It’s all about context

The C# using directive and implicitly typed local variables (i.e. using var) are Good Things whose use should be encouraged in C# programs, not prohibited or severely limited. Used correctly (and it’s nearly impossible to use them incorrectly), they reduce noise and improve understanding, leading to better, more maintainable code. Limiting or prohibiting their use causes clutter, wastes programmers’ time, and leads to programmer dissatisfaction.

I’m actually surprised that I find it necessary to write the above as though it’s some new revelation, when in fact the vast majority of the C# development community agrees with it. Alas, there is at least one shop–the only one I’ve ever encountered, and I’ve worked with a lot of C# development houses–that severely limits the use of those two essential language features.

Consider this small C# program that generates a randomized list of numbers from 0 through 99, and then does something with those numbers:

    namespace MyProgram
    {
        public class Program
        {
            static public void Main()
            {
                System.Collections.Generic.List<int> myList = new System.Collections.Generic.List<int>();
                System.Random rnd = new System.Random();
                for (int i = 0; i < 100; ++i)
                {
                    myList.Add(rnd.Next(100));
                }
                System.Collections.Generic.List<int> newList = RandomizeList(myList, rnd);

                // now do something with the randomized list
            }
            
            static private System.Collections.Generic.List<int> RandomizeList(
                System.Collections.Generic.List<int> theList,
                System.Random rnd)
            {
                System.Collections.Generic.List<int> newList = new System.Collections.Generic.List<int>(theList);
                for (int i = theList.Count-1; i > 0; --i)
                {
                    int r = rnd.Next(i+1);
                    int temp = newList[r];
                    newList[r] = newList[i];
                    newList[i] = temp;
                }
                return newList;
            }
        }
    }

I know that’s not the most efficient code, but runtime efficiency is not really the point here. Bear with me.

Now imagine if I were telling you a story about an experience I shared with my friends Joe Green and Bob Smith:

Yesterday, I went to Jose’s Mexican Restaurant with my friends Joe Green and Bob Smith. After the hostess seated us, Joe Green ordered a Mexican Martini, Bob Smith ordered a Margarita, and I ordered a Negra Modelo. For dinner, Joe Green had the enchilada special, Bob Smith had Jose’s Taco Platter . . .

Wouldn’t you get annoyed if, throughout the story, I kept referring to my friends by their full names? How about if I referred to the restaurant multiple times as “Jose’s Mexican Restaurant” rather than shortening it to “Jose’s” after the first mention?

The first sentence establishes context: who I was with and where we were. If I then referred to my friends as “Joe” and “Bob,” there would be no ambiguity. If I were to write 50 pages about our experience at the restaurant, nobody would get confused if I never mentioned my friends’ last names after the first sentence. There could be ambiguity if my friends were Joe Smith and Joe Green, but even then I could finesse it so that I didn’t always have to supply their full names.

Establishing context is a requirement for effective communication. But once established, there is no need to re-establish it if nothing changes. If there’s only one Joe, then I don’t have to keep telling you which Joe I’m referring to. Doing so interferes with communication because it introduces noise into the system, reducing the signal-to-noise ratio.

If you’re familiar with C#, most likely the first thing that jumps out at you in the code sample above is the excessive use of full namespace qualification: all those repetitions of System.Collections.Generic.List that clutter the code and make it more difficult to read and understand.

Fortunately, the C# using directive allows us to establish context, thereby reducing the amount of noise in the signal:

    using System;
    using System.Collections.Generic;
    namespace MyProgram
    {
        public class Program
        {
            static public void Main()
            {
                List<int> myList = new List<int>();
                Random rnd = new Random();
                for (int i = 0; i < 100; ++i)
                {
                    myList.Add(rnd.Next(100));
                }
                List<int> newList = RandomizeList(myList, rnd);

                // now do something with the randomized list
            }
            
            static private List<int> RandomizeList(
                List<int> theList,
                Random rnd)
            {
                List<int> newList = new List<int>(theList);
                for (int i = theList.Count-1; i > 0; --i)
                {
                    int r = rnd.Next(i+1);
                    int temp = newList[r];
                    newList[r] = newList[i];
                    newList[i] = temp;
                }
                return newList;
            }
        }
    }

That’s a whole easier to read because you don’t have to parse the full type name to find the part that’s important. Although it might not make much of a difference in this small program, it makes a huge difference when you’re looking at code that uses a lot of objects, all of whose type names begin with MyCompany.MyApplication.MySubsystem.MyArea.

