ASP.NET MVC View Engine Resolution Sequence

I created a simple ASP.NET MVC version 1.0 application. I have a ProductController which has one action Index. In the view, I created a corresponding Index.aspx under Product 开发者_开发知识库subfolder.

Then I referenced the Spark dll and created Index.spark under the same Product view folder. The Application_Start looks like

    protected void Application_Start()
    {
        RegisterRoutes(RouteTable.Routes);

        ViewEngines.Engines.Clear();
        ViewEngines.Engines.Add(new Spark.Web.Mvc.SparkViewFactory());

        ViewEngines.Engines.Add(new WebFormViewEngine());

    }

My expectation is that since the Spark engine registers before default WebFormViewEngine, when browse the Index action in Product controller, the Spark engine should be used, and WebFormViewEngine should be used for all other urls.

However, the test shows that the Index action for Product controller also uses the WebFormViewEngine.

If I comment out the registration of WebFormViewEnginer (the last line in the code), I can see that the Index action is rendered by Spark engine and the rest urls generates an error (since the defualt engine is gone), it proves that all my Spark code is correct.

Now my question is how the view engine is resolved? Why the registration sequence does not take effect?

The order in which you register the view engines doesn't matter (much). Rather, view engines take a set of ViewLocationFormats, and if a particular view path fits the formatted name, that engine will be used. Only if you have conflicting formats does the registration order matter.

In the case of spark, views should have the .spark extension. WebFormViewEngine will respond to any with .aspx or .ascx extensions. And of course, as mentioned above, you can override any of this by changing the ViewLocationFormats supplied to the individual view engines.

Updated:

I took a look through the source of both SparkViewFactory and WebFormViewEngine (or more specifically, VirtualPathProviderViewEngine, which the latter derives from), and I can tell you why you're seeing this strange behaviour.

First of all, the Find method in the ViewEngineCollection class works like this (simplified):

foreach (IViewEngine engine in Items) {
    // Query engine for cached view
}

foreach (IViewEngine engine in Items) {
    // Query engine for uncached view
}

In other words, it will always try to find a cached view, in any engine, before resorting to uncached mode.

The way in which individual view engines implement this is the second overload of the FindView method, which takes a bool argument named useCache.

However, and here's where it all gets weird - the VirtualPathProviderViewEngine and SparkViewEngine have very different ideas of what the useCache argument means. There's too much code to repost here but the basic idea is:

• The SparkViewFactory will look only in the cache if useCache is true. If it doesn't find anything, it automatically returns a "cache miss result" - i.e. nothing. On the other hand, if useCache is false, it will not look in the cache at all, it will skip the cache-checking step and go through the normal motions to resolve and create an actual view.

• The VirtualPathProviderViewEngine, on the other hand, looks in the cache if useCache is true, and if it doesn't find the view in the cache, it goes off and creates a new one and adds that to the cache.

Both of these approaches work with respect to the way ViewEngineCollection performs its search.

• In the case of spark, it "misses" on the first iteration of view engines, but "hits" on the second, and after that the view is added to the cache. No problem.

• In the case of VirtualPathProviderViewEngine, it "misses" internally but returns a "hit" anyway on the first iteration, at which point the view is now cached.

So you should be able to see where the problem is here. The VirtualPathProviderViewEngine only appears to be taking precedence over the SparkViewEngine because the former always succeeds on the first (cached) iteration, but Spark only succeeds on the second (uncached) iteration.

In plain English, Spark really does get asked first, but replies: "No, I don't have that view yet. Try it without the cache instead." WebForms gets asked second, but automatically says "I didn't have that view, but I went and made one for you anyway, here it is.". And from that point on, the WebFormViewEngine always gets priority because it has the view cached and Spark doesn't.

Summary: Spark is getting priority, but due to a quirk in the way Spark treats the useCache argument, it's getting left in the dust when the Web Form engine is active at the same time. Either WebForm is over-eager or Spark is lazy, depending on your perspective.

Simply put, the solution is not to have conflicting views! If you've registered multiple view engines, then you should be treating any view name which can be handled by either/both of them as undefined behaviour.

Hmmm... Nope - all due respect webforms doesn't do anything beyond a cache-check when useCache is true. Same as Spark.

Actually - I think someone might have moved my cheese... Spark might have had a quirk added causing a false cache-miss during the useCache==true pass. If that's true it's more of a bug than different rules applied to that parameter.

Updated:

I was looking at MVC 2 originally - which is why I implied @Aaronaught's conclusions were incorrect. MVC 2 does not return a view on the first pass where useCache==true, which is different in MVC 1.0 which will resolve and populate.

So difference is between the way ASP.NET MVC 1.0 and ASP.NET MVC 2 are implemented. Spark and MVC 2 treat the useCache flag the same, and the order they are registered will give them priority.