Skip to content

Namespaces and other objects

If you are familiar with Object Oriented Programming (OOP) concepts and have experience using them from another language, then Dyalog's classes and namespaces should hopefully feel quite straightforward to you.

There is a video Namespaces in Dyalog APL with a basic introduction to the concepts and usage of namespaces.

Two documents are also worth looking at, whether or not you are familiar with OOP:

If nested arrays are arrays inside arrays; namespaces are a bit like a workspace within a workspace. They are objects which contain collections of names, and these names can be listed as before, but using the dot . syntax from object-oriented programming.

      )ns ns         ⍝ Create an empty namespace called ns
      ns.var←1 2 3   ⍝ Create a variable in ns called var
      ns.fn←{⍺,⍵}    ⍝ Create a function in ns called fn
      ⎕nl-9          ⍝ List the names of objects in the current namespace
      ns.⎕nl-⍳9      ⍝ List all names in ns
      )cs ns         ⍝ Change into ns
      ⎕this.⎕nl-⍳9   ⍝ The current namespace is ⎕THIS
      #.⎕nl-⍳9       ⍝ The root namespace is #

Mutable objects

Variables are pass-by-value. This means that if one name is used to assign another name, changes to the first name are not reflected in the second name.

      var1←1 2 3
      var2←var1     ⍝ The value of var1 is assigned to var2
      var1←var1+6   ⍝ The value of var2 is changed
      ⎕←var2        ⍝ var2 retains the previous value
1 2 3

Namespaces are objects and are pass-by-reference. All names which are assigned a reference (or pointer) can be used to refer to the original object.

      )ns ns1


  1. Should there be a difference between (⎕NS⍬)(⎕NS⍬) and (2⍴⎕NS⍬)?
    1. With "by value" semantics?
    2. With "by reference" semantics?
  2. What properties should two namespaces have in order to compare equal? Which of the following scenarios would be faster?
    • Using "by value" semantics?
    • By reference: when they originate from the same call to ⎕NS?
  3. Can you find a way to achieve "by value" semantics when using namespaces?
Discussion notes
  1. Should there be a difference between (⎕NS⍬)(⎕NS⍬) and (2⍴⎕NS⍬)?
    1. No difference in this case, there are just empty namespaces: (2⍴⎕NS⍬)≡(⎕NS⍬)(⎕NS⍬)
    2. Yes. In (⎕NS⍬)(⎕NS⍬) we create two entities. In (2⍴⎕NS⍬) we create only one.
  2. What properties should two namespaces have in order to compare equal? Which of the following scenarios would be faster?
    1. By value: when the two namespace have the exact same content. This is slow as it requires recursively inspecting each namespace for its contents and doing comparisons on every member.
    2. By reference: When the two namespaces originate from the same call to ⎕NS. This is fast as it requires only the comparison of two pointers.
  3. Clone (or make a deep copy of) a namespace
          ns1.vec←1 2 3
          ns3←⎕NS ns1
          ns3.vec[2]←10   ⍝ Only the new copy is changed
          ]disp ns1.vec ns3.vec
    │7 8 9│7 10 9│

Dyadic execute

Namespaces can be used as a simple dictionary structure.

      value ←   ⍝ Get value ← value   ⍝ Set value

But how do we do this when the name can vary? Use variables to store the name.


      ns⍎name,'←1 2 3'
10 20 30


Beware: is potentially dangerous:

      name←'⎕OFF⋄'   ⍝ !!!
      value←1 2 3
      ns {⍺.{⍎⍺,'←⍵'}/⍵} name value

In production, validate your arguments.

Modifying namespaces

In some applications, it is useful to keep a namespace as an object that holds the state of something.

It is reasonable in this case to write a tradfn that uses a namespace reference as both its argument and result. While tradfns can have the side effect of modifying a namespace, we strongly recommend that functions take arguments and return results anyway.

      ∇ nsref ← rate Modify nsref
         nsref.var +← rate

Code organisation

Having modular code organisation is a very sensible idea. However, it is a recent invention relative to APL. In a newer application, it is not hard to imagine a utils namespace, a maths namespace, an interface namespace and so on. In older applications, such compartmentalisation is often achieved using a naming convention. For example:

      ⎕NL 3

This type of code organisation is known as a flat workspace structure.

Names and references

In Dyalog there are both named and unnamed namespaces. Considering our use of the word "name" to refer to a token in the workspace, beginning with a letter, that refers to an array or function or operator, this is certainly a little confusing.

When creating or modifying a namespace, we can assign the namespace to a name which is a namespace reference.

      nsref ← ⎕NS⍬
      nsref.var ← 1 2 3

However, we can also use dyadic ⎕NS to give the namespace a name. This name also becomes the namespace's default display form. The default display form of an unnamed namespace is [Namespace].

