Write smaller, more readable code and catch all boundary cases with pattern matching expressions
My original reason for writing this post was that, some time ago, I felt that the Microsoft documentation for pattern matching in C# did not fully communicate all the cool things you could do with pattern matching in everyday code. Since then, I have come across this very succinctly summary of Pattern matching from Microsoft. However, I thought that this post could still serve benefit as an introduction in using pattern matching to make everyday code more concise.
Pattern matching, in short, is a way you can test a value to see if it has certain characteristics. In many languages, you can write the tests as descriptions of the object under test. Consider the below example:
list is [0, 1, >3]We haven't covered this specific syntax yet, but I can tell you this code resolves to either true or false. It could reasonably be inferred that the code checks the first two elements of list are 0 and 1, and perhaps, that the third element is greater than 3. Less noticably, it is also inferred that the Count of the list is 3.
Patterns often let you extract data directly using the syntax of the pattern. Consider the below example.
list is [0, 1, >3, int number, 4]The above code (or expression) checks the first three elements like the prior example, but in addition to ensuring a Count of five elements, we now see an int number snuggled between the >3 and the 4.
The above code still resolves to true or false. The list may or may not match the pattern. However, if it does, our int number will now contain the element in between >3 and 4. This is what is meant above by using patterns to "extract data".
I should note here that the examples and styles presented in this article reflect my personal preference, and that I am not proposing an objective standard on a "better" way to write code.
Firstly, to properly explain the benefit of pattern matching operators, I have to briefly mention Nullable Reference Types.
C# has had nullable value types (the struct Nullable<T>) since C# 2.0, which serve as wrappers for value types which can hold null. Nullable<T>s are denoted by appending ? to a value type, like so:
public int Age { get; set; } // value type
public int? MaybeAge { get; set; } // nullable value type (Nullable<int>)Nullable Reference Types (or NRTs, as they are called) are denoted by the same syntax of ?. But you may be confused as to why they exist, as reference types are innately nullable in C#. NRTs exist to communicate that null is a valid value for a variable. On the contrary, this communicates that null is not a valid value (or state, if you will) for variables of non-NRTs (so normal reference types). There have always been debates as to whether null should be accepted as a valid state for a variable, and null reference errors have been the source of countless bugs and hours of frustration in almost all programming languages since null's invention.
NRT-enabled projects make the validity of a "null state" part of the variable's type. More than that, the developer actually gets some goodies from the compiler for giving it this extra information. With NRTs enabled, by default, the C# compiler will ensure that - barring System.Reflection, Json deserialization, and cataclysmic world catastrophes - non-NRTs will be guaranteed to be null for the duration of the program.
string name = null; // immediate warning, object is null by default, CS8600 Converting null literal or possible null value to non-nullable type.
string? maybeString = null;As such, if the developer tries to "dereference" (use) a variable without having ensured it is not null, the compiler emits a warning.
name = maybeString; // emits CS8600
if (maybeString != null)
name = maybeString; // ✅ OKIt is worth noting here, that all null state checks associated with enabling NRTs are static and do not add runtime checks, which is why things like System.Reflection can invalidate the state.
The beauty of NRT is that nullable state can become part of your API definitions. Instead of peppering you code with repetitive null checks which can degrade performance and readability, you can define a point between your front-end and back-end precisely where null's must be discarded, which means no more null checks and ArgumentNullException.ThrowIfNulls unless you have a strong reason to not trust the static analysis.
Fair warning: if this seems boring to you, the facets of pattern matching operators I will try to emphasize here might seem underwhelming. However, if this is a feature you'd put a ring on and honeymoon with, this next section should be where it "finally gets good".
You may have seen the following syntax before:
var @object = GetSomeReferenceType();
if (@object is not null)
; // do somethingThis syntax obviously feels a lot like !=, so the clever mind may start substituting null for other values on the right. They'd quickly find that this does not work for non-const variables, emitting a CS0150.
record Person(string Name, int Age, Person? BestFriend);
Person kelly = new("Kelly", 40, null);
var person = GetSomePerson();
if (kelly is person) // CS0150: A constant value is expected
; // do somethingI have received a similar error which essentially says the right-side of a pattern matching expression must be const or an l-value. While as of this writing, I have not been able to replicate this error message, I found that introducing the term l-value is informative as to how pattern matching expressions work.
