vsgi: Write the response head on disposal

The router cannot initialize with 'chunked' encoding, because it would
break empty responses. It must therefore be an explicit choice from the
end-user knowingly that something will be produced.

There's no need for the post-condition since we can be assured that the
head will be written on disposal.

Update the documentation to to include that behaviour.

Author: arteymix

docs/vsgi/index.rst

@@ -42,23 +42,8 @@ line and headers synchronously in the connection stream before returning an
 
 .. _GLib.OutputStream: http://valadoc.org/#!api=gio-2.0/GLib.OutputStream
 
-.. warning::
-
-    The :doc:`response` body must be accessed at least once during the
-    processing to ensure that the headers are written and filters applied.
-
-Some VSGI implementations rely on stream filtering to produce proper responses
-(chunked or gzipped ones) and these are applied when the body is accessed for
-the first time.
-
-.. code:: vala
-
-    new Server ("org.vsgi.App", (req, res) => {
-        res.status = 200;
-        res.headers.append ("Transfer-Encoding", "chunked");
-        // chunked encoding will be applied
-        res.body.write_all ("Hello world!".data, null);
-    });
+In case the body is not accessed, the head is written synchronously on disposal
+of the :doc:`response` object.
 
 Asynchronous processing
 -----------------------

src/valum-router.vala

@@ -421,16 +421,9 @@ namespace Valum {
 		 * raise a {@link Valum.Status.ClientError.METHOD_NOT_ALLOWED}, otherwise
 		 * raise a {@link Valum.Status.ClientError.NOT_FOUND} exception.
 		 *
-		 * The response is initialized with 'chunked' transfer encoding since
-		 * most processing are generally based on stream.
-		 *
 		 * @since 0.1
 		 */
 		public void handle (Request req, Response res) {
-			// sane initialization
-			if (req.http_version == Soup.HTTPVersion.@1_1)
-				res.headers.set_encoding (Soup.Encoding.CHUNKED);
-
 			// initial invocation
 			this.invoke (req, res, () => {
 				var stack = new Queue<Value?> ();

src/vsgi-response.vala

@@ -87,7 +87,7 @@ namespace VSGI {
 						this.write_head (out bytes_written);
 					}
 				} catch (IOError err) {
-					warning ("could not write the head in the connection stream");
+					warning ("could not write the head in the connection stream: %s", err.message);
 				}
 
 				return this.request.connection.output_stream;
@@ -183,5 +183,21 @@ namespace VSGI {
 			return this.head_written;
 		}
 #endif
+
+		/**
+		 * Write the head before disposing references to other objects.
+		 */
+		public override void dispose () {
+			try {
+				if (!head_written) {
+					size_t bytes_written;
+					write_head (out bytes_written);
+				}
+			} catch (IOError err) {
+				warning ("could not write the head in the connection stream: %s", err.message);
+			} finally {
+				base.dispose ();
+			}
+		}
 	}
 }

src/vsgi-server.vala

@@ -73,7 +73,7 @@ namespace VSGI {
 		 * Once dispatched, the {@link Response.head_written} property is
 		 * expected to be true unless its reference still held somewhere else.
 		 */
-		protected void dispatch (Request req, Response res) ensures (res.ref_count > 1 || res.head_written) {
+		protected void dispatch (Request req, Response res) {
 			_application (req, res);
 		}
 	}