Avoid sending Content-Length header when body is an empty string.

Some sites can't handle "Content-Length: 0" header, but twisted
HTTPClientFactory adds Content-Length header unless body is None.

scrapy enforces request.body usage as string, using None is not
possible.

thanks Matt for report, See:
http://groups.google.com/group/scrapy-users/browse_thread/thread/380ffa111879989e?hl=en

--HG--
extra : convert_revision : svn%3Ab85faa78-f9eb-468e-a121-7cced6da292c%40841

scrapy/trunk/docs/ref/request-response.rst

@@ -41,8 +41,10 @@ Request objects
     ``meta`` is a dict containing the initial values for the
     :attr:`Request.meta` attribute. If passed, the dict will be shallow copied.
 
-    ``body`` is a str or unicode containing the request body. If ``body`` is
-    a `unicode` it's encoded to str using the `encoding` passed.
+    ``body`` is a str or unicode containing the request body.
+    If ``body`` is a `unicode` it's encoded to str using the `encoding` passed.
+    If ``body`` is None, an empty string is stored.
+    In any case, the final stored value will be a string (never unicode, never None).
 
     ``headers`` is a multi-valued dict containing the headers of this request
 

scrapy/trunk/scrapy/core/downloader/handlers.py

@@ -53,7 +53,7 @@ def create_factory(request, spider):
     agent = request.headers.pop('user-agent', default_agent)
     factory = HTTPClientFactory(url=url, # never pass unicode urls to twisted
                                 method=request.method,
-                                postdata=request.body,
+                                postdata=request.body or None,
                                 headers=request.headers,
                                 agent=agent,
                                 cookies=request.cookies,