An l-value stands for a "locator value" and generally refers to a variable declaration. That means anything (double-check me on this) that can be used to declare a variable can be used in a pattern matching expression.
if (maybePerson is Person person)
; // do something
// or
if (maybePerson is var person)
; // do somethingThis syntax conveniently moves null-checking and assignment into a single expression.
if (maybePerson != null)
Person person = maybePerson;
// or
Person person;
if (maybePerson != null)
person = maybePerson;
// vs
if (maybePerson is Person person)
; // ...The null-state static analysis ensures that the scope of the variable matches what can be ensured by the logic of the code. For example, person is not valid outside of the if statement, since that is the only place it is guaranteed to not be null.
if (maybePerson is Person person)
; // ✅ use person here
// ❌ person is not valid outside of scopeWhat's really nice, is that that this pattern works with guard clauses too.
if (maybePerson is not Person person)
return;
// can use person hereThis code might look a bit weird at first, since we expect a variable defined in an if statement (like an out parameter) to be valid within the scope of the corresponding block, but in the above example, Person person is actually being defined in the outside scope only. What we're really saying is "if this maybe-null variable maybePerson is not null (it does conform to the non-NRT type), assign it to person and move on".
As David Fowler pointed out, pattern matching (can) move assignment to the right.
You may have been thinking, "I use var a lot... can I use var with pattern matching?". And the answer is: you sure can! Though it may not be as useful as you think, since var corresponds to the nullable version of the inferred type.
_ = person is var alsoPerson;
// translates to
_ = person is Person? alsoPerson;The null-check comes from testing against the non-nullable type, which is only possible (as far as I am aware) when writing out the whole type name and not var. There is, however, a way var could be useful for property patterns, which are covered below.
So far we have been using pattern matching to add some flair to our null checks - which I admit is the bulk of how I use pattern matching syntax these days. However, pattern matching itself is much more powerful and akin to match in F# - which I will not get into as it would be a detour and I do not feel confident enough in F# to expound upon it.
The first way I have started using pattern matching in this section is with multiple inputs, something I like to call "bracket notation".
I like to describe pattern matching with if statements as defining the object you want to receive, then testing and assigning that object in one go according to your description, as opposed to testing a number of conditions on a variable and choosing to assign or discard it. Note the difference in the following case:
Person person = new("Ken", 25, null);
if (person.Name == "Ken" && person.Age >= 18)
; // do something
// vs
if (person is { Name: "Ken", Age: >=18 })
; // do somethingYou can note that both of these examples both do essentially the same thing, but the pattern matching example is literally shorter and arguably more concise.
This bracket notation lets you describe what you want your input to conform to, instead of performing a series of Boolean checks. When you have a lot of properties and fields you want to check, this can - again, arguably - result in cleaner code.
Perhaps you are at the edge of your seat on this syntax. However, it is worth revisiting the non-const restriction above again, as the bulk of if checks performed are against non-const variables. Note the code below does not compile:
int MinimumAge = 18; // non-const!
if (person is {Name: "Ken"} and {Age: >=MinimumAge}) // error CS0150: A constant value is expected
; // do somethingPattern matching expressions must conform to one of the patterns in this C# patterns reference.
A variable is not a pattern, so the correct version of the code above would be the following:
int MinimumAge = 18;
if (person is {Name: "Ken"} and {Age: int age} && age >= MinimumAge)
; // do somethingYou may be thinking that this defeats the purpose of using a pattern matching expression in this instance, and you may be right.
On the other hand, you may find that making checks against several properties on an object with several layers (or nodes in a tree) of hierarchy may be convenient - particularly by using nested property patterns.
if (person is { BestFriend: { Name: "Bob" } })
; // do something if person has a BestFriend named BobTake the below example, where we check if a Node on a tree has a certain structure:
class Node
{
public Node? Left { get; set; }
public Node? Right { get; set; }
public string Value { get; set; }
}
// ...
