X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~mdw/git/hippotat/blobdiff_plain/4224dc197474ed0a2535cdb100193fffaa697d0e..refs/heads/mdw/fixes:/hippotat

diff --git a/hippotat b/hippotat
index 997996d..6aa3121 100755
--- a/hippotat
+++ b/hippotat
@@ -26,13 +26,15 @@ from hippotatlib import *
 
 import twisted.web
 import twisted.web.client
+import urllib.parse
 
 import io
 
 class GeneralResponseConsumer(twisted.internet.protocol.Protocol):
-  def __init__(self, cl, req, desc):
+  def __init__(self, cl, req, resp, desc):
     self._cl = cl
     self._req = req
+    self._resp = resp
     self._desc = desc
 
   def _log(self, dflag, msg, **kwargs):
@@ -41,9 +43,17 @@ class GeneralResponseConsumer(twisted.internet.protocol.Protocol):
   def connectionMade(self):
     self._log(DBG.HTTP_CTRL, 'connectionMade')
 
+  def connectionLostOK(self, reason):
+    return (reason.check(twisted.web.client.ResponseDone) or
+            reason.check(twisted.web.client.PotentialDataLoss))
+    # twisted.web.client.PotentialDataLoss is an entirely daft
+    # exception.  It will occur every time if the origin server does
+    # not provide a Content-Length.  (hippotatd does, of course, but
+    # the HTTP transaction might be proxied.)
+
 class ResponseConsumer(GeneralResponseConsumer):
   def __init__(self, cl, req, resp):
-    super().__init__(cl, req, 'RC')
+    super().__init__(cl, req, resp, 'RC')
     ssddesc = '[%s] %s' % (id(req), self._desc)
     self._ssd = SlipStreamDecoder(ssddesc, partial(queue_inbound, cl.ipif))
     self._log(DBG.HTTP_CTRL, '__init__')
@@ -58,7 +68,7 @@ class ResponseConsumer(GeneralResponseConsumer):
   def connectionLost(self, reason):
     reason_msg = 'connectionLost ' + str(reason)
     self._log(DBG.HTTP_CTRL, reason_msg)
-    if not reason.check(twisted.web.client.ResponseDone):
+    if not self.connectionLostOK(reason):
       self._latefailure(reason_msg)
       return
     try:
@@ -78,8 +88,7 @@ class ResponseConsumer(GeneralResponseConsumer):
 
 class ErrorResponseConsumer(GeneralResponseConsumer):
   def __init__(self, cl, req, resp):
-    super().__init__(cl, req, 'ERROR-RC')
-    self._resp = resp
+    super().__init__(cl, req, resp, 'ERROR-RC')
     self._m = b''
     try:
       self._phrase = resp.phrase.decode('utf-8')
@@ -96,7 +105,7 @@ class ErrorResponseConsumer(GeneralResponseConsumer):
       mbody = self._m.decode('utf-8')
     except Exception:
       mbody = repr(self._m)
-    if not reason.check(twisted.web.client.ResponseDone):
+    if not self.connectionLostOK(reason):
       mbody += ' || ' + str(reason)
     self._cl.req_err(self._req,
             "FAILED %d %s | %s"
@@ -215,8 +224,7 @@ class Client():
       cl.log(DBG.HTTP_FULL, 'requesting: ' + str(mime))
 
       hh = { 'User-Agent': ['hippotat'],
-             'Content-Type': ['multipart/form-data; boundary="b"'],
-             'Content-Length': [str(len(mime))] }
+             'Content-Type': ['multipart/form-data; boundary="b"'] }
 
       bytesreader = io.BytesIO(mime)
       producer = twisted.web.client.FileBodyProducer(bytesreader)
@@ -238,6 +246,17 @@ class Client():
 
 clients = [ ]
 
+def encode_url(urlstr):
+  # Oh, this is a disaster.  We're given a URL as a `str', but the underlying
+  # machinery insists on having `bytes'.  Assume we've been given a sensible
+  # URL, with escaping in all of the necessary places, except that it may
+  # contain non-ASCII characters: then encode as UTF-8 and squash the top-
+  # bit-set bytes down to percent escapes.
+  #
+  # This conses like it's going out of fashion, but it gets the job done.
+  return b''.join(bytes([b]) if b < 128 else '%%%02X' % b
+                  for b in urlstr.encode('utf-8'))
+
 def process_cfg(_opts, putative_servers, putative_clients):
   global clients
 
@@ -262,7 +281,7 @@ def process_cfg(_opts, putative_servers, putative_clients):
       try: c.ifname     = srch(cfg_get_raw, 'ifname_client')
       except NoOptionError: pass
 
-      try: c.url = srch(cfg.get,'url')
+      try: c.url = encode_url(srch(cfg.get,'url'))
       except NoOptionError:
         cfg_process_saddrs(c, ss)
         c.url = c.saddrs[0].url()