1. Top-Down Design
Most of the problems in the SICP book are solved in a top-down way with lower level details deferred until needed. The focus on high level details makes for expressive flexible code since implementation is based on well-defined interfaces and not implementations. Consequently, swapping and improving interfaces is a cinch – the dependency on high level interfaces shields consumers from tight coupling to details.
Picture a calculus system for integer operations (e.g. differentiation); as expected, these higher order functions need support for addition/subtraction/division operators. A top-down design will implicitly assume the add/subtract operators will work appropriately and focus on the higher level operators first. These low-level details can even be mocked and the system unit tested to prove logic accuracy.
The benefit of such a delayed implementation comes when the system needs to be extended to support complex numbers – the higher level operations are still the same, only the low level add/subtract operators need to be changed. Once done, the entire system should work as before.
Design is subjective and this is one approach; there might be better ways for other scenarios but again this can be tried and leveraged.
2. Generics: Data-directed vs Message-Passing Styles
Generics are useful for maintaining and extending existing systems.
The data-directed style maintains a huge table of known procedures and looks up desired operations based on keys. Extending such systems involves adding the new function signatures to the table; this allows the entire system to start using the newly added functions (sounds like the adapter pattern).
The downsides include the extra overhead of a procedures’ table, the need for consumers to know about available operations and the enforcement of contracts in the system. However, benefits such as flexibility, encapsulation and ease of accretion of new features can’t be ignored.
Message passing involves sending messages to objects to trigger the right operations. Sounds familiar? Yes, it does sound like OOP and is one of its major tenets. An object keeps a list of possible operations and invokes the right method based on the message it receives (think of invoking an instance method in an OOP language as sending a ‘message’ to the object e.g. a.b() ).
The question thus comes to mind, how do programming languages handle their primitive operations at a lower level? Do they do message passing? Keep tables?
The importance of context and how they affect semantics in all languages (both human and programming).
Using the example from the book; three is equal to one plus two however ‘three’ is not equal to “one plus two”. The context determines what is meant and humans have long used this to create various literary forms such as puns, sarcasm etc.
4. Huffman Encoding
Some computer science and maths eh, the Huffman encoding scheme allows you to use much less space provided the conditions are met.
5. Understanding the problem deeply before writing any code
Coding should be simple. Unfortunately, a shallow understanding the problem being solved makes it a difficult task.
Spending a good chunk of time to really understand the programming task at hand has great benefits, it leads to well-thought out designs, simple models (simple is difficult!) and less bugs. It also reduces the chances of providing the wrong solution. It makes for a more fulfilling experience too – what beats the thrill of having zero compile errors on the first attempt?
If you liked this post then do check out a couple more exciting discoveries from my SICP journey: