Thoughts On: On the Spectrum of Abstraction
12 August 2017
I came across this talk from React Europe Conference 2016 online and was particularly enlightened by some of the great points mentioned by Cheng Lou.
An interesting question that kicked off what eventually turned into a 30-minute mindblowing session was this:
If B is a subclass of A, why is A the superclass and B the subclass? If B has more properties than A, why do we call B a subclass/subset? Why is B not a superset/superclass?
This question was interesting not because of the technicality that it encompasses, but it sheds some light on the kinds of question that I often overlook in software engineering.
The uncomfortable questions that could only arise if I think deep. Thinking deep is something that I need to work on in my day-to-day life. I am not asking important questions that facilitates the understanding of certain concepts that I fail to grasp fully.
(By the way, it is called a subclass because there are lesser instances rather than methods/properties. We have to be clear of what are we counting.)
The speaker has came up with a ‘mental model’ to think about abstraction. We can view them as trees, with the root being the most abstract, and the leaf nodes as concrete use cases which are “useful”.
An example would be how React is a higher node in the abstraction tree as compared to a product that is built with React.
The notion of power is also defined as such:
“The ability of drill down the level of abstraction to get a concrete use case.”
Utility/Usefulness on the other end of the spectrum is defined as:
“Attached to concreteness, something is immediately useful if I don’t need to understand what abstractions we master to make it work.”
e.g. How React isn’t useful, since it is not useful for to be a product itself.
Cost would be to go up or down a level of abstraction.
Then we have the Principle of Least Power/Pareto Principle
“Use least powerful tools to achieve goals, because of the lesser mental overhead.”
There were many good case studies brought up in the video, but I’ll note down some that were especially meaningful to me.
Case Study: Library vs Framework
This mental model of abstraction is particularly useful when we try to differentiate between a library and a framework.
Libraries are lower in the tree, therefore more useful. Frameworks are higher in the tree, and hence covers more use cases and shared knowledge.
This is a very refreshing way to look at libraries vs framework, as CS2103 tried to explain it in a pretty archaic way if I remember correctly. By looking at abstraction as a tree, I can point out exactly what level of abstraction I am currently on, and my choices for going up or down the abstraction tree.
Case Study: Immutability vs Mutability
I have always had trouble understanding the arguments going for and against immutability. From my experience, it has always been about several zealots throwing buzzwords at me to get me convinced that the more buzzwords a certain paradigm could provide, the better it is.
Cheng Lou expressed this very well with the abstraction tree, and this quote:
Not about the potential power of doing more - it’s about exploiting properties you gain by doing less.
Personally, this suddenly made so much sense.
It’s difficult to reason with mutability when one asks “I can do X with mutability, can you do it in an immutable way?”, and the immutable zealot goes “Well we could do X this way with immutability.” What eventually happens is the mutability lover has an elegant solution for a particular problem that immutability cannot solve as elegantly.
Which totally makes sense, since mutability is higher than immutability in the abstraction tree. But then the quote kicks in. With immutability, we have other cooler features that are suddenly easier to implement, such as time traveling.
An interesting advice given by my professor in CS3217 last semester is that every question in software engineering has a standard answer.
Which implies that there are trade-offs in every solution that we decide on. The quote by Cheng Lou complements well with that, since now we have a way of understanding the different trade-offs that we are making.
Are we drilling up or down in the abstraction tree? How are our requirements going to change over time?
It isn’t just about how high our abstraction is in the current state of the software, it’s about deciding on the right level of abstraction.
Case Study: JS (inline) CSS vs Traditional CSS
These are all things that has to be achieved using different tools, not supported by CSS itself.
One final nail in the coffin for why CSS is at the wrong level of abstraction, preprocessors. We have LESS and SASS and many others to help CSS stay on the correct level of abstraction.
CSS needs actual functions, rather than the current DSL. We need to move stylesheets into a higher level of abstraction to keep the cost low.
What better way than to learn from other people’s mistakes, right?
Here are a few that the speaker mentions:
- Don’t cover every use-case
- Less work = more performance gain
- Clear justification to not cover every use case because of the abstraction tree
- More use-case = more mental overhead
- Not DRY (i.e. repeating code is fine)
- We are paying more cost to abstract things rather than to use them!
- Don’t be swayed by elegance
- Elegance is too loaded for a word
- “The single unifying principle” is too general to be used in a specific way
- When in doubt, use examples
- Examples are leaf nodes on the tree of abstraction
- Use now, understand later
- When lacking ideas of how to express a library, write examples, write documentations
The last point is probably the golden advice. Many times I fret too much about the details and could not move forward with implementation. This would be a great time to fill in some documentation so that I maintain a clear head of which level of abstraction I am in right now.
Also, test cases are the perfect examples to write as well! :)
The talk ended with a pretty apt example of how this talk itself is “the most useless talk of this conference.” But by definition, the talk is also the most power talk of the conference. If we do not apply, but only think in theory, then we won’t have much use for the talk at all.
The next time I think about writing an application or library, it’d probably be good to apply some of the mental model learnt from here.