紧接着上文我们的Action已经执行完毕，现在需要把Action的返回结果转化为HttpResponseMessage 实例，我们也知道转化工作主要在HttpRequestMessage的CreateResponse附加方法中，
public static HttpResponseMessage CreateResponse(this HttpRequestMessage request, HttpStatusCode statusCode, T value, HttpConfiguration configuration) { if (request == null) { throw Error.ArgumentNull("request"); } configuration = configuration ?? request.GetConfiguration(); if (configuration == null) { throw Error.InvalidOperation(SRResources.HttpRequestMessageExtensions_NoConfiguration); } IContentNegotiator contentNegotiator = configuration.Services.GetContentNegotiator(); if (contentNegotiator == null) { throw Error.InvalidOperation(SRResources.HttpRequestMessageExtensions_NoContentNegotiator, typeof(IContentNegotiator).FullName); } IEnumerable formatters = configuration.Formatters; // Run content negotiation ContentNegotiationResult result = contentNegotiator.Negotiate(typeof(T), request, formatters); if (result == null) { // no result from content negotiation indicates that 406 should be sent. return new HttpResponseMessage { StatusCode = HttpStatusCode.NotAcceptable, RequestMessage = request, }; } else { MediaTypeHeaderValue mediaType = result.MediaType; return new HttpResponseMessage { // At this point mediaType should be a cloned value (the content negotiator is responsible for returning a new copy) Content = new ObjectContent (value, result.Formatter, mediaType), StatusCode = statusCode, RequestMessage = request }; } }
首先这里需要一个IContentNegotiator实例，这里有这么一句代码：
IContentNegotiator contentNegotiator = configuration.Services.GetContentNegotiator();在DefaultServices中有 SetSingle<IContentNegotiator>(new DefaultContentNegotiator());这句，我们知道默认的contentNegotiator是DefaultContentNegotiator实例。DefaultContentNegotiator构造函数也比较普通，这里我们还需要一个数据的格式化Formatters，这里有这么一句 IEnumerable<MediaTypeFormatter> formatters = configuration.Formatters;，在
曾提到Formatters主要有JsonMediaTypeFormatter,XmlMediaTypeFormatter, FormUrlEncodedMediaTypeFormatter,JQueryMvcFormUrlEncodedFormatter这4个。接着我们知道要干什么了，需要把我们的value转换为需要的格式，这里创建一个 ContentNegotiationResult result = contentNegotiator.Negotiate(typeof(T), request, formatters);实例，其中Negotiate的主要实现如下：public virtual ContentNegotiationResult Negotiate(Type type, HttpRequestMessage request, IEnumerable<MediaTypeFormatter> formatters)
{ ....。。。。。。 // Go through each formatter to compute how well it matches. Collection<MediaTypeFormatterMatch> matches = ComputeFormatterMatches(type, request, formatters); // Select best formatter match among the matches MediaTypeFormatterMatch bestFormatterMatch = SelectResponseMediaTypeFormatter(matches); // We found a best formatter if (bestFormatterMatch != null) { // Find the best character encoding for the selected formatter Encoding bestEncodingMatch = SelectResponseCharacterEncoding(request, bestFormatterMatch.Formatter); if (bestEncodingMatch != null) { bestFormatterMatch.MediaType.CharSet = bestEncodingMatch.WebName; } MediaTypeHeaderValue bestMediaType = bestFormatterMatch.MediaType; MediaTypeFormatter bestFormatter = bestFormatterMatch.Formatter.GetPerRequestFormatterInstance(type, request, bestMediaType); return new ContentNegotiationResult(bestFormatter, bestMediaType); } return null; } 这里的Negotiate方法实现比较复杂，首先找到当前请求支持的一个MediaTypeFormatterMatch集合，然后再从这个集合中找到找最合适一个bestFormatterMatch （在实际开发中这个实例往往都是JsonMediaTypeFormatter），如果找到那么我们继续找一个bestEncodingMatch （最合适的一个编码很多时候这里是utf-8），这里最后调用GetPerRequestFormatterInstance方法得到一个MediaTypeFormatter实例，然后用这么实例创建一个ContentNegotiationResult实例。现在我们回到CreateResponse方法中来，最后直接返回一个 HttpResponseMessage实例。
MediaTypeHeaderValue mediaType = result.MediaType;
return new HttpResponseMessage
{ // At this point mediaType should be a cloned value (the content negotiator is responsible for returning a new copy) Content = new ObjectContent<T>(value, result.Formatter, mediaType), StatusCode = statusCode, RequestMessage = request };这里的ObjectContent的创建也比较一般，我们也就忽略它吧。
现在我们回到HttpControllerHandler的BeginProcessRequest方法中来，
Task responseBodyTask = _server.Value.SendAsync(request, CancellationToken.None)
