Results of a rewrite pass
[python/dscho.git] / Lib / httplib.py
blobe98fa4637d096a6c29efb0d9e04996cdbdfb839a
1 """HTTP/1.1 client library
3 <intro stuff goes here>
4 <other stuff, too>
6 HTTPConnection go through a number of "states", which defines when a client
7 may legally make another request or fetch the response for a particular
8 request. This diagram details these state transitions:
10 (null)
12 | HTTPConnection()
14 Idle
16 | putrequest()
18 Request-started
20 | ( putheader() )* endheaders()
22 Request-sent
24 | response = getresponse()
26 Unread-response [Response-headers-read]
27 |\____________________
28 | |
29 | response.read() | putrequest()
30 v v
31 Idle Req-started-unread-response
32 ______/|
33 / |
34 response.read() | | ( putheader() )* endheaders()
35 v v
36 Request-started Req-sent-unread-response
38 | response.read()
40 Request-sent
42 This diagram presents the following rules:
43 -- a second request may not be started until {response-headers-read}
44 -- a response [object] cannot be retrieved until {request-sent}
45 -- there is no differentiation between an unread response body and a
46 partially read response body
48 Note: this enforcement is applied by the HTTPConnection class. The
49 HTTPResponse class does not enforce this state machine, which
50 implies sophisticated clients may accelerate the request/response
51 pipeline. Caution should be taken, though: accelerating the states
52 beyond the above pattern may imply knowledge of the server's
53 connection-close behavior for certain requests. For example, it
54 is impossible to tell whether the server will close the connection
55 UNTIL the response headers have been read; this means that further
56 requests cannot be placed into the pipeline until it is known that
57 the server will NOT be closing the connection.
59 Logical State __state __response
60 ------------- ------- ----------
61 Idle _CS_IDLE None
62 Request-started _CS_REQ_STARTED None
63 Request-sent _CS_REQ_SENT None
64 Unread-response _CS_IDLE <response_class>
65 Req-started-unread-response _CS_REQ_STARTED <response_class>
66 Req-sent-unread-response _CS_REQ_SENT <response_class>
67 """
69 import errno
70 import mimetools
71 import socket
72 from urlparse import urlsplit
74 try:
75 from cStringIO import StringIO
76 except ImportError:
77 from StringIO import StringIO
79 __all__ = ["HTTP", "HTTPResponse", "HTTPConnection", "HTTPSConnection",
80 "HTTPException", "NotConnected", "UnknownProtocol",
81 "UnknownTransferEncoding", "UnimplementedFileMode",
82 "IncompleteRead", "InvalidURL", "ImproperConnectionState",
83 "CannotSendRequest", "CannotSendHeader", "ResponseNotReady",
84 "BadStatusLine", "error"]
86 HTTP_PORT = 80
87 HTTPS_PORT = 443
89 _UNKNOWN = 'UNKNOWN'
91 # connection states
92 _CS_IDLE = 'Idle'
93 _CS_REQ_STARTED = 'Request-started'
94 _CS_REQ_SENT = 'Request-sent'
96 class HTTPMessage(mimetools.Message):
98 def addheader(self, key, value):
99 """Add header for field key handling repeats."""
100 prev = self.dict.get(key)
101 if prev is None:
102 self.dict[key] = value
103 else:
104 combined = ", ".join((prev, value))
105 self.dict[key] = combined
107 def addcontinue(self, key, more):
108 """Add more field data from a continuation line."""
109 prev = self.dict[key]
110 self.dict[key] = prev + "\n " + more
112 def readheaders(self):
113 """Read header lines.
115 Read header lines up to the entirely blank line that terminates them.
116 The (normally blank) line that ends the headers is skipped, but not
117 included in the returned list. If a non-header line ends the headers,
118 (which is an error), an attempt is made to backspace over it; it is
119 never included in the returned list.
121 The variable self.status is set to the empty string if all went well,
122 otherwise it is an error message. The variable self.headers is a
123 completely uninterpreted list of lines contained in the header (so
124 printing them will reproduce the header exactly as it appears in the
125 file).
127 If multiple header fields with the same name occur, they are combined
128 according to the rules in RFC 2616 sec 4.2:
130 Appending each subsequent field-value to the first, each separated
131 by a comma. The order in which header fields with the same field-name
132 are received is significant to the interpretation of the combined
133 field value.
135 # XXX The implementation overrides the readheaders() method of
136 # rfc822.Message. The base class design isn't amenable to
137 # customized behavior here so the method here is a copy of the
138 # base class code with a few small changes.
