nil: historical, theoretical, comparative, philosophical, and practical perspectives
Franklin Chen
Pittsburgh Ruby
September 6, 2012

Presenter Notes

About me

  • my first Pittsburgh Ruby meeting: November 2010
  • programmer at CMU, psychology department

Approach to programming

  • be pragmatic
  • improve
  • apply theory to practice
  • understand we are human

Presenter Notes


  • Ruby
  • Java
  • Python
  • Perl
  • Lisp
  • ML
  • Haskell
  • Scala
  • C++
  • JavaScript
  • XSLT
  • etc.

Presenter Notes

Why I'm giving this talk

Steel City Ruby Conf in August 2012: my first Ruby conference

Josh Susser, Thinking in Objects

Josh advocated monkey-patching of nil to remove a nil check in an if/else.

Monkey patching cartoon


Many lightning talks at conference


Presenter Notes

History: "My billion-dollar mistake"

I call it my billion-dollar mistake. It was the invention of the null reference in 1965. At that time, I was designing the first comprehensive type system for references in an object oriented language (ALGOL W). My goal was to ensure that all use of references should be absolutely safe, with checking performed automatically by the compiler. But I couldn't resist the temptation to put in a null reference, simply because it was so easy to implement. This has led to innumerable errors, vulnerabilities, and system crashes, which have probably caused a billion dollars of pain and damage in the last forty years.

Who said this in 2009?

Presenter Notes

Tony Hoare

  • invented and implemented null (1965)
  • invented and implemented Quicksort (1960)
  • invented much other important stuff
  • won Turing Award (1980):

fundamental contributions to the definition and design of programming languages

Tony Hoare in 2011

Presenter Notes

History: "Pornographic programming"

The unexpected appearance of an interpreter tended to freeze the form of the language, and some of the decisions made rather lightheartedly for the ... paper later proved unfortunate. These included ... the use of the number zero to denote the empty list NIL and the truth value false. Besides encouraging pornographic programming, giving a special interpretation to the address 0 has caused difficulties in all subsequent implementations.

Who said this in 1979?

Presenter Notes

John McCarthy

  • invented Lisp in 1960
  • implemented nil
  • invented the term and field of Artificial Intelligence
  • won Turing Award (1971)

John McCarthy in 2006

Presenter Notes

Our heroes: null vs. nil

null: billion-dollar mistake

Tony Hoare in 2011

nil: pornographic programming

John McCarthy in 2006

Presenter Notes

Topics to discuss

  • nil and null in Ruby and other languages
  • the problems
  • reasons for the honest mistakes
  • solutions to the problems

Presenter Notes

Starting with basics: Ruby nil

Who has never seen this output when running a Ruby program?

nil.rb:2:in `<main>': undefined method `name' for nil:NilClass (NoMethodError)

Point of failure:

puts 'Dr. ' +

What went wrong?

(Trick question.)

Presenter Notes

Not only a Ruby problem


C++ is unsafe: fails catastrophically on NULL dereference:

Segmentation fault: 11

Point of failure:

cout << "Dr. " << person->getName() << endl;


Java (like Ruby) is safe: checks for null dynamically:

Exception in thread "main" java.lang.NullPointerException
        at Null.main(

Point of failure:

System.out.println("Dr. " + person.getName());

Presenter Notes

Ruby nil's cousins

  • nil (Smalltalk, Lisp)
  • null (Java)
  • NULL (C)
  • nullptr (C++)
  • undef (Perl)
  • None (Python)
  • None (OCaml, Scala)
  • Nothing (Haskell)
  • etc.

Most of these engage in some form and extremeness level of pornography.

Of those, we focus on Ruby's nil.

Presenter Notes

nil in Ruby

Taken from Smalltalk, invented in 1971, inspired by Lisp.

  • class NilClass
  • singleton object nil of NilClass

Some methods:

  • &
  • nil?
  • to_a
  • to_s
  • etc.

Presenter Notes

Philosophy: ethics


Who does what to whom?


Who intends to do what?


Who should do what?


How does one do something?


How to believe someone has done the right thing?

Presenter Notes

Ruby nil example, revisited


nil.rb:2:in `<main>': undefined method `name' for nil:NilClass (NoMethodError)

Code that crashed:

puts 'Dr. ' +

"What went wrong?"

Three cases:

  • unintentionally uninitialized, left at default value of nil
  • deliberately assigned nil
  • deliberately assigned a reference to some other object, which happened to be nil

Presenter Notes

1: Unintentionally uninitialized

Problem: forgot to initialize person as intended.

Presenter Notes

1: Initialization solution


  • track back
  • find responsible code path
  • insert initialization
  • initialization logic: responsible not to return nil
  • obtain initial value from a source of trust, which never returns nil

New code somewhere:

person = => 'Smith')

Note: will never return nil if it succeeds at all.

Presenter Notes

2: Deliberate nil

Problem: person was intended to be nil, to indicate there is no longer a person.

Elaboration: there was a person in the room, but the person left, so person was set to nil, and there should be no greeting.

