Responding to Richard Grimes article on .NET

Edit #1 - Based on feedback, fixed grammatical errors...

I recently ran across an article written by Richard Grimes in Dr. Dobbs Journal titled Mr. Grimes farewell. I wanted to respond to some of Richard's assertions and points in the article. You should take Richard's article with a grain of salt as he does clearly state that the article is "his opinion". That being said, Richard's article is supposed to be his view of the current state of .NET, but rather than discuss how far we've come and where we're going, he instead argues three points - the .NET Framework is too large blocking adoption, issues with the design of the .NET Framework, and over half the article bashing Visual Basic, and finally that Microsoft is "losing confidence" in the .NET Framework. His quotes, italicized in gray, and my responses are below.

On the size of the .NET Framework blocking adoption

  • RG:The framework redistributable is 25 MB, which is many times larger than the Java redistributable. One of the lessons of the early versions of Visual Basic was that the shareware and freeware market created the popularity of the language. While there are some shareware apps written in .NET, I often hear people complain about the huge redistributable.

My Response: Maybe I'm being too picky, but the size is 23,698K or 23.7MB. While Java's runtime is smaller, it still weighs in at 15MB. Throughout the article, Richard refers to .NET applications when he's really referring to client or public client (meaning not in the firewall) For example, installing the .NET Framework on a server or within an intranet where you can control the environment certainly isn't an issue. Even for public client machines, there's plenty of commercial shareware from games to RSS readers that require the .NET Framework. I've also talked to plenty of shareware developers and they certainly aren't using Java. Many use C/C++, Visual Basic, or Delphi. As far as adoption "and the state of .NET",  which the article is supposed to be about is best summarized by Soma, the Developer Division Vice President in his post on .NET Momentum.

Soma: We have seen over 70 million downloads of the .NET framework from Windows Update and the Microsoft Download center to date. For a simple guy like me, that translates to about 5.5 million downloads a month. Another interesting datapoint is that in 2004, we expect to have about 54 million new PCs shipping with the .NET framework installed/preloaded. We also have over 2.5M developers targeting managed code.  

On the Design of the .NET Framework

  • RG: Almost the first thing I posted on the technical preview newsgroups was a simple console application in Cool, and its equivalent in Java with the rhetoric question to spot the difference.
  • RG: There are classes that are mere wrappers around Win32, but there are other classes that appear to be ported from other frameworks. Before it released .NET, Microsoft had its own Java framework library called WFC, and it also had a managed library as part of the Visual Basic (classic) runtime. It would be nice to know how many classes from WFC and VB were ported to .NET.

My Response: These two points contradict each other. In the first he implies that the .NET Framework is a replica of Java, but in the next statement he claims that the .NET Framework is simply ported Win32 classes, Windows Foundation Classes (WFC), and VB runtime classes. Which one is it? If his point is that you can write a simple application and that it will look the same in C# and Java or C++, I don't think that really proves much. Below is an example of a for loop. 

for ( int i = 0; i < x; i++) {... }

Guess what language it's written in? If you said either C, C++, C#, and Java, then you are correct. I don't see what he's trying to prove here. If he tries to build a more robust application then "Hello World" you're going to run into framework or library specific features (ATL is not, MFC is not EJB, etc).

On Interface-based Programming and Remoting

  • RG: Interfaces are elegant but .NET's preference for class-based solutions has marked the death of interfaces. Look at .NET remoting: This has been provided to allow an object to run in the context where it was created, and be accessed from another context. This means that the object state is kept local, and it is the behavior that is remoted. Thus, remoting is an interface-based facility. You can use .NET remoting with interfaces, but reading through the documentation and all of the “how-tos” on the Web, you wouldn't realize this.

Point #1 - Interfaces are dead

  • My Response: Interfaces are used everywhere in the .NET Framework and are especially valuable given single inheritance in languages like VB and C#. Even the simple string class has IComparable, ICloneable, IConvertible, and IEnumerable interfaces. Going forward, one of the key new features for the .NET Framework 2.0, generics, uses interfaces for constraining data types.

