Beautiful Code is a unique book. It is not bound by a particular system, or programming language, or methodology. Instead, it tries to straddle the entire field of programming with the ostensible aim of exploring beauty in computer code.
What we get is a smorgasborg of essays that posits competing definitions of technical beauty in very different pieces of code. As such, the book becomes a dialectical argument between different conceptions of technical beauty. It may even serve as a rorsarch test to your sensibilities as a programmer. In the process, you may discover algorithms that are startling in their beauty and ingenuity.
Given the broad sweep of the book, the quality of the essays is somewhat of a crap-shoot. What I found more interesting was working out why I found some of the essays elegantly persuasive, whilst others, I found to be a turgid sludge to wade through. In the end, it becomes almost impossible to separate the quality of the writing from the definition of beauty defined in the essay.
Technical beauty is a very strange beast. It is different from the kind of beauty that resides in the curve of Scarlet Johanssen's lips. In physics and maths, there is a tradition of invoking beauty in certain equations. Einstein's equations for general relativity surely merits that description, as does Euler's formula. For me, technical beauty refers to that rare fusion of concision, efficiency and surprisingly deep connections between seemingly unrelated phenomena.
I found that the essays that I liked were the ones that focused on very specific pieces of code. These essays would carefully explain the problem that the programmer was trying to solve, articulate the constraints, and show the straightforward (and usually less elegant) alternatives. They would then go through all the blind-alleys before unveiling the final solution, which would be as surprising as it was elegant. After all, the father of the essay, Montaigne, coined the term "essai", which means attempt in french.
The essays that didn't work would try to describe how the software worked, and skip over the details. But these are technical essays, after all, and without a careful consideration of technique, all that is left is flabby writing. Surprisingly, there were quite a number of essays devoted to scientific programming, with contributions from the writers of BioPerl, Numpy, LAPACK, and the NASA Mars module. Most of these I found rather unsatisfactory, the kind of flabby essays that skipped over interesting technical details. They offered no fundamental insight to how the code was implemented other than the fact that these packages have stood the test of time. The exception was Trevor Oliphant's essay on Numpy, which showed how a powerful iterator implementation can dramatically simplify the organization of a linear algebra library.
How detailed were the techniques? In Henry S. Warren Jr's essay, he describes how one might count bits in C. It's a marvelous essay because through the description of such an elementary operation, Warren illuminates just how close to the metal one can go in a low-level language like C. At the other end of the scale, Jeffre Dean and Sanjay Ghemawat, engineers at Google, describe the MapReduce algorithm, part of the secret sauce that squeezes out information from Google's gigantic parallel arrays of hard-disks and processors.
There is a the famous dictum attributed to Fred Brooks that if you know the data structure of a program, you will know how the program works. The design of the key data structure reveals the inner workings of a piece of software better than any flow-chart ever could. Greg Kroah-Hartman describes an object-oriented design right in the heart of the Linux kernel driver, written in nothing less than C. Here is found one of the scariest looking C macros that I have ever seen. In one of the most twisted pieces of logic, Kroah-Hartman argues that by not simplifying the macro (including such hand-holding as run-type checking), it keeps away people who don't know what they are doing from touching the code. That, for Kroah-Hartman, is beautiful code. Andrew Kuchling's essay explained how dictionaries are implemented in Python. In the process, I finally understood how dictionaries underpin virtually everything in Python, from objects to locally scoped variables. That's why dictionaries in Python need to be as fast as they are.
Two essays introduced me to some radically different programming techniques. Charles Petzold's essay on writing fast image filters shows how you can do custom compilation on very small pieces of code for incredible speed gains. The essay by Andreas Zeller described perhaps the strangest algorithm in the book. He was working a GNU debugger GUI front-end that got broken after the back-end debugger was upgraded to a newer version. Rather that go through each of the 10000 different patches individually by hand, Zeller came up with a perversely beautiful technique to systematically identify which patch broke the GUI front-end.
Some of the essays flew straight over my head, but such is the nature of such an eclectic collection as this. Beauty, some might argue, is the ability to see hidden patterns in the world, so it seems appropriate to mention Brian Kernighan's essay on a terse 35 line C-program that implements a regular expression matcher. It makes extensive use of recursion, of course.