If we assign a reference at the same time, we now have a two ways to refer to the namespace. However, notice that the namespace's name (shown by the display form) stays the same throughout as we are referring to the same namespace object.

      nsref ← 'nsname'⎕NS⍬   ⍝ Assigning to the reference is optional
      ⎕←nsref ⋄ ⎕←nsname
      nsref2 ← nsname

Display form

The display form of an APL array is what is displayed when you execute an expression to display it:

      x ← 1 2 3
      x     ⍝ The display form of x (depending on ]box settings)
1 2 3
      ⎕←x   ⍝ The display form of x (depending on ]box settings)
1 2 3

Namespaces have a default display form. The display form can be altered with ⎕DF:

      abc.⎕DF '' ⋄ abc
      def.⎕DF 'Hello' ⋄ def
      (abc def).⎕DF ⎕NULL ⋄ abc def  ⍝ reset to default
 #.[Namespace]  #.def 

Scripted namespaces

Store the source namespace as a single piece of text:

      )ed ⍟ns  ⍝ equivalent to '⍟'⎕ED'ns'
      ⎕FIX  ':Namespace ns'  'var←123'  ':EndNamespace'
      ⎕SRC ns
 :Namespace ns  var←123  :EndNamespace 


Every time we ”fix”, the namespace is reset per the script. This means that ns.var←456 would become 123 since the definition var←123 is found in the script.

      ⎕SRC ns
 :Namespace ns  var←123  :EndNamespace 
      ⎕FIX ⎕SRC ns  ⍝ similar to editing and fixing

The table below compares some methods for a namespace in the workspace versus a namespace script with text source.

Display Form Script
Set: ⍵.⎕DF Set: ⎕FIX source
Get: ⍕⍵ Get: ⎕SRC ⍵
Reset: ⍵.⎕DF ⎕NULL Reset: ⎕FIX ⎕SRC ⍵


It is possible to have a script depend on other scripts. To do so, use the :require keyword.

Problem set 9

  1. Are namespaces created with )NS scripted or unscripted?
  2. Use ⎕FIX to create an unnamed, scripted namespace

    What happens if you try to edit this script using the Dyalog editor?

  3. Request handler

    1. Create a namespace called req containing
      • a variable status with the value 200
      • a method Method which is the function {4+2×⍵}
    2. Within the req namespace, apply Method to status and store the result in status
  4. Write a function Into that copies a workspace into a namespace

          'dfns.dws' Into dfns
          dfns.disp dfns.morse 'SOS'


    See the documentation for ⎕NS.

  5. Write a function that swaps the values of two variables, the names of which are given as a 2-element nested vector of character vectors .

          ns.(aa bb)←10 20
          ns your_function 'aa' 'bb'
          ns.(aa bb)
    20 10

  6. Write an expression that swaps the values of the variables named x in the namespaces ns1 and ns2 .

          ns1←⎕NS⍬ ⋄ ns2←⎕NS⍬
          ns1.x←10 ⋄ ns2.x←20
          ns1.y←30 ⋄ ns2.y←40
          ns1.x ns1.y ns2.x ns2.y
    20 30 10 40

  7. Write a function ScalarRef that returns a scalar Boolean value indicating whether its argument is a scalar namespace.

          ns←⎕NS⍬ ⋄ ns.a←10
          ]disp ScalarRef¨ ns.a 'abc' (ns ns) ns (⎕ns⍬) 42
    0 0 0 1 1 0

    Use one or more of these scalar namespace properties:

    • Name Class (⎕NC) is 9 (ref; non-scalar arrays and non-nss are 2)
    • Data Representation (⎕DR) is 326 (pointer)
    • Depth (≡) is 0 (simple scalar)
    • Allows dot syntax (
  8. Write a function RefMask that returns an array of the same structure as its argument, but with bits indicating the namespace references.

          ]disp RefMask (⊂⊂⊂⊂1 2 ns) 3 ns (2 2⍴'abc',⎕NS⍬)
    │┌───────────┐│0│1│0 0│
    ││┌─────────┐││ │ │0 1│
    │││┌───────┐│││ │ │   │
    ││││┌─────┐││││ │ │   │
    │││││0 0 1│││││ │ │   │
    ││││└─────┘││││ │ │   │
    │││└───────┘│││ │ │   │
    ││└─────────┘││ │ │   │
    │└───────────┘│ │ │   │

  9. Write a function Fetch which takes namespace reference as left argument and a nested vector of character vector keys as right argument and returns the corresponding values from the namespace.

  10. Write a function IsRoot

          FindRoot ⎕SE.Dyalog.Utils
      FindRoot #
      FindRoot ⎕NS⍬

  11. What are our roots? Write a function FindRoots that takes an arbitrary array of namespaces and finds the root for each namespace.

          Line ⎕SE.Dyalog.Utils
     ⎕SE  ⎕SE.Dyalog  ⎕SE.Dyalog.Utils 

    Line ⎕ ⎕SE ⎕SE.cbbot ⎕SE.cbbot.bandsb2 ⎕ ⎕

  12. Where are my children? Write a function that lists all the children of a given namespace.


    Note: ⎕NL is Name List, not Children List
    Plan: You'll have to crawl through the entire workspace
    Think: How could namespaces still be out of reach?