Node node = CreateSomeTree(); // could also be a Node of an existing tree
if (node.Left?.Left?.Right?.Value == "A"
&& node.Left?.Left?.Value == "B"
&& node.Right?.Value == "C"
&& node.Left?.Value.Length > 0
// you get the picture ...
)
; // do somethingIn the above example, we reach down into a node to verify it has a certain structure, but the only way we can check that structure is with a series of checks, which are especially pesky and require us to add that ? null-safe access operator to avoid NullReferenceExceptions. If we were to rewrite the above expression using pattern matching:
if (node is { Left: { Left: { Right: { Value: "A" }, Value: "B" }, Value: { Length: >0 } }, Right: "C" }
// ...
)
; // do somethingBecause I tried to construct an example that shows the stark difference between the two syntaxes, I will not try to hide the fact that the latter expression is still quite hard on the eyes and may benefit from some line-breaks and tabs. However, by moving our series of assertions connected by && into a single object description with sub-decriptions, we immediately cut down on visual redundancy from reaching into the object several times. If we still want to make Boolean checks on sub-properties use non-const variables, you can still declare any part of the expression as a variable and then add the manual check after the pattern matching expression.
string nameWeCareAbout = GetNameWeCareAbout(); // non-const
if (node is { Left: { Left: { Right: { Value: "A" }, Value: "B" }, Value: { Length: >0 } } subNode, Right: "C" }
&& subNode.Value == nameWeCareAbout)
; // do something if node's Left node's Value is equal to nameWeCareAboutPattern matching includes the helpful or operator of almost self-explanatory use, however, you must remember that pattern matching operators are not the same as Boolean operators. Specifically, a pattern matching expression describes a single object, so or can only be used to describe alternate descriptions of the same object. Take the below example:
if (person is {Name: "Ken"} or {Age: >=18})
; // do somethingThis makes certain types of operations much more concise. For example, since ors can be stacked (as well as and and not), checking whether a string is a certain value is as simple as name is "John" or "Kerry" or "Nami". However, it must be remembered that pattern matching operators are not 1-to-1 replacements for Boolean operators.
Referencing our earlier example, it should be noted that variable extraction is only possible with or or not insofar as it makes sense.
if (person is {Name: "Ken"} or {BestFriend: Person bestFriend}) // error CS8780: A variable may not be declared within a 'not' or 'or' pattern.
; // do something with bestFriendThe above code will not compile because the bestFriend variable cannot be guaranteed to exist inside or outside the if statement, given that if person contains Name: "Ken", the first part of the pattern (before the or) will satisfy the whole pattern and whether bestFriend is not null will never be checked, meaning the variable cannot be defined.
Property patterns are fun to play around with, as they can be used with the null-conditional access operators and can be stacked.
if (root.Child?.Child.Child is { PropertyA: not null, PropertyB: >20})
; // do somethingAs the C# patterns reference is comprehensive, I specifically would like to defer to this section to demonstrate slice and subpatterns with lists.
However, for want of an introductory example, I'd like to include how I used list patterns to parse simple string text here.
Program.cs
string input = @"A 0 3
B 2 6
C 4 4
// ...
";
StringReader sr = new(input);
// ...
while (sr.ReadLine() is string line)
{
string[] tokens = line.Split(' ');
if (tokens is [string name, string arrivalTimeToken, string durationToken]
&& int.TryParse(tokens[1], out int arrivalTime)
&& int.TryParse(tokens[2], out int duration))
{
Jobs.Add(new(tokens[0], arrivalTime, duration));
}
}In the above example, you can see how I
- checked the length of the
string[]is 3 - assigned each element to a variable using a single pattern matching expression.
It may be further clearer to the reader by now that this blog post is not a holistic overview of pattern matching in C# - rather, a fairly involved introduction. I would encourage the reader to further consult the references linked above for exploring the cool and useful ways you can construct pattern matching expressions. Up until now, we have been using patterns inside if statements to perform checks and extract data, and while I find this is my most common use case for patterns, it's arguably the least powerful.
