View source for Boxing (computer programming)

{{Short description|Programming language concept}}
{{redirect|Object type|another use|Class (computer programming)|the universal type|any type}}
{{More citations needed|date=August 2009}}
In [[computer science]], '''boxing''' (a.k.a. wrapping) is the transformation of placing a primitive type within an [[Object (computer science)|object]] so that the value can be used as a [[reference type|reference]]. '''Unboxing''' is the reverse transformation of extracting the primitive value from its wrapper object. '''Autoboxing''' is the term for automatically applying boxing and/or unboxing transformations as needed.

==Boxing==

Boxing's most prominent use is in [[Java (programming language)|Java]] where there is a distinction between [[reference type|reference]] and [[value type]]s for reasons such as runtime efficiency and syntax and semantic issues. In Java, a {{Java|LinkedList<T>}} must be such that <code>T</code> is an instance of {{Java|Object}}. One might desire to have a {{Java|LinkedList<int>}}, but this is not directly possible. Instead Java defines [[primitive wrapper class]]es corresponding to each [[primitive data type|primitive type]]: {{Java|Integer}} and {{Java|int}}, {{Java|Character}} and {{Java|char}}, {{Java|Float}} and {{Java|float}}, etc. Using [[Generics in Java|generic]] parameterized types introduced in [[Java Platform, Standard Edition|J2SE]] 5.0, one can then define a {{Java|LinkedList<Integer>}} and insert {{Java|int}} values into the list by boxing them as {{Java|Integer}} objects.

On the other hand, [[C Sharp (programming language)|C#]] has no primitive wrapper classes, but allows boxing of any value type, returning a generic {{C sharp|Object}} reference. In [[Objective-C]], any primitive value can be prefixed by a {{ObjC|@}} to make an {{ObjC|NSNumber}} out of it (e.g. {{ObjC|@123}} or {{ObjC|@(123)}}). This allows for adding them in any of the standard collections, such as an {{ObjC|NSArray}}.

[[Haskell]] has little or no notion of [[reference type]], but still uses the term "boxed" for the runtime system's uniform pointer-to-[[tagged union]] representation.<ref>{{cite web |title=7.2. Unboxed types and primitive operations |url=https://downloads.haskell.org/~ghc/6.12.1/docs/html/users_guide/primitives.html |website=downloads.haskell.org |access-date=10 August 2022}}</ref>

The boxed object is always a copy of the value object, and is usually [[immutable object|immutable]]. Unboxing the object also returns a copy of the stored value. Repeated boxing and unboxing of objects can have a severe performance impact, because boxing [[dynamic memory allocation|dynamically allocates]] new objects and unboxing (if the boxed value is no longer used) then makes them eligible for [[garbage collection (computer science)|garbage collection]]. However, modern garbage collectors such as the default Java HotSpot garbage collector can more efficiently collect short-lived objects, so if the boxed objects are short-lived, the performance impact may not be severe.

In some languages, there is a direct equivalence between an unboxed primitive type and a reference to an immutable, boxed object type. In fact, it is possible to substitute all the primitive types in a program with boxed object types. Whereas assignment from one primitive to another will copy its value, assignment from one reference to a boxed object to another will copy the reference value to refer to the same object as the first reference. However, this will not cause any problems, because the objects are immutable, so there is semantically no real difference between two references to the same object or to different objects (unless you look at physical equality). For all operations other than assignment, such as arithmetic, comparison, and logical operators, one can unbox the boxed type, perform the operation, and re-box the result as needed. Thus, it is possible to not store primitive types at all.

==Autoboxing==
Autoboxing is the term for getting a reference type out of a value type just through [[type conversion]] (either implicit or explicit). The compiler automatically supplies the extra source code that creates the object.

For example, in versions of Java prior to J2SE 5.0, the following code did not compile:

<syntaxhighlight lang=Java>
Integer i = new Integer(9);
Integer i = 9; // error in versions prior to 5.0!
</syntaxhighlight>

Compilers prior to 5.0 would not accept the last line. {{Java|Integer}} are reference objects, on the surface no different from {{Java|List}}, {{Java|Object}}, and so forth. To convert from an {{Java|int}} to an {{Java|Integer}}, one had to "manually" instantiate the Integer object. As of J2SE 5.0, the compiler will accept the last line, and automatically transform it so that an Integer object is created to store the value {{Java|9}}.<ref>[https://docs.oracle.com/javase/1.5.0/docs/guide/language/autoboxing.html oracle.com Java language guide entry on autoboxing]</ref> This means that, from J2SE 5.0 on, something like {{Java|1=Integer c = a + b}}, where {{Java|a}} and {{Java|b}} are {{Java|Integer}} themselves, will compile now — a and b are unboxed, the integer values summed up, and the result is autoboxed into a new {{Java|Integer}}, which is finally stored inside variable {{Java|c}}. The equality operators cannot be used this way, because the equality operators are already defined for reference types, for equality of the references; to test for equality of the value in a boxed type, one must still manually unbox them and compare the primitives, or use the {{Java|Objects.equals}} method.