140 self.dict = {}
141 self.unixfrom = ''
142 self.headers = list = []
143 self.status = ''
144 headerseen = ""
145 firstline = 1
146 startofline = unread = tell = None
147 if hasattr(self.fp, 'unread'):
148 unread = self.fp.unread
149 elif self.seekable:
150 tell = self.fp.tell
151 while 1:
152 if tell:
153 try:
154 startofline = tell()
155 except IOError:
156 startofline = tell = None
157 self.seekable = 0
158 line = self.fp.readline()
159 if not line:
160 self.status = 'EOF in headers'
161 break
162 # Skip unix From name time lines
163 if firstline and line.startswith('From '):
164 self.unixfrom = self.unixfrom + line
165 continue
166 firstline = 0
167 if headerseen and line[0] in ' \t':
168 # XXX Not sure if continuation lines are handled properly
169 # for http and/or for repeating headers
170 # It's a continuation line.
171 list.append(line)
172 self.addcontinue(headerseen, line.strip())
173 continue
174 elif self.iscomment(line):
175 # It's a comment. Ignore it.
176 continue
177 elif self.islast(line):
178 # Note! No pushback here! The delimiter line gets eaten.
179 break
180 headerseen = self.isheader(line)
181 if headerseen:
182 # It's a legal header line, save it.
183 list.append(line)
184 self.addheader(headerseen, line[len(headerseen)+1:].strip())
185 continue
186 else:
187 # It's not a header line; throw it back and stop here.
188 if not self.dict:
189 self.status = 'No headers'
190 else:
191 self.status = 'Non-header line where header expected'
192 # Try to undo the read.
193 if unread:
194 unread(line)
195 elif tell:
196 self.fp.seek(startofline)
197 else:
198 self.status = self.status + '; bad seek'
199 break
201 class HTTPResponse:
203 # strict: If true, raise BadStatusLine if the status line can't be
204 # parsed as a valid HTTP/1.0 or 1.1 status line. By default it is
205 # false because it prevents clients from talking to HTTP/0.9
206 # servers. Note that a response with a sufficiently corrupted
207 # status line will look like an HTTP/0.9 response.
209 # See RFC 2616 sec 19.6 and RFC 1945 sec 6 for details.
211 def __init__(self, sock, debuglevel=0, strict=0):
212 self.fp = sock.makefile('rb', 0)
213 self.debuglevel = debuglevel
214 self.strict = strict
216 self.msg = None
218 # from the Status-Line of the response
219 self.version = _UNKNOWN # HTTP-Version
220 self.status = _UNKNOWN # Status-Code
221 self.reason = _UNKNOWN # Reason-Phrase
223 self.chunked = _UNKNOWN # is "chunked" being used?
224 self.chunk_left = _UNKNOWN # bytes left to read in current chunk
225 self.length = _UNKNOWN # number of bytes left in response
226 self.will_close = _UNKNOWN # conn will close at end of response
228 def _read_status(self):
229 # Initialize with Simple-Response defaults
230 line = self.fp.readline()
231 if self.debuglevel > 0:
232 print "reply:", repr(line)
233 try:
234 [version, status, reason] = line.split(None, 2)
235 except ValueError:
236 try:
237 [version, status] = line.split(None, 1)
238 reason = ""
239 except ValueError:
240 # empty version will cause next test to fail and status
241 # will be treated as 0.9 response.
242 version = ""
243 if not version.startswith('HTTP/'):
244 if self.strict:
245 self.close()
246 raise BadStatusLine(line)
247 else:
248 # assume it's a Simple-Response from an 0.9 server
249 self.fp = LineAndFileWrapper(line, self.fp)
250 return "HTTP/0.9", 200, ""
252 # The status code is a three-digit number
253 try:
254 status = int(status)
255 if status < 100 or status > 999:
256 raise BadStatusLine(line)
257 except ValueError:
258 raise BadStatusLine(line)
259 return version, status, reason
261 def begin(self):
262 if self.msg is not None:
263 # we've already started reading the response
264 return
266 # read until we get a non-100 response
267 while 1:
268 version, status, reason = self._read_status()
269 if status != 100:
270 break
271 # skip the header from the 100 response
272 while 1:
273 skip = self.fp.readline().strip()
274 if not skip:
275 break
276 if self.debuglevel > 0:
277 print "header:", skip
279 self.status = status
280 self.reason = reason.strip()
281 if version == 'HTTP/1.0':
282 self.version = 10
283 elif version.startswith('HTTP/1.'):
284 self.version = 11 # use HTTP/1.1 code for HTTP/1.x where x>=1
285 elif version == 'HTTP/0.9':
286 self.version = 9
287 else:
288 raise UnknownProtocol(version)
290 if self.version == 9:
291 self.chunked = 0
292 self.will_close = 1
293 self.msg = HTTPMessage(StringIO())
294 return
296 self.msg = HTTPMessage(self.fp, 0)
297 if self.debuglevel > 0:
298 for hdr in self.msg.headers:
299 print "header:", hdr,
301 # don't let the msg keep an fp
302 self.msg.fp = None
304 # are we using the chunked-style of transfer encoding?
