Wishful Thinking

Created at: June 9, 2014 | Reading time: approx. 4 Minutes | View source on GitHub

To me, one of the toughest challenges of creating software is dealing with different levels of abstraction and detail. This mixture of abstraction-layers can lead to harder understandability and difficultly maintainable code.

In April 1984, Harold Abelson and Gerald Jay Sussman coined with the amazing book Structure and Interpretation of Computer Programs, short SICP an interesting term to tackle exactly this problem: Wishful Thinking.

The idea is quite simple: write your algorithm with methods that don’t exist (yet) but read very well and do this until you reach the lowest layer of abstraction. Let’s look at a short code example of a use case from a fictional monster fighting browser game:

class OrderMonsterToArena < UseCase
  Input = Bound.required(
  Denied = Bound.new
  Success = Bound.new(:fight_starts_at)

  def initialize(input)
    @defied_keeper_id = input.defied_keeper_id
    @competing_monster_id = input.competing_monster_id

  def call
    return Denied.new if !challenge_allowed?
    return Denied.new if !monster_obeys?

    Success.new(:fight_starts_at => scheduled_time)

  # ...

If you read the central call method, there is no detailed knowledge needed or present regarding the usage, retrieval or storage of entities. Also, no direct entities can be spotted. Instead, the code forms an abstract Domain Specific Language.

You can note further that none of these methods are actually implemented. I made them up by (wishfully) thinking “How would I name this method if I could name it anything?”

To get this code to work, you can reapply this process of wishful thinking to the individual methods:

class OrderMonsterToArena < UseCase
  # ...

  def challenge_allowed?
    keeper.has_matching_level_for? defied_keeper

  def monster_obeys?

  def scheduled_time

  # ...

In this step, the newly formed layer consists (seemingly) of nothing but different entities and their high-level domain methods. All the structural details of saving and retrieval are still hidden. The appliance of the wished for method led to the actual instances of our existing domain objects, the entities.

But how do we retrieve them? And more importantly, how are they saved to our abstraction of the persistence layer? Well, I think you can guess the answer: “Wishful Thinking”.

class OrderMonsterToArena < UseCase
  # ...

  def keeper

  def defied_keeper
    @defied_keeper ||=

  def competing_monster
    @monster ||=

  def monster_index
    @monster_index ||=

  def scheduled_fight
    @scheduled_fight ||=


This step forms the base layer of this usecase. The next step would be to enhance the used service-classes with the needed interfaces, if they don’t provide it already. I found it very interesting that a clear, layered structure evolved naturally by this approach. Every method can be easily modified without interfering too much with the depending layer. As an abstract graphical representation, this simple use case could be seen as follows:

Layer structure of OrderMonsterToArena

This perspective reveals an interesting pattern. Note how all the dependencies point away from the use case class - this allows for even drastic changes to my use case without interfering with my inner entities and services. Also, the entities have no dependencies at all. They are self-contained and as such very protected from interface changes.

Sidenote about Ordering of Methods

An additional topic relates to this concept of structuring complexity. In which order should one present the methods of different abstractions? A nice approach is to group them by their conceptual level. As in the given examples above, the three groups of methods make sense in the given context. While inspecting such a class, the reader can dive into the concrete implementations if they want to, or just stay at the semantic level. When done correctly, every group of methods will be readable and understandable, without the knowledge of the deeper underlying concepts.

Another approach is ordering by appearance. Here you would mix up the different layers, and always put a method under the method in which it was first used. This ordering of methods is proposed by Robert C. Martin in his Clean Code book.

I am not quite sure which one of these orderings leads to better understanding, so I won’t give a definite answer. My current tendency leans towards the first approach, since it does not mix up the different means of abstraction. I like this.


By programming against (probably not yet existing) interfaces one can assure that these interfaces yield the best possible usability. This approach also produces a nice grouping of the different abstraction layers inside of a class and almost automatically takes care of the dependency directions inside of your classes.

It took me some time to practice this style of thinking, but the reward is well worth the effort. The overall readability of your code will improve which creates better maintainability as well as one or two appreciative comments from your colleagues.

Browse all Articles