Readonly parameters in C# – a step closer to immutability

I’ve been working with C# for many years now and the new additions to the language are very exciting.  They’re focused on immutability.  Well, actually, the Microsoft documentation seem to suggest they’re more focused on Performance, but they go hand-in-hand;  the less you can modify, the faster things are.

The benefits of immutability are often overlooked especially by those who have only experienced programming in C#.  But coming from a C++ background, I long for the immutability that C++ has.  For example, In C++, you can:

  • Declare a method as const, meaning the method cannot modify any fields of the class (or call any other non-const method)
  • Declare a parameter as const, meaning the parameter can’t be modified in that method, and if you passed that parameter around to other methods, those methods would also have to declare that parameter as const and follow by the same rules
  • Declare a return type as const, meaning that you can’t change what you’re given
  • Declare a field as const, meaning that only the constructor can set the field

(these bullet points are generalisations)

const is great.  There’s even been 80’s pop songs written on the virtues of const:

 The new features in C# 7.2 are losely related to the above.  This post describes  one of them; in parameters

in parameters

Here’s a method that takes an in parameter:

It says that the DoStuffTo method takes a Thing and cannot modify it.  If you try, you’ll get:

This is fantastic!  Immutable parameters;  I know that my method is immutable and I know that I don’t want to mutate any parameter.s  And now, when I accidentally do, I get a compiler error.

The example above clearly violates the intent that you wanted a readonly variable, but what about situations where it isn’t clear that doing something mutates the object?  What about this?

This is fine.  No complaints from the compiler. You might naturally think that Conculate doesn’t mutate the instance…

Let’s look at the implementation:

Wait!  Conculate does mutate state?!  How could it, our parameter is readonly! No compiler errors like when we set the property…  This is because the compiler doesn’t know that the method mutates state.

But all is not what it seems. Let’s run the following and see what the output is:

… here’s the output…

We called Conculate, which we know mutates the state (by setting Id to 42), but it turns out that a copy was made and the ID was set on that copy.

Here’s an excerpt from the Microsoft page on reference semantics with value types:

So, it turns out that readonly parameters aren’t as powerful as C++ const parameters.  They  are, technically, read-only (you can’t mutate them), but they’re not const as in C++ const.

I also don’t like the fact that you can legally call methods that mutate the instance, but don’t know about it.  If const methods existed in C#, the compiler would’ve disallowed the call to Conculate because it couldn’t be declared as const (as it mutates a field).

Perhaps in a future version of C#, we’ll get const methods…

But there’s more…

Even though readonly parameters are a step in the right direction, I was disappointed.  I was dissapointed because when I read about readonly parameters, I wanted to immediately jump into every single method I’ve ever written and splash in keywords on every parameter.  But the majority of things passed around are classes and not structs.  Notice in the earlier example that Thing was a struct.  In time, with the performance benefits of reference semantics with value types, I’m sure struct will be the go-to type instead of class.

Anyway, in parameters can be either struct or class.  Here’s what happens when you change Thing to be a class rather than a struct; remember the earlier example where the compiler stopped you setting a property?  Well, now it lets you do whatever you want:

Oh the disappointment!

Here’s another excerpt from the documentation:

The benefits are minimal‘.  I personally think minimal is being generous.  I would’ve use the word Dangerous.  If Thing is a class rather than a struct, then what is passed is a reference to a reference.   This means that the value that the method is dealing with can be changed to point to something else at any time! (even from outside of that method)

Here’s a (contrived)  example:

Here, we start off a task to do something with a Thing.  We wait 1 second and set the reference to null.  Assuming the user hasn’t pressed Enter yet, a NullReferenceException will happen when they do.  This is because we’re passing a reference to a reference and then setting the reference to null.

I can only imagine all the subtle bugs that this will cause if spread throughout a codebase.

ReSharper to the rescue?

I don’t even know if it’s possible, but I’d have liked the compiler to disallow in parameters that aren’t value types.  I’m sure the ReSharper team are working on a new code-inspection to warn when passing in references.

Please follow and like us:
  • Tommy Long

    Thanks for the write up. I had only scanned the new `in` keyword before this, and assumed we were increasingly working towards immutability. But, surely this keyword won’t be able to be used for it now? Given the reverse-compatible nature of .Net (i.e. cannot change the behaviour of `in` going forward), will any new developments towards immutability be able to utilise `in` to the extent we would need for immutability to become a language feature?

  • I am a bit skeptical here. Java for a long time had the final keyword that can be used with both method parameters and local variables, but in practice it turnes out that it is not so useful and only introduces syntactic noise. IMHO if you want to guarantee that a class is not changed unexpectedly then make it immutable – so it will behave like a string or a DateTime. I think that this is a problem with current version of C# that you cannot easily create immutable classes and I think maybe C# 8 should introduce some data or union types.