305 tr_enc = self.msg.getheader('transfer-encoding')
306 if tr_enc and tr_enc.lower() == "chunked":
307 self.chunked = 1
308 self.chunk_left = None
309 else:
310 self.chunked = 0
312 # will the connection close at the end of the response?
313 self.will_close = self._check_close()
315 # do we have a Content-Length?
316 # NOTE: RFC 2616, S4.4, #3 says we ignore this if tr_enc is "chunked"
317 length = self.msg.getheader('content-length')
318 if length and not self.chunked:
319 try:
320 self.length = int(length)
321 except ValueError:
322 self.length = None
323 else:
324 self.length = None
326 # does the body have a fixed length? (of zero)
327 if (status == 204 or # No Content
328 status == 304 or # Not Modified
329 100 <= status < 200): # 1xx codes
330 self.length = 0
332 # if the connection remains open, and we aren't using chunked, and
333 # a content-length was not provided, then assume that the connection
334 # WILL close.
335 if not self.will_close and \
336 not self.chunked and \
337 self.length is None:
338 self.will_close = 1
340 def _check_close(self):
341 if self.version == 11:
342 # An HTTP/1.1 proxy is assumed to stay open unless
343 # explicitly closed.
344 conn = self.msg.getheader('connection')
345 if conn and conn.lower().find("close") >= 0:
346 return True
347 return False
349 # An HTTP/1.0 response with a Connection header is probably
350 # the result of a confused proxy. Ignore it.
352 # For older HTTP, Keep-Alive indiciates persistent connection.
353 if self.msg.getheader('keep-alive'):
354 return False
356 # Proxy-Connection is a netscape hack.
357 pconn = self.msg.getheader('proxy-connection')
358 if pconn and pconn.lower().find("keep-alive") >= 0:
359 return False
361 # otherwise, assume it will close
362 return True
364 def close(self):
365 if self.fp:
366 self.fp.close()
367 self.fp = None
369 def isclosed(self):
370 # NOTE: it is possible that we will not ever call self.close(). This
371 # case occurs when will_close is TRUE, length is None, and we
372 # read up to the last byte, but NOT past it.
374 # IMPLIES: if will_close is FALSE, then self.close() will ALWAYS be
375 # called, meaning self.isclosed() is meaningful.
376 return self.fp is None
378 def read(self, amt=None):
379 if self.fp is None:
380 return ''
382 if self.chunked:
383 return self._read_chunked(amt)
385 if amt is None:
386 # unbounded read
387 if self.will_close:
388 s = self.fp.read()
389 else:
390 s = self._safe_read(self.length)
391 self.close() # we read everything
392 return s
394 if self.length is not None:
395 if amt > self.length:
396 # clip the read to the "end of response"
397 amt = self.length
398 self.length -= amt
400 # we do not use _safe_read() here because this may be a .will_close
401 # connection, and the user is reading more bytes than will be provided
402 # (for example, reading in 1k chunks)
403 s = self.fp.read(amt)
405 return s
407 def _read_chunked(self, amt):
408 assert self.chunked != _UNKNOWN
409 chunk_left = self.chunk_left
410 value = ''
412 # XXX This accumulates chunks by repeated string concatenation,
413 # which is not efficient as the number or size of chunks gets big.
414 while 1:
415 if chunk_left is None:
416 line = self.fp.readline()
417 i = line.find(';')
418 if i >= 0:
419 line = line[:i] # strip chunk-extensions
420 chunk_left = int(line, 16)
421 if chunk_left == 0:
422 break
423 if amt is None:
424 value += self._safe_read(chunk_left)
425 elif amt < chunk_left:
426 value += self._safe_read(amt)
427 self.chunk_left = chunk_left - amt
428 return value
429 elif amt == chunk_left:
430 value += self._safe_read(amt)
431 self._safe_read(2) # toss the CRLF at the end of the chunk
432 self.chunk_left = None
433 return value
434 else:
435 value += self._safe_read(chunk_left)
436 amt -= chunk_left
438 # we read the whole chunk, get another
439 self._safe_read(2) # toss the CRLF at the end of the chunk
440 chunk_left = None
442 # read and discard trailer up to the CRLF terminator
443 ### note: we shouldn't have any trailers!
444 while 1:
445 line = self.fp.readline()
446 if line == '\r\n':
447 break
449 # we read everything; close the "file"
450 # XXX Shouldn't the client close the file?
