# slama.dev

## The C# programming language

19. 2. 2021; lecture notes

### Preface

This website contains my lecture notes from a lecture by Pavel Ježek from the academic year 2020/2021. If you find something incorrect/unclear, or would like to contribute either text or an image, feel free to submit a pull request (or let me know via email).

### Strings

• internally an array of chars
• immutable (neither length nor contents)
• concatenation creates new strings
• String.concat(s1, s2, s3, ...); is good for concatenating a lot of them
• s = "a" + "bcd" + "ef"; internally uses this method, so it isn’t slow
• System.String is the same as string
• can’t ever be a name of a variable!
• removes the confusion – what if someone decides to implement a String class
• == compares contents, char by char (same as .Equals())
• we can use object.ReferenceEquals(o1, o2) if we want reference equality
Action Code
split s.Split(char);
split on multiple delimiters s.Split(delimiters);, delimiters is char[]
convert to integer int.Parse(string);, can throw!
length s.Length;

#### Interning

• hashmap for reusing already created strings
• only for constants, not variables (only exception being "")!
• we can do this ourselves using String.Intern(str); – if the table contains it, return it; else add it to the table
• should be limited use: strings can’t be removed from the table

#### System.Text.StringBuilder

• essentially a mutable string
• internally behaves like a dynamic array
• quick operations like concatenation
• .ToString() returns a proper string (since we can’t use StringBuilder anywhere we want a string)
• no copying happens – internally, the new string points to the contents of the StringBuilder (so we don’t have to copy twice)
• if we were to modify the string builder again, it gets copied

#### Interpolation

• Console.Write("cislo = {0} a {1}", i, j");
• same as Console.Write("cislo = " + i.ToString() + " a " + j.ToString()),
• calls String.Format("format string", i, j)
• creates new string
Console.WriteLine("{0}: Hello, {1}!", s1, s2);

// careful, what if s2 == "{0}"
Console.WriteLine("{0}: Hello, " + s2 + "!", s1, s2);
• interpolated strings\$"My name is {name} and I'm {age}."
• is translated into a String.Format call (at compile time)… most of the time
• if assigned to a FormattableString, its instance is created instead
• .GetArgument(i) returns the i-th argument
• .Format() creates a string
• the {} blocks can be additionaly formatted using :<format> and , <format>

### Chars

• System.Char == char (keyword)
• 2-byte UTF-16 character
• some characters must be at least a string, since some UTF-16 characters can be up to 4 bytes
• \xABCD – unicode code in hex, consumes one to four hex characters
• \uABCD – unicode code in hex, consumes exactly four hex characters
• \UABCDEFGH – unicode code in hex, consumes exactly eight hex characters
• note that while it does mean a single character, this has to be a string

### File I/O

• if we’re dealing with binary I/O, use FileStream instead; this is for text
Action Code
reading System.IO.StreamReader f = new System.IO.StreamReader(path);
read line f.ReadLine();
read chars int chars_read = f.Read(buffer, 0, BUFFER_SIZE);
writing System.IO.StreamWriter f = new System.IO.StreamWriter(path);
write line f.WriteLine(line);
close f.Dispose();

### Classes

#### Constructor

• same syntax as C++, but it’s a good idea to add a public before it
• if none is specified, a default one without parameters is created
• note that we have to write it explicitly if we want it besides one with parameters!
• translated to a .ctor method (see disassembly)
• useful to know, since debug messages can contain this
class A {
int x = stuff;

A () {}
}

// is equivalent to (for each constructor!)
// note that stuff can be arbitrary code

class A {
A () {
int x = stuff;
}
}
• everything in C# inherits : object == System.Object (if not inheriting anything)
• when inheriting, the constructor of the predecessor is called like this:
class A : Z {
int x = something;

A () {stuff}
}

// is equivalent to (for each constructor!)
// note that stuff can be arbitrary code

class A : Z {
A () {
int x = something;