Point #2 - Lack of documentation on using interfaces with .NET Remoting

  •  My Response: I am by no means saying that our documentation is flawless, but here's a link to the .NET Framework SDK Samples on Remoting. Notice the fifth sample down is using Interfaces with remoting. I also outline our guidance on using interfaces for remoting below. 
  • RG: .NET can use interfaces but the preferred way is to use classes.
  • RG: Instead, Microsoft prefers people to use a class-based approach, which often results in the bizarre situation of people deploying their server assembly to the client so that the client has the metadata of the service objects available, or a soapsuds assembly, which basically was a hack to get around the problems of having class-based remoting.

My Response: We don't "prefer" any mechanism, per se. While we may offer guidance, developers can choose to develop their applications as they see fit. Our Patterns and Practices group does provide guidance and best practices on these and other points and even includes guidelines on how to design a remote interface. I don't see a class-based favoritism over interfaces, in fact, there has been an increasing movement to use messages and service orientation rather than object orientation. I concede his point on deploying server assemblies to the client, but poor design is poor design. People deploying a server assembly to a client just so that the metadata of the service objects is available could have easily used an interface or schema instead. That being said, there are some situations where you do in fact want to have server code in each client, the example being a peer-to-peer chat where each client acts as both a client and a server.

 

On Microsoft using the .NET Framework for their applications

  • RG: Microsoft treats .NET as a useful library to extend its products, and to date, it has not shown any more conviction to the framework. There have been a few .NET products written entirely in .NET; one such product is Microsoft CRM....They do not want the expense of rewriting their existing code for .NET, and there is no compulsion to provide all new code in .NET; instead, .NET will be hosted as and when it's needed, particularly to allow extensibility through user-supplied code.  

My Response: We should dissect exactly what Richard says here. He says that Microsoft is using .NET to extend existing products and that Microsoft doesn't want the expense of rewriting applications from scratch in .NET. This makes perfect sense to me, why would we re-write perfectly good code? .NET code can interoperate with existing code, and you bet we're going to take advantage of the interoperability layer to add new features that exploit the best managed code has to offer.  As I pointed out previously, Microsoft is using .NET in all sorts of software from operating systems, to developer tools, to Office.

  • RG: Microsoft's current operating systems, XP and Windows 2003, do not depend on .NET; and with XP, .NET is an optional component.

My Response: This is a half-truth at best. While Windows XP Professional does not use the .NET Framework, that's because the .NET Framework was released after Windows XP Professional shipped. Let's look at the operating systems that shipped after the .NET Framework was released:

  • Windows XP Media Center edition requires the .NET Framework for MCE-specific applications.
  • Windows XP Professional Tablet PC Edition requires the .NET Framework for handwriting recognition which is a managed application.
  • Windows Server 2003 requires the .NET Framework to use ASP.NET, UDDI Services, or Sharepoint Team Services.
  • Windows Small Business Server 2003 requires the .NET Framework for ASP.NET as do SBS specific applications like Remote Web Workplace, and the Backup Snap-in.

On Longhorn and the death of browser applications

  • RG: My opinion is that Avalon, or more specifically, XAML, will mark the death of ASP. The reason is that Avalon is a client-side technology, but the browser is an important part of the distribution model. XAML is so rich that a browser-contained XAML application will look no different to a process-based Avalon application, and coupled with Web Services or Indigo (as the mechanism to access remote code), an XAML application will make an ASP.NET application look paltry and antiquated. Why would Microsoft want to kill ASP? Well, with installation of ASP.NET Microsoft sells a single copy of Windows 2003 and perhaps a handful of copies of Visual Studio.NET. The clients don't have to be Windows, so there is no extra sale to Microsoft (whether as a product or licence). This is hardly a great revenue generator, and worse, ASP.NET actually makes it easy to write the application so that it can be used by browsers other than IE.