451 self.close()
453 return value
455 def _safe_read(self, amt):
456 """Read the number of bytes requested, compensating for partial reads.
458 Normally, we have a blocking socket, but a read() can be interrupted
459 by a signal (resulting in a partial read).
461 Note that we cannot distinguish between EOF and an interrupt when zero
462 bytes have been read. IncompleteRead() will be raised in this
463 situation.
465 This function should be used when <amt> bytes "should" be present for
466 reading. If the bytes are truly not available (due to EOF), then the
467 IncompleteRead exception can be used to detect the problem.
469 s = ''
470 while amt > 0:
471 chunk = self.fp.read(amt)
472 if not chunk:
473 raise IncompleteRead(s)
474 s = s + chunk
475 amt = amt - len(chunk)
476 return s
478 def getheader(self, name, default=None):
479 if self.msg is None:
480 raise ResponseNotReady()
481 return self.msg.getheader(name, default)
484 class HTTPConnection:
486 _http_vsn = 11
487 _http_vsn_str = 'HTTP/1.1'
489 response_class = HTTPResponse
490 default_port = HTTP_PORT
491 auto_open = 1
492 debuglevel = 0
493 strict = 0
495 def __init__(self, host, port=None, strict=None):
496 self.sock = None
497 self._buffer = []
498 self.__response = None
499 self.__state = _CS_IDLE
501 self._set_hostport(host, port)
502 if strict is not None:
503 self.strict = strict
505 def _set_hostport(self, host, port):
506 if port is None:
507 i = host.find(':')
508 if i >= 0:
509 try:
510 port = int(host[i+1:])
511 except ValueError:
512 raise InvalidURL("nonnumeric port: '%s'" % host[i+1:])
513 host = host[:i]
514 else:
515 port = self.default_port
516 self.host = host
517 self.port = port
519 def set_debuglevel(self, level):
520 self.debuglevel = level
522 def connect(self):
523 """Connect to the host and port specified in __init__."""
524 msg = "getaddrinfo returns an empty list"
525 for res in socket.getaddrinfo(self.host, self.port, 0,
526 socket.SOCK_STREAM):
527 af, socktype, proto, canonname, sa = res
528 try:
529 self.sock = socket.socket(af, socktype, proto)
530 if self.debuglevel > 0:
531 print "connect: (%s, %s)" % (self.host, self.port)
532 self.sock.connect(sa)
533 except socket.error, msg:
534 if self.debuglevel > 0:
535 print 'connect fail:', (self.host, self.port)
536 if self.sock:
537 self.sock.close()
538 self.sock = None
539 continue
540 break
541 if not self.sock:
542 raise socket.error, msg
544 def close(self):
545 """Close the connection to the HTTP server."""
546 if self.sock:
547 self.sock.close() # close it manually... there may be other refs
548 self.sock = None
549 if self.__response:
550 self.__response.close()
551 self.__response = None
552 self.__state = _CS_IDLE
554 def send(self, str):
555 """Send `str' to the server."""
556 if self.sock is None:
557 if self.auto_open:
558 self.connect()
559 else:
560 raise NotConnected()
562 # send the data to the server. if we get a broken pipe, then close
563 # the socket. we want to reconnect when somebody tries to send again.
565 # NOTE: we DO propagate the error, though, because we cannot simply
566 # ignore the error... the caller will know if they can retry.
567 if self.debuglevel > 0:
568 print "send:", repr(str)
569 try:
570 self.sock.sendall(str)
571 except socket.error, v:
572 if v[0] == 32: # Broken pipe
573 self.close()
574 raise
576 def _output(self, s):
577 """Add a line of output to the current request buffer.
579 Assumes that the line does *not* end with \\r\\n.
581 self._buffer.append(s)
583 def _send_output(self):
584 """Send the currently buffered request and clear the buffer.
586 Appends an extra \\r\\n to the buffer.
588 self._buffer.extend(("", ""))
589 msg = "\r\n".join(self._buffer)
590 del self._buffer[:]
591 self.send(msg)
593 def putrequest(self, method, url, skip_host=0):
594 """Send a request to the server.
596 `method' specifies an HTTP request method, e.g. 'GET'.
597 `url' specifies the object being requested, e.g. '/index.html'.
600 # check if a prior response has been completed
601 # XXX What if it hasn't?
602 if self.__response and self.__response.isclosed():
603 self.__response = None
606 # in certain cases, we cannot issue another request on this connection.
607 # this occurs when:
608 # 1) we are in the process of sending a request. (_CS_REQ_STARTED)
609 # 2) a response to a previous request has signalled that it is going
610 # to close the connection upon completion.