// constructor of Z (without parameters)

stuff
}
}
• if no constructor without parameters exist, the code wouldn’t compile; in this case (or if we just want to call a different one), we can do this:
class A : Z {
A () : base(/*parameters for constructor of Z*/) {stuff}
}

// is equivalent to (for each constructor!)
// note that stuff can be arbitrary code

class A : Z {
A () {
int x = something;

// constructor of Z (with the appropriate parameters)

stuff
}
}
• calling one constructor from another constructor:
• A(): this(constructor parameters) {}
• the stuff that would be called before A() isn’t called (so we don’t do it twice)

#### Class constructor

• same as a regular constructor but for static variables
• called before the first time an object of the class is instantiated
• if no object is constructed, it is never called
• if we do Class a;, then it’s also not called!
• if none is specified, a default blank one is instantiated
• this means that it can never have parameters
• internally called .cctor
class A {
static A() { }
}

#### Inheritance

class A { }      // some stuff
class B : A { }  // some stuff + some more stuff

A a = new A(); // is fine
a = new B();   // is also fine
• when inheriting, everything is inherited, except constructors
• it would be a mess – which constructor would get called when?
• each class has exactly one predecessor
• if none is specified, System.Object == object is used automatically
##### Virtual/abstract/new methods
• member hiding (same name, higher priority)
• it sometimes make sense to implement a function differently in a child
• if we’re looking at B like it’s A, the A implementation would be called (if it’s not virtual)
• a warning is issued – did we really want to do this?
• what if we called the method from some other method in A?
• we can use new to let it know that we really wanted to do this
class A {
public virtual void f() { Console.WriteLine("A"); }
public virtual void g() { Console.WriteLine("A"); }
}

class B : A {
public override void f() { Console.WriteLine("B"); }

// new is optinal, suppresses a warning
public new void g() { Console.WriteLine("B"); }
}

(new B()).f();  // prints B
(new B()).g();  // prints B

((A)(new B())).f();  // prints B
((A)(new B())).g();  // prints A
class Animal {
public virtual void Name() { Console.WriteLine("Animal"); }
}

class Mammal: Animal {
public override void Name() { Console.WriteLine("Mammal"); }
}

class Dog: Mammal {
// new is optinal, suppresses a warning
public new virtual void Name() { Console.WriteLine("Dog"); }
}

class Beagle: Dog {
public override void Name() { Console.WriteLine("Beagle"); }
}

Dog pet = new Beagle();
pet.Name();              // prints Beagle

Animal pet = new Beagle();
pet.Name();              // prints Mammal
• new creates a new record, override overrides the current one:
• it doesn’t matter that a method with the same name already existed
A M D B
A.Name A M M M
D.Name     D B
• if abstract is used, an entry in VTM is created but no implementation is provided
• the entire class has to be abstract so it can’t be instantiated, since the method has to be overriden
• virtual methods can be called from the constructor and act correctly, since Type (and its VMT) must have been initialized by then
• virtual is essentially a promise to the user that it’s fine to provide an alternate implementation in the child without breaking the entire class
• base can be used to call a method from the parent non-virtually (even if it were)
##### Superclass to subclass conversion
• take A and B from the code above; B b = (B)a; has to be checked at runtime
• the conversion is explicit, because it’s something to think about
• if it’s wrong InvalidCastException is thrown, so test the code!
###### is
• checks, whether object is of the given Type (or type of any of its children)
• runtime calls a method, not too quick of an operation – has to go through the class tree!
• if object is null, false is always returned, though it seems that null can be anything
###### as
• returns the object if it is of the given Type (or type of any of its children), else null
B b = a as B;  // assigns a to b if it's of the valid type
// this is the reason why null is A returns false

if (b != null) {
// do stuff with B b
}

// is almost equivalent to (since C# 7.0)

