On things you don't have – Alexey Raga

Alexey Raga

throw new InvalidIdeaException()

On things you don't have

Any (good) programming language offers a lot of very useful things. These things are usually pretty much well known among developers programming in this language. However, there are many other things that particular language does not provide, and these things are typically remain completely unknown or totally misunderstood by the same developers, well, because they do not exist in their universe. It is like early iPhone users who would say “who needs copy/paste on a phone, nonsense!”. Right until this feature became available to them and they realised how useful it was.

Now, why one would even look at another programming language?
One reason would be to escape from the Blub paradox and to better understand his options.
Another reason is that a programming language is just a tool in your toolbox. No carpenter would ever say “I am comfortable with a Sledge Hammer so I don’t even need to know about other options”. That would probably be a silly thing to say. We, developers, are no different.
It is very pragmatic to know the options and then to be able to choose the most effective tool for the job rather than the most familiar one.

In this article I point to some of the very useful “features” that C# doesn’t offer, but other languages do. For the purpose of this article (and to stay within the .NET boundaries) I will only mention things that exist in F# but not in C#.

Note, that it is not a comprehensive list by any means, just some features I picked and decided to write about this time.

Type Providers

Type providers” is a mind blowing feature that is, as far as I know, unique to F# and does not exist in other languages.

It is better to start with a screenshot:

FSharp SQL Type Provider

What happens here is just fantastic. When SQL Type Provider is pointed to a database (given a connection string), it gets the schema, interrogates it and “automagically” creates the types. In design time, so on the very next line you have the access to all the tables and fields, with an intellisense and an autocomplete. All statically typed.
It is like if you generated these types using some ORM such as EntityFramework, except that you don’t need to do anything. It is even better because the type provider and the compiler make sure that the code is always in sync with the database schema.
And it works for any database, all you need is to provide it with a connection string. ORM in two lines of code, isn’t it great?!

“Not a big deal, I am just comfortable with my EntityFramework”, one can say.
OK, and what about CSV? Or JSON? Or XML? Or Atom? What if this JSON or XML or Atom is a feed somewhere in the Internet? Like stock market statistics, or Netflix, or Freebase, or WorldBank data sources? Or your own JSON API?

Here is an example of how you could programmatically query StackOverflow in F#:

FSharp Atom Type Provider

Similar two lines of code except that now the OData Type Provider is used. As you can guess, not only StackOverflow, but any OData-compatible source (such as atom) can be used with this provider.

Type providers is a mind blowing feature. There are lots of type providers exist of all sorts: Amazon S3, COM, Azure, HTTP, Json, XML, SQL, OData, R, Yaml, MathLab, DynamicsCRM, WorldBank, etc., etc.
There is even WMI Type Provider so you can easily query and manage your operating system from F# accessing pretty much everything, in 2 lines of code.

And it is not that hard to build a type provider for something if there isn’t one already. For example, you can build a type provider for your product and it will serve as a nearly perfect API that you can build on top or use when integrating with other systems.

If you are a bit intrigued by now then have a look at the WorldBank type provider examples and play with it trying to find some interesting correlations.

Record types

Records are simple types that contain named values and, optionally, methods.

type FirstName = FirstName of string
type LastName  = LastName  of string
type Age       = Age       of int

type Person = { name: FirstName; surname: LastName; age: Age }

The first 3 lines are just immutable type declarations, consider them as wrappers. You could (and, as we agreed before, should) do it in C# too. Unfortunately it will take you about 50 lines of code for each type, but it can never be an excuse, can it?
In F# it is a one-liner, you just declare the essence of your type and the compiler does all the heavy lifting.

The last line is declaring a record type. It is also immutable, and it also gives you all these Equals, GetHashCode, IComparable, etc.
On top of that it declares a “copying constructor” so you can easily “update” the record. Of course, in F# you don’t think much in terms of “constructors”, it is taken care of, so you can concentrate on your business logic.

A toy example:

// creating an instance of Person
let mary26 = { name = FirstName "Mary"; surname = LastName "Stuart"; age = Age 26 }

// create a copy with `age` incremented
let mary27 = { mary26 with age = Age 27 }

// evaluates to `true`
let older = mary27 > mary26

You can look at these type declarations as at just syntactic sugar and “not a big deal”, but it actually is a big deal. It changes how you think about your types and your code.
And anyway, isn’t it great to be able to express something fairly complex in just a couple of lines of a very simple code? Hurray, no more ridiculous rules like “one file per type”!

Sum types

C# only has what is called product types. When you declare a class that contains a couple of properties - it is a product type. It is typically written in literature as P = A * B * C. For languages like C# it means that type P is a composition of A and B and C. For example, Point is a composition two ints, and Tuple<A, B> is a composition product of A and B.
In fact, the type Person that is declared above is a product type too, as most of the types that you create daily.

Sum types are different. These are disjunctions. What is written as P = A + B means that type P can be either A or B.

This is something that is very hard to model in C#. We are working around it by inventing all sorts of complexity and ceremony simply because C# doesn’t offer anything better. And because we don’t know any better we think that this complexity is normal as it should be.
F# has sum types baked in the language, they are called discriminated unions (because MS likes naming things differently).

In fact one of the examples of sum types we have already discussed before, it is the Option type that is declared as:

type Option<'a> = Some of 'a | None

This means exactly how it reads: the Option is a generic type that can either be None or Some, and if it is Some then it also carries a value of type 'a.

type DownloadResult =
  | OK of string
  | NotChanged
  | Error of string

This is how you could model some download result: simply declare a type that can represent either a success with the content, or indicate that nothing has changed since you did it last time, or an error.