611 # 3) the headers for the previous response have not been read, thus
612 # we cannot determine whether point (2) is true. (_CS_REQ_SENT)
614 # if there is no prior response, then we can request at will.
616 # if point (2) is true, then we will have passed the socket to the
617 # response (effectively meaning, "there is no prior response"), and
618 # will open a new one when a new request is made.
620 # Note: if a prior response exists, then we *can* start a new request.
621 # We are not allowed to begin fetching the response to this new
622 # request, however, until that prior response is complete.
624 if self.__state == _CS_IDLE:
625 self.__state = _CS_REQ_STARTED
626 else:
627 raise CannotSendRequest()
629 if not url:
630 url = '/'
631 str = '%s %s %s' % (method, url, self._http_vsn_str)
633 self._output(str)
635 if self._http_vsn == 11:
636 # Issue some standard headers for better HTTP/1.1 compliance
638 if not skip_host:
639 # this header is issued *only* for HTTP/1.1
640 # connections. more specifically, this means it is
641 # only issued when the client uses the new
642 # HTTPConnection() class. backwards-compat clients
643 # will be using HTTP/1.0 and those clients may be
644 # issuing this header themselves. we should NOT issue
645 # it twice; some web servers (such as Apache) barf
646 # when they see two Host: headers
648 # If we need a non-standard port,include it in the
649 # header. If the request is going through a proxy,
650 # but the host of the actual URL, not the host of the
651 # proxy.
653 netloc = ''
654 if url.startswith('http'):
655 nil, netloc, nil, nil, nil = urlsplit(url)
657 if netloc:
658 self.putheader('Host', netloc)
659 elif self.port == HTTP_PORT:
660 self.putheader('Host', self.host)
661 else:
662 self.putheader('Host', "%s:%s" % (self.host, self.port))
664 # note: we are assuming that clients will not attempt to set these
665 # headers since *this* library must deal with the
666 # consequences. this also means that when the supporting
667 # libraries are updated to recognize other forms, then this
668 # code should be changed (removed or updated).
670 # we only want a Content-Encoding of "identity" since we don't
671 # support encodings such as x-gzip or x-deflate.
672 self.putheader('Accept-Encoding', 'identity')
674 # we can accept "chunked" Transfer-Encodings, but no others
675 # NOTE: no TE header implies *only* "chunked"
676 #self.putheader('TE', 'chunked')
678 # if TE is supplied in the header, then it must appear in a
679 # Connection header.
680 #self.putheader('Connection', 'TE')
682 else:
683 # For HTTP/1.0, the server will assume "not chunked"
684 pass
686 def putheader(self, header, value):
687 """Send a request header line to the server.
689 For example: h.putheader('Accept', 'text/html')
691 if self.__state != _CS_REQ_STARTED:
692 raise CannotSendHeader()
694 str = '%s: %s' % (header, value)
695 self._output(str)
697 def endheaders(self):
698 """Indicate that the last header line has been sent to the server."""
700 if self.__state == _CS_REQ_STARTED:
701 self.__state = _CS_REQ_SENT
702 else:
703 raise CannotSendHeader()
705 self._send_output()
707 def request(self, method, url, body=None, headers={}):
708 """Send a complete request to the server."""
710 try:
711 self._send_request(method, url, body, headers)
712 except socket.error, v:
713 # trap 'Broken pipe' if we're allowed to automatically reconnect
714 if v[0] != 32 or not self.auto_open:
715 raise
716 # try one more time
717 self._send_request(method, url, body, headers)
719 def _send_request(self, method, url, body, headers):
720 # If headers already contains a host header, then define the
721 # optional skip_host argument to putrequest(). The check is
722 # harder because field names are case insensitive.
723 if 'Host' in (headers
724 or [k for k in headers.iterkeys() if k.lower() == "host"]):
725 self.putrequest(method, url, skip_host=1)
726 else:
727 self.putrequest(method, url)
729 if body:
730 self.putheader('Content-Length', str(len(body)))
731 for hdr, value in headers.items():
732 self.putheader(hdr, value)
733 self.endheaders()
735 if body:
736 self.send(body)
738 def getresponse(self):
739 "Get the response from the server."
741 # check if a prior response has been completed
742 if self.__response and self.__response.isclosed():
743 self.__response = None
746 # if a prior response exists, then it must be completed (otherwise, we
747 # cannot read this response's header to determine the connection-close
748 # behavior)
750 # note: if a prior response existed, but was connection-close, then the
751 # socket and response were made independent of this HTTPConnection
752 # object since a new request requires that we open a whole new
753 # connection
755 # this means the prior response had one of two states:
756 # 1) will_close: this connection was reset and the prior socket and
757 # response operate independently
758 # 2) persistent: the response was retained and we await its
759 # isclosed() status to become true.