Consider the below code that I just had Bing Chat generate for me:
// Define an enum for weather conditions
public enum Weather
{
Sunny,
Cloudy,
Rainy,
Snowy
}
// Define an enum for temperature ranges
public enum Temperature
{
Hot,
Warm,
Mild,
Cool,
Cold
}
// Define a tuple for outerwear items
public struct Outerwear
{
public string Hat;
public string Jacket;
public string Gloves;
}
// Write a switch statement that takes a weather and temperature value and assigns an outerwear value
public static Outerwear ChooseOuterwear(Weather weather, Temperature temperature)
{
// Declare an outerwear variable to store the result
Outerwear outerwear;
// Use a switch statement to evaluate the weather and temperature combination
switch (weather, temperature)
{
// For each possible case, assign an appropriate outerwear value
case (Weather.Sunny, Temperature.Hot):
outerwear = new Outerwear { Hat = "Sun hat", Jacket = "None", Gloves = "None" };
break;
case (Weather.Rainy, Temperature.Cool):
outerwear = new Outerwear { Hat = "Rain hat", Jacket = "Raincoat and sweater", Gloves = "Waterproof gloves" };
break;
case (Weather.Snowy, Temperature.Cold):
outerwear = new Outerwear { Hat = "Beanie and scarf", Jacket = "Winter coat and sweater", Gloves = "Woolen gloves" };
break;
case (Weather.Cloudy, _): // Use a discard pattern to match any temperature with cloudy weather
outerwear = new Outerwear { Hat = "None", Jacket = "Warm jacket", Gloves = "None" };
break;
case (_, Temperature.Mild): // Use a discard pattern to match any weather with mild temperature
outerwear = new Outerwear { Hat = "None", Jacket = "Light jacket", Gloves = "None" };
break;
// For any other case, throw an exception
default:
throw new ArgumentException("Invalid weather or temperature value");
}
// Return the outerwear value
return outerwear;
}You may find online many cases of people arguing about switch statements and if else / if else if statements. Some say that if statements easily introduce edge case bugs where certain uncovered cases fall through the cracks or into the wrong "bucket". switch statements - on the other hand - are not as versatile as Boolean conditionals in if statements, but their simplicity helps force the programmer to keep control flow simple. Additionally, a handy default case can be added to catch any uncovered cases.
Some complain that switch statements are unnecessarily restrictive and verbose and that edge case bugs can easily be caught using if statement guard clauses to filter out bad states one by one, instead of handling all possible states in a switch statement or an if else / else if set of blocks.
switch expressions are a possible solution to this dilemma in two parts:
- They are concise and easy to read.
- Uncovered cases produce compiler warnings
I asked Bing Chat to modify the above code to use switch expressions instead, and you may note the ChooseOuterwear method appears about half as tall.
public static Outerwear ChooseOuterwear(Weather weather, Temperature temperature)
{
// Use a switch expression to evaluate the weather and temperature combination
return (weather, temperature) switch
{
// For each possible case, assign an appropriate outerwear value
(Weather.Sunny, Temperature.Hot) => new Outerwear { Hat = "Sun hat", Jacket = "None", Gloves = "None" },
(Weather.Rainy, Temperature.Cool) => new Outerwear { Hat = "Rain hat", Jacket = "Raincoat and sweater", Gloves = "Waterproof gloves" },
(Weather.Snowy, Temperature.Cold) => new Outerwear { Hat = "Beanie and scarf", Jacket = "Winter coat and sweater", Gloves = "Woolen gloves" },
(Weather.Cloudy, _) => new Outerwear { Hat = "None", Jacket = "Warm jacket", Gloves = "None" }, // Use a discard pattern to match any temperature with cloudy weather
(_, Temperature.Mild) => new Outerwear { Hat = "None", Jacket = "Light jacket", Gloves = "None" }, // Use a discard pattern to match any weather with mild temperature
// For any other case, throw an exception
_ => throw new ArgumentException("Invalid weather or temperature value")
};
}Instead of switching on a value to execute a case, a switch expression works by resolving to a value associated with the first matching expression.