The use of using is ubiquitous in C# code. Every significant Microsoft example uses it. Every bit of open source code I’ve ever seen uses it. Every C# project I’ve ever worked on uses it. I wouldn’t think of not using it. Nearly every C# programmer I’ve ever met, traded email with, or seen post on StackOverflow and other programming forums uses it. Even the very few who don’t generally use it, do bend the rules, usually when LINQ is involved, and for extension methods in general.

I find it curious that extension methods are excepted from this rule. I’ve seen extension methods cause a whole lot more programmer confusion than the using directive ever has. Eliminating most in-house created extension methods would actually improve the code I’ve seen in most C# development shops.

The arguments against employing the using directive mostly boil down to, “but I can’t look at a code snippet and know immediately what the fully qualified name is.” And I agree: somebody who is unfamiliar with the C# base class library or with the libraries being used in the current project will not have the context. But it’s pretty unlikely that a company will hire an inexperienced C# programmer for a mid-level job, and anybody the company hires will be unfamiliar with the project’s class layout. In either case, a new guy might find the full qualification useful for a week or two. After that, he’s going to be annoyed by all the extra typing, and having to wade through noise in order to find what he’s looking for.

As with having friends named Joe Green and Joe Smith, there is potential for ambiguity if you have identically named classes in separate namespaces. For example, you might have an Employee class in your business layer and an Employee class in your persistence layer. But if your code is written rationally, there will be very few source files in which you refer to both. And in those, you can either revert to fully-qualified type names, or you can use namespace aliases:

    using BusinessEmployee = MyCompany.MyProject.BusinessLogic.Employee;
    using PersistenceEmployee = MyCompany.MyProject.Persistence.Employee;

Neither is perfect, but either one is preferable to mandating full type name specification in the entire project.

The code example still has unnecessary redundancy in it that impedes understanding. Consider this declaration:

    List<int> myList = new List<int>();

That’s like saying, “This variable called ‘myList’ is a list of integers, and I’m going to initialize it as an empty list of integers.” You can say the same thing more succinctly:

    var myList = new List<int>();

That eliminates the redundancy without reducing understanding.

There is some minor debate in the community about using var (i.e. implicit typing) in other situations, such as:

    var newList = RandomizeList(myList, rnd);

Here, it’s not plainly obvious what newList is, because we don’t know what RandomizeList returns. The compiler knows, of course, so the code isn’t ambiguous, but we mere humans can’t see it immediately. But if you’re using Visual Studio or any other modern IDE, you can hover your mouse over the RandomizeList, and a tooltip will appear, showing you the return type. And if you’re not using a modern IDE to write your C# code, you have a whole host of other problems that are way more pressing than whether or not a quick glance at the code will reveal a function’s return type.

I’ve heard people say, “I use var whenever the type is obvious, or when it’s necessary.” The “necessary” part is a smokescreen. What they really mean is, “whenever I call a LINQ method.” That is:

    var oddNumbers = myList.Select(i => (i%2) != 0);

The truth is, they could easily have written:

    IEnumerable<int> oddNumbers = . . .

The only time var is absolutely necessary is when you’re working with anonymous types. For example, this code that creates a type that contains the index and the value:

    var oddNumbers = myList.Select((i, val) => new {Index = i, Value = val});

I used to be in the “only with LINQ” camp, by the way. But after a while I realized that most of the time the return type was perfectly obvious from how the result was used, and in the few times it wasn’t, a quick mouse hover revealed it. I’m now firmly in the “use implicit typing wherever it’s allowed” camp, and my coding has improved as a result. With the occasional exception of code snippets taken out of context, I’ve never encountered a case in which using type inference made code harder to understand.

If you find yourself working in one of the hopefully very few shops that restricts the use of these features, you should actively encourage debate and attempt to change the policy.

If you’re one of the hopefully very few people attempting to enforce such a policy (and I say “attempting” because I’ve yet to see such a policy successfully enforced), you should re-examine your reasoning. I think you’ll find that the improvements in code quality and programmer effectiveness that result from the use of these features far outweigh the rare minor inconveniences you encounter.

Comments are closed.

Categories

A sample text widget

Etiam pulvinar consectetur dolor sed malesuada. Ut convallis euismod dolor nec pretium. Nunc ut tristique massa.

Nam sodales mi vitae dolor ullamcorper et vulputate enim accumsan. Morbi orci magna, tincidunt vitae molestie nec, molestie at mi. Nulla nulla lorem, suscipit in posuere in, interdum non magna.