if (a is B b) {
// do stuff with B b
}

// b is not initialized here!
##### System.Object
• inherited by all classes
• has the following members:
• protected object MemberwiseClone();
• creates a shallow copy of the object (byte by byte)
• doesn’t make much sense for reference types
• note that it is protected!
• implement public A Clone() {...} to make it public
• there is an IClonable interface, but it was before <T>
• public Type GetType();
• done for all types (new Type("A"),…)
• contains .Name and other stuff for reflection
• also contains a virtual method table (VMT)
• public virtual bool Equals(object o);
• for checking, whether two objects are equal
• public virtual string ToString();
• the string representation of the object
• shouldn’t be something too complicated (like a giant recursive function)
• public virtual int GetHashCode();
• hash code of the object; for hashmap-like data structures
• public static bool ReferenceEquals(object objA, object objB);
• whether the two objects are equal by reference
##### System.ValueType
• is inherited by all types (int, Nullable, bool, user-defined structures…)
• overrides Equals to be byte-equal (since it is a value type)
• if it has reference members, uses reflection to check their values too – string, for example
• better to implement it ourselves in this case
##### sealed
• prevents inheritance and (likely) allows for some code optimizations
• no virtual function calls
• sealed class A {} – is not inheritable
• sealed override void m() – is not overridable

### Variable scope

#### Local variables

• created each time we enter {, deleted each time we leave }
• but… is extremely fast and can be optimized by the compiler, if a variable is repeatedly created and discarded
• the same name can’t be reused within the same scope, but be careful:
if <something> {int b;} else {int b;}  // this is ok
int b;  // this is not (already declared in a nested block)
• also, the compiler must know that the variable is initialized before use
• a hack is to assign something garbage to f (and add a comment)
int e = 1, f;
if (e == 1) {f = 2;}
e = f; // error, f is not initialized in all paths

### Exceptions

• all exceptions must inherit the System.Exception class
• even the ones from other languages are wrapped in a SEHException
• takes a long time (a lot of things have to be collected), though a try block is basically free
• watch out for exceptions that
1. can’t be caught (like StackOverflowException)
2. probably shouldn’t be caught (like OutOfMemoryException)

#### System.Exception

Property/Function Meaning
e.Message string error message
e.StackTrace string trace of method call stack
e.Source string app/object that threw it
e.ToString() string the formatted exception

#### Syntax

try {
// stuff
} catch (Exception e) {
// stuff only executed if the type matches the exception raised
} finally {
// stuff always executed, even if the exception is not handled
// for example, for closing IO/network resources
}

#### Common exception classes

• it’s important to distinguish exceptions due to bugs (NullReferenceException, IndexOutOfRangeException) and due to incorrect program usage (possibly FileNotFoundException and many others)

• Exception
• SystemException
• ArithmeticException
• DivideByZeroException
• OverflowException
• ….
• NullReferenceException
• IndexOutOfRangeException
• InvalidCastException
• ApplicationException
• user-defined exceptions
• IOException
• FileNotFoundException
• DirectoryNotFoundException
• WebException

### using

Type x;
try {
x = new Type();  // could raise an exception!

// some code
} finally {
if (x != null) x.Dispose();
}
• is (sort of, because x has a slightly different scope) equivalent to
using (type x = new Type()) {
// some code
}

// is also equivalent to (since C# 8.0)
// the Dispose is called when the variable goes out of scope

using type x = new Type();
• only works with objects that are disposable (inherit IDisposable $\implies$ have a Dispose method)

### Properties

int Property {
get { /* stuff to do */ }
set { /* stuff to do (with a variable value) */ }
}
• syntactic sugar for defining a T get() and void set(T value) methods
• must be used with =, although it does generate methods get_* and set_* (so don’t name other methods like that)
• is nice when we want to let the programmer know that we’re just setting/getting something (although it may do something more complex)
• it’s generally a good idea to not make it too slow, since the programmer expects it to be instant
• when doing assemblies, properties make it so we don’t need to rebuild code that uses it, since the API stays the same
• interfaces can also contain them (not the implementation, just that it has to have one)

