This patch changes the original implementation of HTTPResponse'
retry() function so that the response instance is reused
instead of replaced with a new one (after a Retry exception was
raised).  This fixes a bug in zopedoctests' http() function
(Testing/ZopeTestCase/zopedoctest/functional.py:113): the doctest code
assumes that the HTTPResponse instance passed to publish_module() (line
177) is used to handle to complete request, so it can be used to get the
status, headers etc later on (lines 183-186).  Normally this is okay,
but raising a Retry will create a new response instance, which will then
hold that data (relevant for evaluating the doctest) while the original
(passed in) instance is still empty.  So to fix this (quickly) retry()
now cleans up and returns itself...

--- ZPublisher/HTTPResponse.py~	2006-05-30 07:08:18.000000000 +0200
+++ ZPublisher/HTTPResponse.py	2006-09-12 19:51:52.000000000 +0200
@@ -182,19 +182,20 @@
 
     def retry(self):
         """Return a response object to be used in a retry attempt
         """
 
         # This implementation is a bit lame, because it assumes that
         # only stdout stderr were passed to the constructor. OTOH, I
         # think that that's all that is ever passed.
 
-        return self.__class__(stdout=self.stdout, stderr=self.stderr)
+        self.__init__(stdout=self.stdout, stderr=self.stderr)
+        return self
 
     _shutdown_flag = None
     def _requestShutdown(self, exitCode=0):
         """Request that the server shut down with exitCode after fulfilling
            the current request."""
         import ZServer
         ZServer.exit_code = exitCode
         self._shutdown_flag = 1