sheleketto

Most of the reviews here seem to praise this book. Engineers I work with, themselves skeptical of the agile movement, found the book an outdated artifact of the long-deprecated waterfall model of software development. I found the truth somewhere in between: the book is full of important nuggets of wisdom, many of which are under-appreciated in the broader software engineering culture, but these bits are mixed with tons of more dubious advice that's often as bureaucratic as the book itself. That said, this book is quite thoughtful, and nearly all of the topics are important issues to consider, even if I found many of the specific suggestions less useful. Random examples of ideas that resonated strongly with me: * Ch5: Design is "wicked", and you're best off iterating with both top-down and bottom-up approaches and using loose heuristics rather than rigid algorithms. * Ch9: "One of the biggest differences between hobbyists and professional programmers is the difference that grows out of moving from superstition into understanding." Amen. Similarly, in the debugging chapter, I strongly agreed with the emphasis on truly understanding the root cause of an issue and all of the observed behavior before fixing the problem. * Ch10: All of the comments on scoping and the reasons to minimize scopes. * Ch23: Totally agree with the general attitude towards debugging, and the notion that programmers learn a lot from debugging. I've probably learned 2/3 of what I know from my experiences debugging, and those experiences inform every line of code that I write. That said, the notion that one should seriously consider giving up on a bug after less than an hour of debugging is ridiculous. * Ch24: Parts of this chapter on refactoring and "design-ahead code" are remarkably concise and forceful. I can't tell you how many times I've written code for a feature I was sure we were going to want, but didn't quite flesh out because we weren't sure exactly how it should work, and that code turned out to be poorly exercised, buggy, and ultimately ripped out -- not because of the poor quality, but because we simply didn't need the feature, or the feature needed to work very differently than we initially anticipated. Of course, like all things, there's a balance, and one has to consider each case individually. * Ch32: Bad idea to refer to bugids and other historical artifacts in code. The source should represent the current state of the system, and the bug tracker and source control document what changes were made when, by whom, and why. By contrast, there's lots I took issue with: * Ch5: Many of the examples, particularly in the design section and with respect to OOP, are highly simplistic. * Ch6: Encapsulation as typically lauded in OOP texts is often a sham, in my experience. Realistic problems rarely lend themselves to simple, clean abstractions. Even traditionally successful abstractions, like Unix file descriptors and POSIX open/read/write/close, reveal a lot about the underlying implementation, and one cannot use them successfully in complex programs without understanding details about the implementation. * Ch7: The Bjarne Stroustrup quote about macros (that each macro demonstrate a flaw in the programming language) is exactly the kind of dogmatic lunacy that gave us C++[11]. * Ch8: The discussion of error handling misses what I consider the most important question in determining what a component should do in handling an error, which is: can the current layer be *sure* it knows what the caller wants to do? If not, it should either propagate the error to the caller so that it can decide what to do with it, or else provide options so that the caller can indicate ahead of time what it wants to do (e.g., ignore, retry N times, or fail). If there's more than one consumer, err on the side of providing rich information to the caller so it can decide what it wants to do. There's nothing more infuriating than library functions that retry operatons after errors that you know aren't transient or which try to deal with a fatal error by papering over it. It's much worse than having to retry something the library should have known should be retried. * Ch9: The pseudocode-writing technique suggests thinking both implementation-agnostic and low-level enough that code generation is automatic. These two are always competing, and (in my experience) usually exclusive. * Ch13: There's an over-emphasis on a superficial notion of "safety", and the suggestion that things like void pointers in C lead to serious runtime bugs. Despite the arguments for "proper" generics in higher-level languages, in my experience, improper use of void*'s in data structures is not even remotely an important category of runtime bugs. One of the "C" examples in this chapter isn't even valid C. * Ch16 is full of small errors and some braindead code (e.g., "if (...) { assert(false); }"). The book is divided into several sections: * The first few chapters on metaphors and the notion of construction are thoughtful and unusually self-aware for this kind of "advice" book. * The next several chapters on requirements and design are a microcosm of much of the rest of the book, containing a smattering of useful ideas in a sea of more dubious, often bureaucratic advice. * The middle chapters on actual writing and structuring of code are probably the best. * Some of the later chapters on debugging, refactoring, style, and so on, are pretty thoughtful. Some of the others (like collaborative construction and tools) feel like useless catalogs of techniques and software packages: they're not immediately useful to the reader, nor are they a particularly good reference, but they seem to be there to flesh out an exhaustive description of software construction. One nice thing about the book is that the text itself is organized with useful headings and bolded main ideas, and each major section ends with main idea bullet points and a complete checklist, making it relatively easy to skim. This could help an experienced engineer pick out good parts that are worth reading more about. Unfortunately, the kinds of things that are most useful to learn are typically the things you wouldn't necessarily realize are important from a bullet point. As I mentioned, the book is very thoughtful, and the issues raised are all important and relevant for practicing engineers. I found about 30% of the advice strongly resonated with me, at least 50% was highly suspect, and the rest was pretty thought-provoking. For less experienced engineers who are good at picking things up from books like this, it might be a useful text. For experienced engineers, or the majority of us who learn subtle lessons much better by making the mistakes (or at least seeing real-life, personally meaningful examples), you may be better off learning by doing.

2023-07-26 08:44