#### Auto-implemented

• if we’re just doing get => value and set = value, we can just do get; and set; and the code will get automatically generated
• what if we want to make an interface out of it in the future?
• what if we want to add constraints to the values in the future?
• we can also set default values: int X { get; set; } = 12;
• we can also explicitly set the accessibility of the functions: public int Length { get; private set; }
• public affects get, since it doesn’t have an explicit modifier, but not set, because it has one

#### Instantiating objects with properties

A a = new A { x = 3; y = 6 };

// is literally the same as

A a = new A();
a.x = 3;
a.y = 6;
• this implies that the setter can’t be private, since it’s just syntactic sugar!

#### Parametric properties (indexers)

• for defining indexing on an object of the given class
• can be overloaded, since it’s just a method (but might not be the smartest thing to do)
• are compiled into get_Item and set_Item
class File {
FileStream s;

public int this [int index] {
get {
s.Seek(index, SeekOrigin.Begin);
return s.ReadByte();
}

set {
s.Seek(index, SeekOrigin.Begin);
return s.WriteByte((byte) value);
}
}

}

### CLI Type System

• compared to C++, we can’t choose whether something is a value or a reference – it is determined by its type

#### Value types

• variable is always the value
• assigning value copies it
• allocated in-place (with exceptions)
• contains:
• structures
• simple types (int, char)
• Nullables (it’s just a struct)
• enumerations
• new is also used, but it doesn’t allocate anything – it just calls the constructor
• is weird, since it’s completely different to references that are allocated
• can implement interfaces, but boxing happens when we assign to I var
• watch out for passing stuctures to function and doing something to them, since they will be boxed and nothing will change…
##### Simple types
• things like byte, short, int, char, bool
• CLR and JIT knows about it, so operators (think +) is not a function call
• sizes are (almost) always defined (not like C++), except:
• nint and nuint – native, platform-dependent (u)int (since C# 9.0)
• bool is only true/false, so it isn’t defined too (implementation detail)
• decimal – exponent is decimal, so numbers like 0.1 are precies
• the downside is that it is much slower
• note that they are not normalized, so decimal b = 1.000M; will print the zeroes
• all of them are perpendicular to one another in the type hierarchy, but there are conversions
• it is an actual conversion, not like for reference types (just checks in that case)
• implicit conversions happen automatically, while explicit are manual
• although a conversion is implicit, it can have data loss (long $\rightarrow$ float)
• they are not free, it can sometimes take quite a while (int $\rightarrow$ float)
• no conversions from/to bool (unlike C++)

• by default, all numberic constants (without a period) are int
• we can do long a = 10;, it’s optimized to not do a conversion
• we sometimes want to call a specific method with constants and it has overrides:
• unsigned int/long: u
• long, unsigned long: l
• unsigned long: ul
• by default, all decimal number constants are double
• we can’t do float f = 2.5;; do 2.5f
• we can (and should) also do
• d for double d = a / 2d;
• m for decimal m = a / 2m;
• when designing APIs, it might be a good idea to do Int32 instead of int, so programmers from other languages now exactly what we mean
##### Nullable types
• int? x is a nullable variable – can contain its value, or null
• internally imlemented as a struct Nullable<T> that has T Value and bool HasValue memobers
• bool?’s null value is “i don’t know” so we can use | and & and get results we expect