.Then(response => ConvertResponse(httpContextBase, response, request));SendAsync已经执行完毕了，我们再来看看ConvertResponse吧，它主要就是把HttpResponseMessage的信息写到HttpResponseBase信息做。
internal static Task ConvertResponse(HttpContextBase httpContextBase, HttpResponseMessage response, HttpRequestMessage request)
{ Contract.Assert(httpContextBase != null); Contract.Assert(request != null); // A null response creates a 500 with no content if (response == null) { CreateEmptyErrorResponse(httpContextBase.Response); return TaskHelpers.Completed(); } CopyResponseStatusAndHeaders(httpContextBase, response); CacheControlHeaderValue cacheControl = response.Headers.CacheControl; // TODO 335085: Consider this when coming up with our caching story if (cacheControl == null) { // DevDiv2 #332323. ASP.NET by default always emits a cache-control: private header. // However, we don't want requests to be cached by default. // If nobody set an explicit CacheControl then explicitly set to no-cache to override the // default behavior. This will cause the following response headers to be emitted: // Cache-Control: no-cache // Pragma: no-cache // Expires: -1 httpContextBase.Response.Cache.SetCacheability(HttpCacheability.NoCache); } // Asynchronously write the response body. If there is no body, we use // a completed task to share the Finally() below. // The response-writing task will not fault -- it handles errors internally. Task writeResponseContentTask = (response.Content == null) ? TaskHelpers.Completed() : WriteResponseContentAsync(httpContextBase, response, request); return writeResponseContentTask.Finally(() => { request.DisposeRequestResources(); request.Dispose(); response.Dispose(); }); }这里的CopyResponseStatusAndHeaders方法比较简单我也不多说了，就是把HttpResponseMessage中的StatusAndHeaders信息写到 httpContextBase.Response中去，
我们还是关注一下WriteResponseContentAsync的实现吧：
internal static Task WriteResponseContentAsync(HttpContextBase httpContextBase, HttpResponseMessage response, HttpRequestMessage request)
{ HttpResponseBase httpResponseBase = httpContextBase.Response; HttpContent responseContent = response.Content; var unused = response.Content.Headers.ContentLength; CopyHeaders(response.Content.Headers, httpContextBase); // Select output buffering based on the user-controlled buffering policy bool isBuffered = _bufferPolicySelector.Value != null ? _bufferPolicySelector.Value.UseBufferedOutputStream(response) : true; httpResponseBase.BufferOutput = isBuffered; return isBuffered ? WriteBufferedResponseContentAsync(httpContextBase, responseContent, request) : WriteStreamedResponseContentAsync(httpResponseBase, responseContent); }前面CopyResponseStatusAndHeaders方法把HttpResponseMessage中的Headers信息写到 httpContextBase.Response中去了，这里继续把response.Content.Headers信息写到 httpContextBase.Response中去。
private static readonly Lazy<IHostBufferPolicySelector> _bufferPolicySelector =
new Lazy<IHostBufferPolicySelector>(() => GlobalConfiguration.Configuration.Services.GetHostBufferPolicySelector());在GlobalConfiguration中有这么一句 config.Services.Replace(typeof(IHostBufferPolicySelector), new WebHostBufferPolicySelector());所以我们知道_bufferPolicySelector这里其实是一个WebHostBufferPolicySelector实例，这里调用WebHostBufferPolicySelector的UseBufferedOutputStream方法来获取当前输出信息是否采用输出缓存，一般情况下这个个方法返回true，主要实现如下：
public virtual bool UseBufferedOutputStream(HttpResponseMessage response)
{ HttpContent content = response.Content; if (content != null) { long? contentLength = content.Headers.ContentLength; if (contentLength.HasValue && contentLength.Value >= 0) { return false; } return !(content is StreamContent || content is PushStreamContent); } return false; } }一般情况下response.Content.Headers.ContentLength为null。所以后面调用的是WriteBufferedResponseContentAsync方法，而该方法的主要实现就一句代码
Task writeResponseContentTask = HttpResponseMessage.Content.CopyToAsync(HttpContextBase.Response.OutputStream);把当前HttpResponseMessage.Content中的数据全部写到HttpContextBase.Response.OutputStream中去。到这里我们的BeginProcessRequest就差不多执行完了，后面就是调用回调函数的事情了。
到本章为止，整个web api的主要流程就说完了，这个系列中有关参数的具体绑定和返回值的格式化我是忽略了的，他们的实现都相对比较复杂，后面再抽时间来看看他们是如何实现 吧，个人对web api也不是很熟悉，发现它和mvc 中的很多代码相似，相比之下mvc的使用比web api要广泛得多，所以这里建议大家多读读mvc的源码，读了之后再来读web api源码相对要轻松很多了。