Building and Maintaining Frameworks: Video Podcast Transcript
Building and Maintaining Frameworks (video transcription)
Welcome to OnMicrosoft—conversations with thought leaders in Microsoft technologies. In this session, .NET Team Program Manager Krzysztof Cwalina discusses his work building frameworks and offers valuable insights for developers building framework libraries.
Ted Neward: Krzysztof, why don't you give us a little bit of background about who you are, what you do, and what makes you famous?
Krzysztof Cwalina: Well, I'm not sure I'm famous.
Ted: To a select portion of the population, you're moderately famous.
Krzysztof: Yes, yes. I'm a program manager on the .NET Framework team. We have a subteam inside the .NET Framework called Application Frameworks Core, and the team is responsible for two separate charters. One is developing features for the .NET Framework—crosscutting, general-purpose APIs for application models. A second part of the charter is to ensure design quality of the .NET Framework. So we work on things like API design guidelines [and] architectural guidelines. We do reviews of APIs in the .NET Framework, reviews of architecture; also work on layering; componentization; and crosscutting, long-term design activities.
Ted: And you were part of the group that built the FCL [Framework Class Library] over time, correct? So you've seen [it] through this a couple of iterations?
Krzysztof: Yes, I've been on the .NET Framework basically since the group started. So when I joined Microsoft, I started on the VB team and then three months after I joined Microsoft, the team was reorganized and we created the .NET Framework team. I started to work on basic APIs, roughly associated with the BCL [Base Class Library]. Later, in .NET Framework 2.0, I worked on collections, basically on the BCL team. And now, as I said, I work on the application frameworks.
Ted: So you've had a lot of time, in terms of building frameworks that people use that have been around for a while, right? One question that frequently comes up (I'm sure you get it, but I've seen it as well) is, "I want to build something that will be useful beyond just the next six months." What sort of things have you learned, what sort of general tips can you give us about trying to build a framework, or a component, or some reusable entity? What are some things that you've learned the hard way?
Krzysztof: I've spent a lot of time building frameworks and also getting involved with other groups who develop frameworks, helping them get their APIs to be well integrated with the rest of the framework—usable. I work a lot with usability engineers, and I do have a lot of experience in terms of knowing how framework APIs start their lifecycle and what happens over time. At the Tech Ed in Orlando, I gave a talk about evolution of framework libraries, and the talk was specifically about what happens with a framework over time. How you avoid a situation we're all familiar with: When you look at new code, it's usually better than average, correct? When you look at old code, you think, like, "Who the heck created it?"
Ted: Question is, the old code was generally not written by me.
Ted: New code is written by me. Old code was written by "that guy" that we fired, or the contractor, right? We always blame the contractor.
Krzysztof: Exactly. The natural tendency is to say that the person who created it in the past just did not do a very good job. I looked at what happens with frameworks designed by people who are just great framework designers—and they also, after 10 years, start to show flaws, right? You look at the flaws, and you say, "Why did they happen?" I think the main reason is that the environment around the framework changes. So, initially, you design your library, and you say, "These are the scenarios, this is what my customers want today," and so on, and so on. Over time, the scenarios change, the customers' expectations change, and so on, and so on. Unfortunately, framework APIs are not something that can be easily changed over time. So the ecosystem around [it] changes, but the APIs don't, and the reason is backwards compatibility, right? The moment you release the framework outside your immediate organization, you're pretty much bound with a lot of backwards-compatibility restrictions.
Ted: Yes. Don Box used to call this the "two-cubicle rule." If your code—in whatever fashion—got more than two cubicles away from where you were sitting, you couldn't just stand up and say, "Hey, I changed this interface, all of you need to get it from source control and recompile." So, once it's outside of that radius, people can't hear you when you say that stuff was refactored and changed and so forth.
Krzysztof: Definitely. There has been a lot of research done about managing code change, correct? Like this whole object-oriented design community research. A lot of the research focuses on how we separate parts of the system, so if one changes, the other one is not affected.
Ted: Right. Decoupling.
Krzysztof: Exactly. For internal implementation, what you need to do is just enough decoupling so you don't have to change the other system every time something changes in the other subsystem, correct? But the goal is to minimize change. It's not to eliminate change. Now, for APIs, you cannot actually do any change. If one part of the subsystem changes, you cannot change APIs in other parts, even if they are affected, correct? What happens is it limits the changes in subsystems that have dependencies between each other. So, on a high level, or to summarize, I would say that, in general, software is very difficult to modify after it's been shipped. This is past the second cubicle. But for API design and for framework APIs, the problem is even bigger.
Ted: Right. You know, one of the things that the Agile movement, the Agile community really rests on in terms of bedrock is this notion of being able to refactor. If I release something and I discover that it's not quite what we want, that the API was flawed, that our needs have changed, I can go back and refactor this code so that I don't end up with these large collections of deprecated classes and deprecated methods that never go away. Is it possible—because it almost seems like you're saying it's not—is it possible to refactor APIs? Or are we basically stuck with the BCL as it is until the end of time?
Krzysztof: I think there are some changes that you, of course, can make in APIs. For example, you can add new members to existing types. You can imagine a situation where that would be a breaking change, but it's extremely corner case, and for all practical purposes we say that additions to existing types are not breaking, correct? And there is some number of other additional changes that are also not breaking. For example, changing a member from non-virtual to virtual is probably not breaking.
Krzysztof: Unless you take dependency through...
Ted: You can see the scenario where it might create a problem, but for the most part you assume that that's not a breaking change.
Krzysztof: Exactly. So there are some means of evolving frameworks in place. One is making those changes that are not breaking. There are some others, some features in the CLR that we tried to add, to allow us to modify the framework. For example, .NET Framework 2.0 or CLR 2.0 has a feature called type forwarders, and the feature basically allows you to move types from one assembly to another without causing breaking changes for clients that compiled against the older version of the framework.
Ted: When you say "move types"—like the CLR, physically, even though it was defined in assembly A, it appears to be part of assembly B?
Krzysztof: Well, actually, you take type from assembly A and you actually physically move it to assembly B.
Krzysztof: In assembly A, you leave a marker that says, "Hey, runtime, when you try to load this type, you're not gonna find it here—it's in the other assembly."
Ted: Oh, I see. Sort of a source code move. You physically wanted that type to come out of assembly B instead of assembly A.
Krzysztof: Yeah, yeah. So this is just an example of a feature. We are thinking about more features that allow you more malleability in the framework without causing breaking changes. You can see extension methods to be such a feature. Of course, this is not only for binary compatibility, but for source compatibility. You can move a member from one type to another without causing source breaks, correct? Let's say you had a method, you know, "Do something on type Foo." If the method is non-virtual, you can create an extension method on a different type, extension method for the type Foo. And then source, at least C# source and VB source that support extension methods, wouldn't be broken. So, as I said, there are some changes in the framework that are just naturally allowed. There are some that we added features to the runtime—and will be adding more, probably, in the future—to allow us to modify the framework to basically ensure that, as you said, some of the flaws that we do have in the Base Class Library, over time, get fixed.
For more information, visit onpodcastweekly.com and subscribe to all our podcasts. Brought to you by the publishing imprints and information portal of Pearson Education.