#### Reference types

• variable is always the reference
• assigning reference creates a new reference to it
• the actual value is allocated on managed heaps
• can only ever be assigned a reference to the heap
• can contain a null
• actually, since C# 8.0, we can toggle this behavior and forbid it to contain a null
• only gives warnings, since the compiler can miss paths where it’s not null
• this makes it so we can make objects it Nullable using ?
• contains:
• classes
• interfaces
• arrays
• delegates
• takes up more memory, since it has to contain:
• the reference to the heap, where the value is
• heap overhead
1. which type it is (reference to an instance of the class MethodTable)
• is either $4/8B$, depending on the architecture
2. syncblock – related to threads and locking
• always $4B$ - padded to nearest $8B$
##### Arrays
• each new type inherits System.Array and is a new .NET type
• is sort of like generic types – code is generated for each variant
Action Code
create int[] array = new int[size];
create statically int[] array = {1, 2, 3};
fill with stuff Array.Fill(table, -1, 0, 256);, requires System
sort Array.Sort(array);, highly optimized
length .Length
• array of struct automatically calls constructors on them so we don’t have to
• array of class is null by default
• jagged multidimensional arrays:
• int[][] a = new int[2][]
• can have different row lengths
• takes a lot of memory (each row is a pointer) and is not generally nice to it
• rectangular multidimensional arrays:
• single memory object
• has to have same row lengths
• nicer to memory
• int[,] a = new int[2, 3]

#### Pointers

• dangerous, won’t be discussed too much in the first semester
• is never checked – can be pointed anywhere (actually)
• isn’t tracked by the garbage collector, so it can delete the instance without us knowing… the solution is tracking references (see ref section)

#### (un)boxing

• crossing the bounday between a reference type and a value type
• object o = 5; creates a new object on the heap where the reference points to
• is immutable for simple value types, because… how would we modify it?

### Structures

• only makes sense when we want the value semantics
• Vector, Color, Complex structures, for example
• no inheritance, since the variable is always the value – assigning to variables would be broken (and is one of the main reasons for inheritance), which means:
• no abstract, no virtual methods – all user-defined structures are essentially sealed
• as opposed to a class, a structure always has an empty constructor without parameters
• even if we define some other one
• this means that we can’t create one, since it’s always there
• if they’re a part of a class/struct, the constructor is called automatically
• if they’re on a stack, they can’t be accessed until they are initialized
using System;

struct A {
public int num;
public int num2;
}

struct B {
public static A a1;
public A a2;
}

static class Program {
static void Main(string[] args) {
B b = new B();
Console.WriteLine(B.a1.num);  // is ok
Console.WriteLine(b.a2.num);  // is also ok

A a;
// Console.WriteLine(a.num);  // is not ok

a.num = 5;
Console.WriteLine(a.num);  // is ok

// A a2 = a;  // is not ok, the entire struct has not been initialized

a.num2 = 5;

A a2 = a;  // is now ok
}
}
• careful with putting them in List<S> (or some other structure) and making them mutable – if we attempt to do something like myList[i].IncreaseByOne(), it will just modify the returned copy and not do anything
• this doesn’t happen with arrays, since they’re not using an indexer
• it’s generally a good idea to make structures immutable, because users will use it like this

### readonly

• used to set fields not assignable, except in the constructor
• can also be used in autoimplemented properties (int X { get; } = 5;)

### Visibility

Access Modifiers Access is…
public not restricted.
private limited to the containing type.
protected limited to the containing class derived types.
internal limited to the current assembly.
protected internal limited to the current assembly OR same/derived types.
private protected limited to the current assembly AND same/derived types.
• by default:
• visibility of classes/interfaces/structs is internal
• visibility in classes is private
• visibility in interfaces is public
• it doesn’t make much sense to be private, since it’s a public contract
class A {
private int x;

// is ok, x is private in A
public int GetX() {
return x;
}

// is ALSO OK, since the code is inside A (B : A)
public static void SetXOnB(B b) {
b.x = 30;
}
}

### =>

• syntactic sugar for:
• when a non-void method returns something
• when a void method has only one command

### ??, ??=, ?.

• conditional code execution, depending on whether something is null or not
• a ?? b returns a, if it isn’t null, else b
• x ??= y will be assigned if y isn’t null
• x?.m() will call the method if x isn’t null
• x?.f will get the field value, if x isn’t null