For example, (Weather.Cloudy, _) could resolve to
(Weather.Cloudy, Temperature.Hot)(Weather.Cloudy, Temperature.Cold)- any other
Temperature
If the _ expression were not included at the bottom, the compiler would automatically identify that not all cases are covered ((Weather.Sunny, Temperature.Cold), for example) and produce a CS8509 warning.
public static Outerwear ChooseOuterwear(Weather weather, Temperature temperature)
{
// Use a switch expression to evaluate the weather and temperature combination
return (weather, temperature) switch // The switch expression does not handle all possible values of its input type (it is not exhaustive). For example, the pattern '(Weather.Sunny, Temperature.Warm)' is not covered.
{
// For each possible case, assign an appropriate outerwear value
(Weather.Sunny, Temperature.Hot) => new Outerwear { Hat = "Sun hat", Jacket = "None", Gloves = "None" },
(Weather.Rainy, Temperature.Cool) => new Outerwear { Hat = "Rain hat", Jacket = "Raincoat and sweater", Gloves = "Waterproof gloves" },
// ... but not _ and not all cases
};
}This behavior is similar to default in switch statements, except instead of just falling out of the cases block when no cases match and depending on the programmer to write some "catch-all" code, the switch expression emits a warning when you fail to cover all edge cases, and throws a runtime error when hitting the unsupported case if you ignore this warning.
One consequence of this is that those cases where your switch statements do nothing but set some other variable have become a lot simpler. A switch expression resolves to a value, where a switch statement conditionally executes some code. In the same way that ChooseOuterwear directly returns a value from the switch expression, you can also set a variable directly from the result of a switch expression.
enum Access { Restricted, Moderated, Unrestrained }
// ...
var access = (age, isModerator) switch
{
(<18, false) => Access.Restricted,
(>=18, false) => Access.Moderated,
(_, true) => Access.Unrestrained
};You may be wondering how to introduce custom Boolean checks as we did above with if statements. This is accomplished with the when keyword in switch expressions. Instead of trying to get Bing to shoehorn a when into our mutilated and transmogrified example, I encourage the reader to refer to this example in the C# reference which compares tuple inputs to each other.
public record Point(int X, int Y);
static Point Transform(Point point) => point switch
{
var (x, y) when x < y => new Point(-x, y),
var (x, y) when x > y => new Point(x, -y),
var (x, y) => new Point(x, y),
};Furthermore, the when keyword can even be used in a try/catch block to conditionally catch exceptions!
when in a catch statement - modified from Microsoft
try
{
return await new HttpClient.GetStringAsync("https://localHost:10000");
}
catch (HttpRequestException e) when (e.Message.Contains("301"))
{
return "Site Moved";
}
// ...
catch (HttpRequestException e)
{
return e.Message;
}What's fun about switch expressions, is that they allow you to let your data define parts of decision-making in your code. I'm not sure if I would necessarily call them "declarative" or "functional" (I'll let the F# devs argue about that), but the static analysis certainly cuts down on some of the uncertainty when using if statements.
The API that powers this very blog uses switch expressions! They work well with minimal APIs, which already support lambdas and expressions.
group.MapGet("/", async (string key, IPostsTableAccess tableAccess, IPostClient postClient)
=> await tableAccess.GetRow(key) switch
{
Row row => await postClient.GetPost(row.Key) switch
{
{ IsPublished: bool p } post when isEditor || p => Results.Ok(post),
_ => Results.Problem(),
},
_ => Results.NotFound()
});You can use switch expressions with razor code too!
<div>
@((RenderFragment)(State switch
{
FetchState.Success when Post is not null => @<BlogPost Post="Post" />,
FetchState.Loading => @<div>Loading</div>,
_ when Error is not null => @<div class="alert alert-danger">@Error</div>,
_ => @<div class="alert alert-danger">Please try again.</div>
}))
</div>In this post, we didn't (necessarily) cover positional patterns, recursive patterns, or using pattern matching for type conversion such as with interfaces and inheritance. I would encourage the interested reader to check out these examples provided by Microsoft for building type-driven and data-driven algorithms.