761 if self.__state != _CS_REQ_SENT or self.__response:
762 raise ResponseNotReady()
764 if self.debuglevel > 0:
765 response = self.response_class(self.sock, self.debuglevel,
766 strict=self.strict)
767 else:
768 response = self.response_class(self.sock, strict=self.strict)
770 response.begin()
771 assert response.will_close != _UNKNOWN
772 self.__state = _CS_IDLE
774 if response.will_close:
775 # this effectively passes the connection to the response
776 self.close()
777 else:
778 # remember this, so we can tell when it is complete
779 self.__response = response
781 return response
783 # The next several classes are used to define FakeSocket,a socket-like
784 # interface to an SSL connection.
786 # The primary complexity comes from faking a makefile() method. The
787 # standard socket makefile() implementation calls dup() on the socket
788 # file descriptor. As a consequence, clients can call close() on the
789 # parent socket and its makefile children in any order. The underlying
790 # socket isn't closed until they are all closed.
792 # The implementation uses reference counting to keep the socket open
793 # until the last client calls close(). SharedSocket keeps track of
794 # the reference counting and SharedSocketClient provides an constructor
795 # and close() method that call incref() and decref() correctly.
797 class SharedSocket:
799 def __init__(self, sock):
800 self.sock = sock
801 self._refcnt = 0
803 def incref(self):
804 self._refcnt += 1
806 def decref(self):
807 self._refcnt -= 1
808 assert self._refcnt >= 0
809 if self._refcnt == 0:
810 self.sock.close()
812 def __del__(self):
813 self.sock.close()
815 class SharedSocketClient:
817 def __init__(self, shared):
818 self._closed = 0
819 self._shared = shared
820 self._shared.incref()
821 self._sock = shared.sock
823 def close(self):
824 if not self._closed:
825 self._shared.decref()
826 self._closed = 1
827 self._shared = None
829 class SSLFile(SharedSocketClient):
830 """File-like object wrapping an SSL socket."""
832 BUFSIZE = 8192
834 def __init__(self, sock, ssl, bufsize=None):
835 SharedSocketClient.__init__(self, sock)
836 self._ssl = ssl
837 self._buf = ''
838 self._bufsize = bufsize or self.__class__.BUFSIZE
840 def _read(self):
841 buf = ''
842 # put in a loop so that we retry on transient errors
843 while 1:
844 try:
845 buf = self._ssl.read(self._bufsize)
846 except socket.sslerror, err:
847 if (err[0] == socket.SSL_ERROR_WANT_READ
848 or err[0] == socket.SSL_ERROR_WANT_WRITE):
849 continue
850 if (err[0] == socket.SSL_ERROR_ZERO_RETURN
851 or err[0] == socket.SSL_ERROR_EOF):
852 break
853 raise
854 except socket.error, err:
855 if err[0] == errno.EINTR:
856 continue
857 if err[0] == errno.EBADF:
858 # XXX socket was closed?
859 break
860 raise
861 else:
862 break
863 return buf
865 def read(self, size=None):
866 L = [self._buf]
867 avail = len(self._buf)
868 while size is None or avail < size:
869 s = self._read()
870 if s == '':
871 break
872 L.append(s)
873 avail += len(s)
874 all = "".join(L)
875 if size is None:
876 self._buf = ''
877 return all
878 else:
879 self._buf = all[size:]
880 return all[:size]
882 def readline(self):
883 L = [self._buf]
884 self._buf = ''
885 while 1:
886 i = L[-1].find("\n")
887 if i >= 0:
888 break
889 s = self._read()
890 if s == '':
891 break
892 L.append(s)
893 if i == -1:
894 # loop exited because there is no more data
895 return "".join(L)
896 else:
897 all = "".join(L)
898 # XXX could do enough bookkeeping not to do a 2nd search
899 i = all.find("\n") + 1
900 line = all[:i]
901 self._buf = all[i:]
902 return line
904 class FakeSocket(SharedSocketClient):
906 class _closedsocket:
907 def __getattr__(self, name):
908 raise error(9, 'Bad file descriptor')
910 def __init__(self, sock, ssl):
911 sock = SharedSocket(sock)
912 SharedSocketClient.__init__(self, sock)
913 self._ssl = ssl
915 def close(self):
916 SharedSocketClient.close(self)
917 self._sock = self.__class__._closedsocket()
919 def makefile(self, mode, bufsize=None):
920 if mode != 'r' and mode != 'rb':
921 raise UnimplementedFileMode()
922 return SSLFile(self._shared, self._ssl, bufsize)
924 def send(self, stuff, flags = 0):
925 return self._ssl.write(stuff)
927 sendall = send
929 def recv(self, len = 1024, flags = 0):
930 return self._ssl.read(len)
932 def __getattr__(self, attr):
933 return getattr(self._sock, attr)
936 class HTTPSConnection(HTTPConnection):
937 "This class allows communication via SSL."