The following example is interesting because the structure is recursive. It is a tree where each node can either be a leaf or a branch of two trees (left and right). Every developer has implemented this structure at least once, here how it looks like when you have sum types in your language:

type Tree<'a> =
  | Branch of left: Tree<'a> * right: Tree<'a> //'left' and 'right' trees
  | Leaf of 'a

Now imagine (or remember) implementing it in C# and compare the effort required.

In C# we often abuse enums for things they don’t really supposed to be used for. Enums are often used for result codes, colours, statuses, roles, etc.
Sum types are often much better than enums in these cases because:

  • Sum types are real types, not just aliases to int or byte.
  • Sum types are closed. Enums are just numbers, and you can assign 9876 to your enum value of “role”, even if it is “incorrect” and is “outside of the range” of your enum. With sum types it is not possible.
  • Sum types can carry more information. Like Some that carries a value, and Branch that carries a couple of trees. (Look at the DownloadResult example again)
  • Compiler won’t let you shoot yourself in the foot! It will not be possible to use NotChanged where OK is expected, etc.

Consider this example where we want to model a small kingdom:

type Rank = Sovereign | Heir | Peasant

type Command =
    | RegisterBirth of Person * Rank
    | IncreaseRank  of PersonId
    | RegisterDeath of PersonId

let personId = sendCommand RegisterBirth(mary, Heir)

//after 6 days Mary's father dies...
let newRank  = sendCommand IncreaseRank(personId) //and Mary is a queen now!

This example looks simple, it is declarative and easy to understand, but it alone worth pages and pages of C# code in several files. Note also that the code above is trivial to write, read and maintain.

Pattern matching

Pattern matching is often viewed as a “switch on steroids”. Indeed, it can be used like one:

let guess a =
    match a with
    | 1         -> "One"
    | 2 | 3 | 4 -> "Not too many"
    | value     -> sprintf "%d is a lot!" value

Here the function guess matches the value of the argument a against some cases and produces the result, in the same way as the switch statement in C# would be used.

But there is much more to pattern matching. Starting with a feature that is so heavily requested in C# that it will probably be included in a one of its next releases:

let someTuple = ("Mia", 27)
let (name, age) = someTuple

The first line just creates a tuple of two elements.
The second line is interesting: it “decomposes” the tuple into two individual values so you don’t have to write things like name = someTuple.Item1 etc. When implemented in C# it will probably be some special case, a syntactic sugar only for tuples (let’s hope that I’m wrong).
In F# it is yet another example of pattern matching.

There is another one:

let guessList list =
    match list with
    | []                    -> "Empty list"
    | [ x ]                 -> sprintf "Singleton list with %O" x
    | [ x ; y ] when x = y  -> sprintf "%O twice!" y
    | [ _ ; y ]             -> sprintf "Two elements with second %O" y
    | [ x ; _ ;  z]         -> sprintf "Three elements, starts with %O and ends with %O" x z
    | _                     -> "Didn't expect that..."

Here is how we could match on lists. This cannot be done with the regular switch because patterns here are more complex: the length of the list is involved, elements are extracted, conditions are applied (the when clause in the 3rd pattern).
Now you can probably see why it is called pattern matching.

A slightly more complicated example would be:

let buyAlcohol person =
    match person with
    | {Person.name = FirstName("Mary")} -> "Mary is a queen, buy at will"
    | {Person.age = a} when a >= Age 21 -> "Here is your beer"
    | {Person.name = FirstName(nm)}     -> sprintf "%s is not allowed to buy alcohol" nm

This example above is matching against the record type (as well as against our “wrapper” types such as FirstName).
The first pattern matches any person whose first name is “Mary”.
The second one doesn’t care about anything but the person’s age, and it makes the age accessible as a. The when guard guarantees that the pattern only matches if the person is older than 21.
The last pattern matches any person at all. It just extracts person’s first name into the value named nm, so it can be used in the result string.

Pattern matching goes way beyond this, you can match against records, arrays, lists, tuples, types, unions, etc. You can have conditions for patterns to express that “it matches this shape, but only if that condition holds”. You can even define your own patterns.
Look at the list of the supported patterns on MSDN and you will be truly surprised of what it can do for you. Then, to be impressed even more, read the Active Patterns section that explains how you can define your own patterns for your specific cases and then to use them as they were built in the language.

Type inference

You have probably noticed but none of the examples above specifies types in function declarations. Not even parameters’ types. But it doesn’t mean that types are not there! F# is a statically typed language (just like C#), but it has much better type inference system.

In C# type inference pretty much ends with the var keyword and lambda functions. In F# specifying types explicitly is almost never required. The compiler infers types for you!
You still can specify types explicitly if you want, but it is not how people typically write F# code.

Conclusion

In this article I picked some of the features I really miss in C#. It is not a comprehensive list, and I only lister things from F#, simply because I tried to make this article to be more relevant for developers who want or have to stay in .NET ecosystem.

For those I could advice: give F# a try. Do it not for the sake of learning another language, but for very pragmatic and result-oriented reasons: you write much less code, you write much simpler and trivial code, you write much less boilerplate and ceremony. And you still stay with .NET, so you don’t change anything except the words you type in your IDE :)

F# offers, perhaps, everything that C# does, and it offers more. Someone said once that for a long time there was no feature introduced in C# language without a disclaimer of “and F# already can do it”.

Not falling into a trap of “I am comfortable with a Sledge Hammer”, looking around and knowing your options is always beneficial for pragmatic reasons. The Blub paradox is yet another trap to avoid, and knowing things we don’t have could be just as important as knowing things we do have. It is because it gives us choice and lets us to be more efficient at what we do.

Have a nice day. Use languages.

Written on September 1, 2015