The difference between struct and record
This question mixes two axes
struct and record answer two different questions:
- struct vs class - a question about memory semantics (value type vs reference type)
- record vs an ordinary declaration - a question about equality semantics
record is NOT an alternative to struct - it is a modifier you can use together with class (the default) or together with struct (explicitly, via record struct).
Four combinations, not two
public class OrderClass { } // reference type, equality by reference
public record OrderRecord { } // reference type (!), equality by value
public struct PointStruct { } // value type, equality by value (but manual/reflection)
public record struct PointRecordStruct { } // value type, equality by value (generated)
A record without the struct suffix generates a class by default.
public record OrderRecord(int Id); // to jest DE FACTO klasa!
OrderRecord r1 = new OrderRecord(1);
OrderRecord r2 = r1;
// r1 and r2 are STILL two references to THE SAME object on the heap (as with a class)
// BUT r1 == r2 returns true regardless, because it is a VALUE comparison, not a reference one
Comparing a "pure" struct with a "pure" record
struct |
record (= record class) |
|
|---|---|---|
| Memory type | Value (stack/inline) | Reference (heap) |
| Copying on assignment | Copies the WHOLE object | Copies only the reference |
| Default equality | Field comparison via reflection (slow!) in ValueType.Equals |
Generated, efficient code comparing the fields |
Can be null |
No | Yes |
| Inheritance | No | Yes |
with expression |
No | Yes |
A surprising fact: a plain struct also has "equality by value" in a sense:
public struct PointStruct { public int X; public int Y; }
var p1 = new PointStruct { X = 1, Y = 2 };
var p2 = new PointStruct { X = 1, Y = 2 };
p1.Equals(p2); // true - BUT via reflection, SLOW (ValueType.Equals uses reflection)
record struct solves exactly this performance problem.
record struct - where the two concepts meet
public record struct PointRecordStruct(int X, int Y);
var p1 = new PointRecordStruct(1, 2);
var p2 = p1; // KOPIUJE (struct)
p2.X = 100; // CAN be changed - a record struct WITHOUT readonly is not init-only
Console.WriteLine(p1.X); // 1 - niezmienione
var p3 = p1 with { X = 100 }; // "with" works for a record struct too
An important difference from record class: a record struct is NOT init-only by default - the properties have { get; set; }, not { get; init; } (unless you add readonly record struct). This is the opposite of what happens in record class.
public record struct MutablePoint(int X, int Y); // set, CAN be changed after creation
public readonly record struct ImmutablePoint(int X, int Y); // init, niemutowalne, jak record class
A practical rule
Do you need to represent DATA (not behaviour/identity)?
├── Small (≤16 bytes), a value like a primitive (Point, Money, coordinates)?
│ └── record struct (the best of both worlds)
├── Larger data, DTO, Command/Query (CQRS)?
│ └── record (= record class)
└── Need mutable state with identity (an EF entity, a ViewModel, a service)?
└── a plain class