### ref

• a tracking reference
• address is passed and automatically de-referenced
• can point to heap/stack/static data
• parameter must be a variable, not an expression (like a number)
• makes sense if we really want to modify the passed variable
void Inc(ref int x) { x += 1; }

void f() {
int val = 3;
Inc(ref val);  // val will be 4
}
• calling a method on an object is just a shortcut for doing Class.function(ref this, <other parameters>)

### out

• an alternative to ref
• useful for returning/setting multiple things in a function
• the CIL code is exactly the same, but the compiler checks, whether each out parameter has been assigned to (since the caller wouldn’t know about the state of the variable)
• we can’t not assign the parameters (we have to at least assign dummy values)
void Read(out int first, out int next) {
first = Console.Read();
next = Console.Read();
}

void f() {
int first, next;
Read(out first, out next);
}
• the variable can be declared as a part of the function call
int a;
if (tryParse(something, out a)) { Console.WriteLine("We failed: " + a); }
Console.WriteLine("We didn't fail: " + a);

// is the same as

if (tryParse(something, out int a)) { Console.WriteLine("We failed: " + a); }
Console.WriteLine("We didn't fail: " + a);

### var

• derived at compile time, depending on what is on the right side
• if we can’t determine the type, the code won’t compile
• var x;
• var y = (1, 2, 3);
• var z = null
• the declaration is a comment – it’s unwise to write var everywhere:
• var name = GetName(); – what does it return?
• var d = new List<int>(); – what if I want to change List to a HashSet later?

### Interfaces

• a contract – we can assign any class that implements an interface to variables with an interface (when we only require the functionality of the given interface)
• we can do the same for a struct, but it involves boxing
• can’t be instantiated, since it has no code
• not entirely an abstract class
• classes/interface can implement multiple interfaces
• classes can inherit only a single class
• interfaces can’t have code
• note that each interface „inherits“ System.Object, since all objects inherit it too
• not literally, it’s #justcompilerthings

### Heaps and GC

• two heaps; behavior of the garbage collector to the two heaps is different
• real limit is around $1.4\ GB$ (for 32-bit systems)
• OutOfMemoryException could happen sooner, since we could have a lot of holes in the given heap and the new object wouldn’t fit in – heap fragmentation
• happens easily when resizing (dynamic array, for example), since we’re creating a new array of twice the size, that has to co-exist with the old one for a bit
• segment – reservation (virtual memory); around $~16\ MB$
• varies greatly on the architecture and GC configuration!
• kvantum – commit (physical memory); around $~8\ kB$
• varies greatly on the architecture and GC configuration!
• GC is generational
• gen 0 – allocated here
• gen 1 – survives a GC
• gen 2 – survives another GC
• GC of a given generation checks the generation and all above
• the next generations are checked only if the previous ones didn’t free too much memory
• ephemeral segment – the current newest segment of the garbage collector
• this is where GC happens
• once it is full, a new one is added and the old one becomes generation 2 segment
• GC class – for interacting with the GC:
• checking the generation of the object
• checking the number of generation collections
• forcing a collection of a given generation
• should be the last resort, since it usually does things well
• GC $\implies$ no memory leaks is not true (or, well, to an extent…):
• a global cache used in some classes doesn’t get collected, since there is a reference to it

#### Large Object Heap (LOH)

• if it’s larger than $85\ 000\ B$
• usually reached when the object contains an array
• no heap compacting (don’t move objects when others die)

#### Small Object Heap (SOH)

• if it’s smaller than $85\ 000\ B$

#### CIL

• common intermediate language
• formerly known Microsoft Intermediate Language (MSIL) or Intermediate Language (IL)
• intermediate language binary instruction set for the CLI specification
• executed by CLI-compatible runtime like CLR
• object-oriented, stack-based, typically JITted into native code

#### GAC

• global assembly cache – stores assemblies to be shared by computer applications