Presenter Notes

2: Deliberate nil solution


  • caller trusts that person could be nil
  • caller is responsible to check for nil
  • caller is responsible for adding appropriate action in case of nil

New caller code:

if person.nil?
  puts 'I see nobody here'
  puts 'Dr. ' +

Note: putting in a nil check is not enough; must provide entire code path.

Presenter Notes

3: Deliberately assigned reference

Problem: person was assigned some_method(...) at some point.

person = room.find_tallest_person
puts 'Dr. ' +

Presenter Notes

3: No solution: problem propagated

New problem: locally, without further context,

  • don't know whether person is ever intended to possibly be nil
  • don't know whether some_method is responsible for giving us nil or non-nil
  • don't know whether the caller code has any responsibility to check nil
  • we don't know what action to take, and whom to trust

The big problem with nil: any assignment other than to nil itself or a constructor call might or might not result in nil.

Presenter Notes

3: "Defensive" nil checks?

Solution (?):

  • caller has trusts nobody: check every single variable read with nil?
  • caller always responsible for some action for any possible nil branch


  • code bloat
  • inefficiency
  • loss of clear intention and trust
  • loss of incentive for anyone to be responsible for guaranteeing non-nil, since everyone will check anyway

How many people code in this way?

Presenter Notes

3: Give up goal of no nil exceptions?

Solution (?):

  • use a few defensive nil checks
  • document (with YARD?) what can or cannot be nil
  • write unit tests, acceptance tests, etc.
  • handle exception at some high level in program


  # deeply nested code...
rescue NoMethodError => e
  # restart the app at some good state

Presenter Notes

Bad monkey-patching

Ruby allows monkey-patching: modifying a class.

Solution (?):

class NilClass
  def name
    'Default Name'

person = nil
puts 'Dr. ' +


Dr. Default Name


  • global nil responsible for a person's name??
  • worse than failing: produces wrong output!!

Presenter Notes

Even worse monkey-patching

Not a solution either:

class NilClass
  def method_missing(*args)

person = nil
puts 'Dr. ' +


nil-monkey-all.rb:8:in `+': can't convert NilClass to String (NilClass#to_str
                            gives NilClass) (TypeError)
        from nil-monkey-all.rb:8:in `<main>'


  • silently swallows up a whole series of nonexistent methods
  • fails far worse than original code that failed fast at name
  • puts would have succeeded, unintentionally

Presenter Notes

Philosophy: what is nothingness?

absence of a Person

should not be the same concept as

presence of a NilClass

Presenter Notes

History: why null and nil?

Tony Hoare and John McCarthy knew how to do things right in the 1960s, but chose not to!

  • Hoare invented null for a statically typed language for efficiency and his convenience as a compiler writer
  • McCarthy implemented "pornographic" properties of nil for a dynamically typed language for efficiency and his convenience as an interpreter writer

"Those who cannot remember the past are condemned to repeat it." (George Santayana, 1905)

Presenter Notes

Fixing null and nil

  • null: replace with static type system as in functional languages
  • nil: replace with object-oriented design pattern

Explore both: duality between functional and object-oriented languages.

Start with functional.

Presenter Notes

A basic static type system

  • computation as a branch of philosophy
  • static type system: a language for understanding programming languages

Examples in Haskell syntax for convenience.

(Note: intended semantics is ML: Haskell's lazy types introduce complications.)

Presenter Notes

Primitive types

Primitive types:

example1 :: Int
example1 = 42

example2 :: Char
example2 = 'c'

example3 :: ()
example3 = ()


Presenter Notes

Product types (records)

Tuples of type A and type B and ...

Most languages add syntactic sugar and semantics, as records or structs or classes.

type Person = (Int, String)

person :: Person
person = (42, "Sally")

example :: (Bool, Person)
example = (True, (42, "Sally"))

Real Haskell has records:

data Person = P { id :: Int, name :: String }

person :: Person
person = P { id = 42, name = "Sally" }

Presenter Notes

Sum types (tagged unions)

Type A or type B or ..., value tagged with which one it is.

Familiar from ALGOL (1960s), Pascal (1970s), etc.

data Color = Red | Blue
example1 = Red
example2 = Blue

data Bool = False | True

data Key = Name String | ID Int
example1 = Name "Sally"
example2 = ID 12

data Login = Guest | User Key
example1 = Guest
example2 = User (Name "Sally")
example3 = User (ID 12)

Presenter Notes

Function types

Function type from type A to type B:

addInts :: Int -> Int -> Int
addInts x y = x + y

greeting :: Person -> String
greeting person = "Dr. " ++ name person

Note: greeting never gets a null and never fails.

The type system guarantees that greeting can only be called on a valid Person.

Presenter Notes

But we need nothingness!

In real life, need a way to express: x can be bound to a Person or to nothing.