Another example: J2SE 5.0 allows the programmer to treat a collection (such as a {{Java|LinkedList}}) as if it contained {{Java|int}} values instead of {{Java|Integer}} objects. This does not contradict what was said above: the collection still only contains references to dynamic objects, and it cannot list primitive types. It cannot be a {{Java|LinkedList<int>}}, but it must be a {{Java|LinkedList<Integer>}} instead. However, the compiler automatically transforms the code so that the list will "silently" receive objects, while the source code only mentions primitive values. For example, the programmer can now write {{Java|list.add(3)}} and think as if the {{Java|int}} {{Java|3}} were added to the list; but, the compiler will have actually transformed the line into {{Java|list.add(new Integer(3))}}.

===Automatic unboxing===
With automatic unboxing the compiler automatically supplies the extra source code that retrieves the value out of that object, either by invoking some method on that object, or by other means.

For example, in versions of Java prior to J2SE 5.0, the following code did not compile:

<syntaxhighlight lang=Java>
Integer k = new Integer(4);
int l = k.intValue(); // always okay
int m = k;            // would have been an error, but okay now
</syntaxhighlight>

C# doesn't support automatic unboxing in the same meaning as Java, because it doesn't have a separate set of primitive types and object types. All types that have both primitive and object version in Java, are automatically implemented by the C# compiler as either primitive (value) types or object (reference) types.

In both languages, automatic boxing does not downcast automatically, i.e. the following code won't compile:

C#:
<syntaxhighlight lang="csharp">
int i = 42;
object o = i;         // box
int j = o;            // unbox (error)
Console.WriteLine(j); // unreachable line, author might have expected output "42"
</syntaxhighlight>

Java:
<syntaxhighlight lang="java">
int i = 42;
Object o = i;          // box
int j = o;             // unbox (error)
System.out.println(j); // unreachable line, author might have expected output "42"
</syntaxhighlight>

== Boxing in Rust ==
[[Rust (programming language)|Rust]] has the {{Code|Box}} type, which represents a uniquely owned, heap-allocated value:<ref>{{cite web |title=std::boxed - Rust |url=https://doc.rust-lang.org/std/boxed/index.html |website=doc.rust-lang.org |access-date=2 June 2025}}</ref>
<syntaxhighlight lang="rust">
let number: Box<i32> = Box::new(42);
</syntaxhighlight>

<code>std::boxed::Box</code> in Rust is equivalent to <code>std::unique_ptr</code> in [[C++]], a kind of [[smart pointer]].

<syntaxhighlight lang="cpp">
using std::unique_ptr;

unique_ptr<int> number = std::make_unique<int>(42);
</syntaxhighlight>

If the value needs to have shared ownership (e.g. between threads), one can use {{Code|Arc}}, which represents a reference-counted, heap-allocated value.<ref>{{cite web |title=Arc in std::sync - Rust |url=https://doc.rust-lang.org/std/sync/struct.Arc.html |website=doc.rust-lang.org |access-date=18 January 2025}}</ref><ref>{{cite web |title=Arc - Rust By Example |url=https://doc.rust-lang.org/rust-by-example/std/arc.html |website=doc.rust-lang.org |access-date=18 January 2025}}</ref> This is similar to <code>std::shared_ptr</code> in C++.

==Type helpers==
Modern [[Object Pascal]] has yet another way to perform operations on simple types, close to boxing, called type helpers in [[FreePascal]] or record helpers in [[Delphi (programming language)|Delphi]] and [[FreePascal]] in Delphi mode.<br>
The dialects mentioned are Object Pascal compile-to-native languages, and so miss some of the features that C# and Java can implement. Notably run-time [[type inference]] on strongly typed variables.<br>
But the feature is related to boxing.<br>
It allows the programmer to use constructs like
<syntaxhighlight lang="pascal">
{$ifdef fpc}{$mode delphi}{$endif}
uses sysutils;  // this unit contains wraps for the simple types
var
  x:integer=100;
  s:string;
begin
  s:= x.ToString;
  writeln(s);
end.
</syntaxhighlight>

==References==
{{Reflist}}

{{Data types}}

[[Category:Data types]]
[[Category:Java (programming language)]]
[[Category:Programming language concepts]]