#### JIT

• just in time – when it’s needed
• the default unit is 1 method
• function calls are translated to calls to stubs
• short calls that call JIT and fix references to the method in the code
• compilation is spaced into the running time of the program
• can mess benchmarks up – make sure we’re testing when the method has already been JITted
• also takes care of optimizations, since optimizing CIL code doesn’t make much sense
• means that they can’t be overly aggressive
##### AOT
• ahead-of-time – before it’s needed
• ngen.exe – pre-JITting code
• more intensive optimizations
• JITting can still happen when it’s needed, though
• will be an important feature in .NET 5

#### CLR

• run-time environment for the .NET framework
• responsible for running .NET programs, regardless of the language
• contains things like GC and JIT compiler

#### BCL

• base class library
• types for built-in CLI data types, basic file access, collections, formatting,…

• BCL + CLR

#### Tiered compilation

• fast-running JIT generating bad code / slow-running JIT generating good code
• bad code is JITted for the first run of a method
• after a number of calls, better code is JITted
• can run in a separate thread and replace the bad code later
• makes benchmarking more difficult

#### Self-contained .exe

• executable containing the program and all needed assemblies
• produced using dotnet publish

#### Method inlining

• compiler optimization that moves the body of the method to where it is called
• can’t always be done (recursive functions)
• generally happens for smaller ($<32\ kB$) methods that aren’t too complicated (no try/catch, for example)
• can be forced/disabled using [MethodImpl]

#### Demand loading

• assemblies are used and loaded into memory only when it’s necessary
• it it’s missing, then the program won’t crash, unless it’s explicitly used

#### (De)compiling

• ILSpy – open-source .NET assembly browser and decompiler
• ilasm – generates an executable from a text representation of CIL code

### Data structures

#### Dictionaries

• using System.Collections.Generic;
Action Code
create Dictionary<int, int> d = new Dictionary<int, int>();
contains d.ContainsKey(element);
add d[index] = element;

#### Queues

Action Code
create Queue<string> q = new Queue<string>();
add q.Enqueue(element);
pop q.Dequeue(element);
size q.Count;
peek q.Peek();

### Miscellaneous

#### NuGet

• package manager

• benchmarks

#### prg.exe.config (XML)

• configures .NET (garbage collection settings,…)
• server mode doesn’t do garbage collection as frequently, to be faster
• client mode collects a lot (could be up to 30%)

#### CLS

• common language specification
• what a language must do in order to be .NET-compliant
• specifies things like minimum types:
• sbyte, ushort are not compliant, so making them parameters of a function called from some other DLL might not be the best idea; on the other hand, implementation details are quite fine

#### Arithmetic overflows

checked {
// kód
}
• is not controlled when functions are called from this block (how could it, when they’re probably already translated…)

#### typeof(Class)

• return the Type instance of the class
• useful when we’re comparing types of variables

#### nameof(x) [Microsoft Docs]

• useful when debugging, when showing exceptions to users…
• better to do than "x", since we can rename using any IDE easily
Console.WriteLine(nameof(System.Collections.Generic));  // output: Generic
Console.WriteLine(nameof(List<int>));                   // output: List
Console.WriteLine(nameof(List<int>.Count));             // output: Count
Console.WriteLine(nameof(List<int>.Add));               // output: Add

var numbers = new List<int> { 1, 2, 3 };
Console.WriteLine(nameof(numbers));        // output: numbers
Console.WriteLine(nameof(numbers.Count));  // output: Count
Console.WriteLine(nameof(numbers.Add));    // output: Add

#### NUnit tests

using System;
using NUnit.Framework;

public class Tests
{
[SetUp]
public void Setup() { /* do something for setup (optional) */ }

[Test]
public void NameOfTheTest()
{
Assert.AreEqual("a", "a");
Assert.AreNotEqual("a", "b");
Assert.IsTrue(true);
Assert.IsFalse(false);

Assert.Throws<ArgumentException>(() =>
{
This doesn't throw!
});
}
}