Very easy: use a sum type!

data MaybePerson = Nobody
                 | One Person

-- OK
person :: Person
person = P { id = 42, name = "Sally" }

-- type error!
personFAIL = Nobody

-- OK, nobody
possiblePerson1 :: MaybePerson
possiblePerson1 = Nobody

-- Wrapped person
possiblePerson2 :: MaybePerson
possiblePerson2 = One (P { id = 12, name = "Jack" })

Presenter Notes

Dealing with a possible Person

Functions can be defined to require a Person or MaybePerson:

greeting :: Person -> String
greeting person = "Dr. " ++ name person

-- will not type check
greetingForMaybePersonFAIL :: MaybePerson -> String
greetingForMaybePersonFAIL person = "Dr. " ++ name person

-- type checks
greetingForMaybePerson :: MaybePerson -> String
greetingForMaybePerson maybePerson =
  case maybePerson of
    Nobody     -> "I see nobody here"
    One person -> "Dr. " ++ name person

Presenter Notes

No silver bullet

This technique does not prevent buggy code!

-- type checks, but compiler warns of missing case of Nobody
greetingForMaybePersonBAD :: MaybePerson -> String
greetingForMaybePersonBAD maybePerson =
  case maybePerson of
    One person -> "Dr. " ++ name person

-- run this code
greetingForMaybePersonBAD Nobody


Exception: Non-exhaustive patterns in case

But at least the compiler warns you of code branches you were supposed to write.

Presenter Notes

Wrapping up on types: parameterized

Parameterized types (generics) use type variables such as a, b.

C++ templates, C# generics, Java generics, etc.

type NestedTriple a b c = (a, (b, c))

example1 :: NestedTriple Boolean Int String
example1 = (True, (42, "Sally"))

Presenter Notes

Maybe as parameterized type

The static typing way to mitigate the nil problem:

data Maybe a = Nothing | Just a

Avoid having to create a special MaybeX wrapper type for every type we want to have a nil possibility for.

possiblePerson:: Maybe Person
possiblePerson = Maybe (P { id = 12, name = "Jack" })

Standard ML, OCaml, Scala also have this parameterized type, except that they use different words: Option instead of Maybe, None instead of Nothing, and Some instead of Just.

Presenter Notes

Maybe example: intention

(Back to ethics.)

Every variable has a static type.

A type is an intention.

Presenter Notes

Maybe example: responsibility

Every variable, because it has a declared type, forces an accessor to take responsibility for how to use it.

A variable not of a Maybe type can never be bound to Nothing.

The type checker of the compiler enforces everyone's responsibility.

Presenter Notes

Maybe example: means

Means: static type checker of compiler.

Presenter Notes

Maybe example: trust

Trust (for type correctness) is delegated to the compiler.

(But type correctness does not guarantee code correctness.)

Presenter Notes

Maybe example: action

Type forces us to take action: unwrap a Maybe value to treat both cases.


  • no guarantees that you will write the correct code for each case
  • no guarantee that you won't just defeat the type system by declaring everything to be of Maybe type

Maybe is a tool, not a solution to programming.

Java and similar so-called statically typed languages: really duck typed languages!

  • every reference can quack like its declared type or quack like null
  • null quacks unconditionally by either a segmentation fault or an exception

Presenter Notes

Maybe, conclusion


  • static type system: tool to distinguish between nothing and something
  • for each type a, the type Maybe a generates its own unique Nothing: no global nil
  • helps with: intention, responsibility, means, trust
  • we still have to act for ourselves: think and code

Presenter Notes

Maybe: back to history

  • Hoare knew how to do null right in the 1960s
  • Hoare weakened his static type system for efficiency and convenience
  • ML had Maybe in the 1970s and 1980s
  • C++, Java and similar languages give static type systems a bad name

Reminder: purpose of static type system is to aid human reasoning; not primarily for efficiency!

Presenter Notes

Back to Ruby and nil

  • Ruby: not statically typed
  • How apply the lessons from types?


  • Use Ruby-oriented Maybe wrapper
  • Use "null object" object-oriented design pattern

Presenter Notes

Maybe for Ruby

  • many Maybe-inspired Ruby libraries

I have not used any of these examples, but should examine for correctness and usefulness

Presenter Notes

Null object design pattern

Exploit duality of functional and object-oriented design:

  • each part of a sum type becomes its own class
  • each function that does a case analysis becomes a method

Original code:

def greeting(person)
  if person.nil?
    puts 'I see nobody here'
    puts 'Dr. ' +

person = nil
greeting person

Presenter Notes

Transform functional to OO

New code:

class Person
  # ...

  def greeting
    puts 'Dr. ' + @name

class NoPerson
  def greeting
    puts 'I see nobody here'

person =

Presenter Notes

Drawbacks of null object pattern

  • inversion of design may not be natural
  • works with simple examples with limited case switching to extract
  • not possible when switching on more than one value


Presenter Notes

History continues

  • Efficiency still matters: Scala debate at this very moment (September 2012) about whether to optimize away the indirection of Option at compile time: Hoare's concern in 1960s!
  • Still much discussion in Ruby community about nil-handling patterns

Presenter Notes


  • nil handling is a real problem
  • think in terms of ethics
  • static type systems give one approach
  • Ruby can be used in functional or object-oriented style
  • we should know and learn from history
  • it's not over
  • we still have to write correct code

Franklin Chen

This slideshow:

Presenter Notes