939 default_port = HTTPS_PORT
941 def __init__(self, host, port=None, key_file=None, cert_file=None,
942 strict=None):
943 HTTPConnection.__init__(self, host, port, strict)
944 self.key_file = key_file
945 self.cert_file = cert_file
947 def connect(self):
948 "Connect to a host on a given (SSL) port."
950 sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
951 sock.connect((self.host, self.port))
952 realsock = sock
953 if hasattr(sock, "_sock"):
954 realsock = sock._sock
955 ssl = socket.ssl(realsock, self.key_file, self.cert_file)
956 self.sock = FakeSocket(sock, ssl)
959 class HTTP:
960 "Compatibility class with httplib.py from 1.5."
962 _http_vsn = 10
963 _http_vsn_str = 'HTTP/1.0'
965 debuglevel = 0
967 _connection_class = HTTPConnection
969 def __init__(self, host='', port=None, strict=None):
970 "Provide a default host, since the superclass requires one."
972 # some joker passed 0 explicitly, meaning default port
973 if port == 0:
974 port = None
976 # Note that we may pass an empty string as the host; this will throw
977 # an error when we attempt to connect. Presumably, the client code
978 # will call connect before then, with a proper host.
979 self._setup(self._connection_class(host, port, strict))
981 def _setup(self, conn):
982 self._conn = conn
984 # set up delegation to flesh out interface
985 self.send = conn.send
986 self.putrequest = conn.putrequest
987 self.endheaders = conn.endheaders
988 self.set_debuglevel = conn.set_debuglevel
990 conn._http_vsn = self._http_vsn
991 conn._http_vsn_str = self._http_vsn_str
993 self.file = None
995 def connect(self, host=None, port=None):
996 "Accept arguments to set the host/port, since the superclass doesn't."
998 if host is not None:
999 self._conn._set_hostport(host, port)
1000 self._conn.connect()
1002 def getfile(self):
1003 "Provide a getfile, since the superclass' does not use this concept."
1004 return self.file
1006 def putheader(self, header, *values):
1007 "The superclass allows only one value argument."
1008 self._conn.putheader(header, '\r\n\t'.join(values))
1010 def getreply(self):
1011 """Compat definition since superclass does not define it.
1013 Returns a tuple consisting of:
1014 - server status code (e.g. '200' if all goes well)
1015 - server "reason" corresponding to status code
1016 - any RFC822 headers in the response from the server
1018 try:
1019 response = self._conn.getresponse()
1020 except BadStatusLine, e:
1021 ### hmm. if getresponse() ever closes the socket on a bad request,
1022 ### then we are going to have problems with self.sock
1024 ### should we keep this behavior? do people use it?
1025 # keep the socket open (as a file), and return it
1026 self.file = self._conn.sock.makefile('rb', 0)
1028 # close our socket -- we want to restart after any protocol error
1029 self.close()
1031 self.headers = None
1032 return -1, e.line, None
1034 self.headers = response.msg
1035 self.file = response.fp
1036 return response.status, response.reason, response.msg
1038 def close(self):
1039 self._conn.close()
1041 # note that self.file == response.fp, which gets closed by the
1042 # superclass. just clear the object ref here.
1043 ### hmm. messy. if status==-1, then self.file is owned by us.
1044 ### well... we aren't explicitly closing, but losing this ref will
1045 ### do it
1046 self.file = None
1048 if hasattr(socket, 'ssl'):
1049 class HTTPS(HTTP):
1050 """Compatibility with 1.5 httplib interface
1052 Python 1.5.2 did not have an HTTPS class, but it defined an
1053 interface for sending http requests that is also useful for
1054 https.
1057 _connection_class = HTTPSConnection
1059 def __init__(self, host='', port=None, key_file=None, cert_file=None,
1060 strict=None):
1061 # provide a default host, pass the X509 cert info
1063 # urf. compensate for bad input.
1064 if port == 0:
1065 port = None
1066 self._setup(self._connection_class(host, port, key_file,
1067 cert_file, strict))
1069 # we never actually use these for anything, but we keep them
1070 # here for compatibility with post-1.5.2 CVS.
1071 self.key_file = key_file
1072 self.cert_file = cert_file
1075 class HTTPException(Exception):
1076 # Subclasses that define an __init__ must call Exception.__init__
1077 # or define self.args. Otherwise, str() will fail.
