Very well-written. At 190 pages, quite likely finish-able in a weekend. But if you get 70% of the way through it, that is still very valuable. It makes sense as an underlying set of principles that explain why you would use design patterns.
Beyond his credibility as the creator of Tcl, in the talk he gave at Google about the book [1] he describes the course he teaches that was the inspiration for the book. The course is to write a word processor with a team and extended it with features over the semester, swapping codebases between teams so occasionally you have to work with other's code. Essentially the largest possible code project you can fit in one semester. Along the way, Ousterhout will give a deep code review of the each project. I think it must be very insightful to see the same project implemented so many ways, and to see what works in terms of simplicity and extensibility.
"This book does not consider functional programming, or the fact that terrible software in modern times is often the result of layers of incorrect abstraction that are often induced by the overuse of "object oriented" (OO) programming techniques. Instead, it takes OO programming for granted as being a good thing (visible in its preferences for two OO languages, C++ and Java), and then it reasons in that domain, making it totally uninteresting to many potential readers. The book also assumes that formal design doesn't work (he calls it "waterfall"), and takes alternatives for granted. I doubt the author is even aware of his inherent bias, but it is incredibly strong."
Does anyone here agree or disagree with these points?
The text isn’t about implementation or paradigm. It’s about things applicable in software design in general. For example, good interfaces are deep, rather than shallow. The examples of narrow interfaces are in Java because they were good examples. If you have to instantiate 5, 6 objects to use an interface, the interface is not deep. But if you had to do the same thing in a CLI interface or a GUI, it’s the same idea. You could apply the same ideas to functional APIs.
Although I think that the book provides a lot of very valuable insights, I missed a more differentiated view of the concepts presented in the book. For example, in chapter 4 ("Modules should be deep") Ousterhout dismisses the use of small classes and methods. He argues that using many small classes and methods makes the system as a whole more complex because many little pieces have to be managed. I completely agree with that statement. But only under the assumption that software systems are static and never-changing. If we consider software systems as ever-changing, I would argue that using smaller pieces often provides the flexibility necessary for maintaining large systems.
That's a very good point. Designing small components probably requires more effort and can also cause a lot of harm if done incorrectly.
Prior HN discussion: https://news.ycombinator.com/item?id=12120752
It is better to provide a clear and memorable picture that is only slightly wrong than a hedging and muddled picture that is still slightly wrong.