/readlink.py requires print_function
   tests/revlog-formatv0.py not using absolute_import
   tests/run-tests.py not using absolute_import
   tests/seq.py not using absolute_import
   tests/seq.py requires print_function
   tests/silenttestrunner.py not using absolute_import
   tests/silenttestrunner.py requires print_function
   tests/sitecustomize.py not using absolute_import
   tests/svn-safe-append.py not using absolute_import
   tests/svnxml.py not using absolute_import
   tests/test-ancestor.py requires print_function
   tests/test-atomictempfile.py not using absolute_import
   tests/test-batching.py not using absolute_import
   tests/test-batching.py requires print_function
   tests/test-bdiff.py not using absolute_import
   tests/test-bdiff.py requires print_function
   tests/test-context.py not using absolute_import
   tests/test-context.py requires print_function
   tests/test-demandimport.py not using absolute_import
   tests/test-demandimport.py requires print_function
   tests/test-dispatch.py not using absolute_import
   tests/test-dispatch.py requires print_function
   tests/test-doctest.py not using absolute_import
   tests/test-duplicateoptions.py not using absolute_import
   tests/test-duplicateoptions.py requires print_function
   tests/test-filecache.py not using absolute_import
   tests/test-filecache.py requires print_function
   tests/test-filelog.py not using absolute_import
   tests/test-filelog.py requires print_function
   tests/test-hg-parseurl.py not using absolute_import
   tests/test-hg-parseurl.py requires print_function
   tests/test-hgweb-auth.py not using absolute_import
   tests/test-hgweb-auth.py requires print_function
   tests/test-hgwebdir-paths.py not using absolute_import
   tests/test-hybridencode.py not using absolute_import
   tests/test-hybridencode.py requires print_function
   tests/test-lrucachedict.py not using absolute_import
   tests/test-lrucachedict.py requires print_function
   tests/test-manifest.py not using absolute_import
   tests/test-minirst.py not using absolute_import
   tests/test-minirst.py requires print_function
   tests/test-parseindex2.py not using absolute_import
   tests/test-parseindex2.py requires print_function
   tests/test-pathencode.py not using absolute_import
   tests/test-pathencode.py requires print_function
   tests/test-propertycache.py not using absolute_import
   tests/test-propertycache.py requires print_function
   tests/test-revlog-ancestry.py not using absolute_import
   tests/test-revlog-ancestry.py requires print_function
   tests/test-run-tests.py not using absolute_import
   tests/test-simplemerge.py not using absolute_import
   tests/test-status-inprocess.py not using absolute_import
   tests/test-status-inprocess.py requires print_function
   tests/test-symlink-os-yes-fs-no.py not using absolute_import
   tests/test-trusted.py not using absolute_import
   tests/test-trusted.py requires print_function
   tests/test-ui-color.py not using absolute_import
   tests/test-ui-color.py requires print_function
   tests/test-ui-config.py not using absolute_import
   tests/test-ui-config.py requires print_function
   tests/test-ui-verbosity.py not using absolute_import
   tests/test-ui-verbosity.py requires print_function
   tests/test-url.py not using absolute_import
   tests/test-url.py requires print_function
   tests/test-walkrepo.py requires print_function
   tests/test-wireproto.py requires print_function
   tests/tinyproxy.py requires print_function