My Response: I respectfully disagree. XAML will allow for rich interfaces, but ASP.NET and HTML are not going away. Our value is that we can take the best of both worlds, and provide an optimized experience to XAML browsers while still maintaining compatibility with old computers. It should also be noted that there is a difference between client applications and server applications. The server market itself is *totally different* then the client/consumer market. While people talk about Microsoft's dominance in client operating systems at around 90%, we are nowhere near that number in the server market. We're competing against products and companies like IBM WebSphere, hundreds of middleware products, Oracle in the database market, etc. If we want to win the server market, we need to have the fastest, most reliable, most secure, most productive and affordable solution for creating Web applications. To say that browser applications are a threat to Microsoft is so....1996. The threat is not the Web. If it was, wouldn't Microsoft have "lost" already given that the Web is already incredibly successful and popular. How much more popular does the Web have to be before this is proven untrue? Richard then goes on to say that Microsoft needs to do this because of client revenue. The client operating system market, as stated above ~90% is pretty saturated. The server market is where the opportunity for revenue growth truly lies (PS the server and tool business grew 18% last quarter! - see slide 9). On the question of revenue, a typical server deal is in the thousands of dollars as you're paying for several parts including:

  • Operating Systems
  • Transaction Engine
  • Middleware
  • Database

Depending on the complexity of the solution, this can range from thousands of dollars to several million dollars. Server products are expensive. If you look at the market for Web content-management solutions, the average price can be around $50,000 for a one-proc enterprise license. My point being that there are plenty of revenue opportunities and competitive threats from the likes of IBM and others in the server market. Do you know how large the market for database software is in terms of revenue? Would you agree that it's billions of dollars? Do you know that Oracle is the #2 software company and their primary revenue is from databases? That's just *one* of the market opportunities for server software. Back to my point - We are totally committed to our server products and to making Windows Server and ASP.NET the best platform for creating Web applications. Period.

On Longhorn

  • RG: I take the decision to make Avalon available to other versions of Windows as a lack of confidence in the sales of Longhorn.
  • RG: However, Microsoft's announcement that Avalon will be available for other versions of Windows indicates to me that they are not so confident on the uptake of Longhorn, and developers will not write applications for Avalon if they are not convinced that there will be the clients that will run it.

My Response: The decision to make Avalon available to other versions of Windows was driven by one thing, customer demand. Any rudimentary Web search turns up results of customers complaining about not being able to have this functionality on down-level operating systems.

On using the .NET Framework for shipping products

  • RG: So, with the announcements they have made last year, Microsoft has indicated that Longhorn will not be the great .NET innovation that we were lead to believe it was from PDC 2003. This indicates to me that Microsoft is losing confidence in .NET.  

My Response: Richard, the key parts of Longhorn you've mentioned in your article, Avalon and Indigo, are written in managed code. How does that indicate we are losing confidence in .NET when we've decided to bet the success of our next operating system on the .NET Framework? You might then make the follow up argument that because the entire operating system isn't managed, that we have "lost confidence". While that's your opinion, I don't think managed code is right for every scenario and Microsoft has never claimed that it is. Microsoft still fully supports C/C++ and we have a very large existing C/C++ code base and C++ customer constiuency. We'll use managed code where it makes sense.

  • RG: The framework has become Visual Basic—it's intended for users to develop applications, but not for Microsoft to create operating systems or the revenue generating products that they base their profits on.

My Response: I'll avoid responding to the VB bashing as someone is already working on that. There are two points here, one is using the .NET Framework for creating operating systems, the other on Microsoft revenue. On creating operating systems using the .NET Framework, I don't think we've *ever* said that you should be creating an operating system from scratch based solely on managed code. The truth is, gasp, the vast majority of our customers are not creating operating systems.  For those customers who are or who need that level of control and performance, we have C/C++ and we absolutely have not abandoned that. On revenue generating profits, I've already listed several revenue generating applications using the .NET Framework. Microsoft is divided into the seven business groups listed below and let's see which ones are using managed code. 

  • Client - Check
  • Information Worker - Check
  • Server & Tools - Check
  • Home and Entertainment - Check
  • MSN - Check
  • MBS - Check
  • Mobile and Embedded Devices - Check

I hope this clears up and FUD, half-truths and any misconceptions on managed code. If something is incorrect here, please let me know!