1078 pass
1080 class NotConnected(HTTPException):
1081 pass
1083 class InvalidURL(HTTPException):
1084 pass
1086 class UnknownProtocol(HTTPException):
1087 def __init__(self, version):
1088 self.args = version,
1089 self.version = version
1091 class UnknownTransferEncoding(HTTPException):
1092 pass
1094 class UnimplementedFileMode(HTTPException):
1095 pass
1097 class IncompleteRead(HTTPException):
1098 def __init__(self, partial):
1099 self.args = partial,
1100 self.partial = partial
1102 class ImproperConnectionState(HTTPException):
1103 pass
1105 class CannotSendRequest(ImproperConnectionState):
1106 pass
1108 class CannotSendHeader(ImproperConnectionState):
1109 pass
1111 class ResponseNotReady(ImproperConnectionState):
1112 pass
1114 class BadStatusLine(HTTPException):
1115 def __init__(self, line):
1116 self.args = line,
1117 self.line = line
1119 # for backwards compatibility
1120 error = HTTPException
1122 class LineAndFileWrapper:
1123 """A limited file-like object for HTTP/0.9 responses."""
1125 # The status-line parsing code calls readline(), which normally
1126 # get the HTTP status line. For a 0.9 response, however, this is
1127 # actually the first line of the body! Clients need to get a
1128 # readable file object that contains that line.
1130 def __init__(self, line, file):
1131 self._line = line
1132 self._file = file
1133 self._line_consumed = 0
1134 self._line_offset = 0
1135 self._line_left = len(line)
1137 def __getattr__(self, attr):
1138 return getattr(self._file, attr)
1140 def _done(self):
1141 # called when the last byte is read from the line. After the
1142 # call, all read methods are delegated to the underlying file
1143 # obhect.
1144 self._line_consumed = 1
1145 self.read = self._file.read
1146 self.readline = self._file.readline
1147 self.readlines = self._file.readlines
1149 def read(self, amt=None):
1150 assert not self._line_consumed and self._line_left
1151 if amt is None or amt > self._line_left:
1152 s = self._line[self._line_offset:]
1153 self._done()
1154 if amt is None:
1155 return s + self._file.read()
1156 else:
1157 return s + self._file.read(amt - len(s))
1158 else:
1159 assert amt <= self._line_left
1160 i = self._line_offset
1161 j = i + amt
1162 s = self._line[i:j]
1163 self._line_offset = j
1164 self._line_left -= amt
1165 if self._line_left == 0:
1166 self._done()
1167 return s
1169 def readline(self):
1170 s = self._line[self._line_offset:]
1171 self._done()
1172 return s
1174 def readlines(self, size=None):
1175 L = [self._line[self._line_offset:]]
1176 self._done()
1177 if size is None:
1178 return L + self._file.readlines()
1179 else:
1180 return L + self._file.readlines(size)
1182 def test():
1183 """Test this module.
1185 A hodge podge of tests collected here, because they have too many
1186 external dependencies for the regular test suite.
1189 import sys
1190 import getopt
1191 opts, args = getopt.getopt(sys.argv[1:], 'd')
1192 dl = 0
1193 for o, a in opts:
1194 if o == '-d': dl = dl + 1
1195 host = 'www.python.org'
1196 selector = '/'
1197 if args[0:]: host = args[0]
1198 if args[1:]: selector = args[1]
1199 h = HTTP()
1200 h.set_debuglevel(dl)
1201 h.connect(host)
1202 h.putrequest('GET', selector)
1203 h.endheaders()
1204 status, reason, headers = h.getreply()
1205 print 'status =', status
1206 print 'reason =', reason
1207 print "read", len(h.getfile().read())
1208 print
1209 if headers:
1210 for header in headers.headers: print header.strip()
1211 print
1213 # minimal test that code to extract host from url works
1214 class HTTP11(HTTP):
1215 _http_vsn = 11
1216 _http_vsn_str = 'HTTP/1.1'
1218 h = HTTP11('www.python.org')
1219 h.putrequest('GET', 'http://www.python.org/~jeremy/')
1220 h.endheaders()
1221 h.getreply()
1222 h.close()
1224 if hasattr(socket, 'ssl'):
1226 for host, selector in (('sourceforge.net', '/projects/python'),
1228 print "https://%s%s" % (host, selector)
1229 hs = HTTPS()
1230 hs.set_debuglevel(dl)
1231 hs.connect(host)
1232 hs.putrequest('GET', selector)
1233 hs.endheaders()
1234 status, reason, headers = hs.getreply()
1235 print 'status =', status
1236 print 'reason =', reason
1237 print "read", len(hs.getfile().read())
1238 print
1239 if headers:
1240 for header in headers.headers: print header.strip()
1241 print
1243 if __name__ == '__main__':
1244 test()