Prototype of chunked transfer encoded POST.

net/url_request/url_request.h

Index: net/url_request/url_request.h
diff --git a/net/url_request/url_request.h b/net/url_request/url_request.h
index 3561a294624dccccc8eeb5d653a794eb13b10034..f4a1d875fcf9c610fe1965d4d80d837827c57828 100644
--- a/net/url_request/url_request.h
+++ b/net/url_request/url_request.h
@@ -312,6 +312,21 @@ class URLRequest : public base::NonThreadSafe {
     AppendFileRangeToUpload(file_path, 0, kuint64max, base::Time());
   }
 
+  // Indicates that the request body should be sent using Chunked-Transfer

[wtc, 2011/01/20 00:29:47]

Nit: Chunked-Transfer => chunked transfer

This comment also applies to line 320 and line 327 below.

The HTTP 1.1 RFC (RFC 2616) actually hyphenates it like this:
  chunked transfer-coding
I would omit the hyphen.

[willchan no longer on Chromium, 2011/01/20 08:51:04]

Push back if you feel I'm being unreasonable. I just saw this and felt like I
had to chime in. We're adding an extra 3 member functions to this API. When
reading this API, it's not obvious what's the difference between all of these
APIs. It also seems to be the case that the UploadData* itself is directly
exposed.

I'm a bit concerned about the complication of the API. I feel like perhaps there
should be just two functions, a simple one and a complicated one. The simple one
could just be AppendBytesToUpload. For everything else, I think URLRequest
should just expose the UploadData since that's what we're doing already.

I'm fine with having these member functions be in URLFetcher, since that
object's purpose is to help simplify access to URLRequest from chrome/browser,
and it's far less cluttered than URLRequest. But I personally would like to push
back on additions to the URLRequest interface except where absolutely necessary.

[Satish, 2011/01/20 18:02:36]

I'm happy to make this change. @wtc, @vandebo - are you ok if I change the
URLRequest API as suggested here?

[vandebo (ex-Chrome), 2011/01/20 20:48:45]

Feel free to make any changes that you and Will agree on here.

Will, it seems that it might be better to clean up this API in a separate CL.  I
see your point about UploadData being exposed, but it seems that Satish has
followed the convention that's already been established here.

+  // encoding. This method may only be called before Start() is called.
+  void EnableChunkedUpload();
+
+  // Appends the given bytes to the requests's upload data to be sent

[wtc, 2011/01/20 00:29:47]

Nit: requests's => request's

+  // immediately via chunked-transfer encoding. When all data has been sent,
+  // call MarkEndOfChunks() to indicate the end of stream.

[wtc, 2011/01/20 00:29:47]

Nit: end of stream => end of upload data
                or    end of request body

+  //
+  // This method may be called only after calling EnableChunkedUpload() and
+  // Start().
+  void AppendChunkToUpload(const char* bytes, int bytes_len);
+
+  // Indicates the end of a chunked-transfer encoded request body.
+  void MarkEndOfChunks();
+
   // Set the upload data directly.
   void set_upload(net::UploadData* upload);