Linux Core Kernel Commentary: Guide to Insider's Knowledge on the Core Kernel of the Linux Code [Englisch] [Taschenbuch]

LCKC's main purpose is to show you how a real, modern OS works inside, by acting as your guide to its code. LCKC is also of interest to programmers who -- like me -- are interested in what happens to code in extreme conditions. In the kernel's case, those extreme conditions include unusually strong requirements for speed, clarity, robustness, correctness, portability, and extensibility. The inevitable conflicts among these goals only make the results more interesting.

More information is available at the book's support site, though I have unabashedly repeated the best of that information here for your convenience. (My site links back to Amazon.com, in case your "Back" button is broken. :-)

My combative temperament compels me to address the negative comments that a few reviewers have made about the book, though it's hard to do that without sounding defensive. Skip the rest, if you like.

It's true that, in round numbers, LCKC is 400 pages of code followed by 150 pages of commentary. But this is misleading in two ways. First, because C and English have different readability requirements, the code is printed in a different font from the commentary -- a larger, monospaced font -- which greatly exaggerates the difference in the relative sizes of the two sections. Second, the unusual (albeit very attractive) typeface and page layout in the commentary portion of the book shrinks its page count. By my estimate, the commentary portion would be about 250 pages if laid out normally.

Why include printed source with the book? It's an article of faith in the publishing world that people won't read large amounts of text from a computer screen, so the publishers didn't want to include only an electronic copy of the code. Splitting the book into two physical volumes, one for code and one for commentary, introduces problems of its own: potential buyers have to *know* that they need to print the code themselves or buy both volumes; if you try to solve this by putting both volumes in a shrinkwrapped slipcase, you drive up the cost of the book and make it impossible for bookstore browsers to peruse it. The chosen solution was the least of the available evils.

I cannot agree, either, that the book is badly organized. If you want to know about process scheduling, you read that chapter. If you want to know about memory management, you read that chapter, and so on. LCKC is emphatically not a reference book, though I suppose it can be used that way -- the code part of the book has pointers to the page numbers on which each discussed function is covered. (Of course, not all the code is covered -- once you've read the chapter, you can generally find your way around the remaining related code yourself. So it's not *primarily* useful as a reference work, and this may have been the problem for reviewers who were expecting something else.)

I am most sympathetic to the criticism that the book should have been arranged with commentary next to the code it's commenting on. The reason not to do it that way is this: code is code, and commentary is commentary. The best order to arrange the code may not match the best order to explain the overall structure of the code in English, and for the purposes of *this* book, it's better not to mutilate either the commentary or the code for the sake of the other. The point is to see what the code actually looks like, but without having to do all the work of figuring it out yourself, yet without sacrificing the integrity and coherence of either code or commentary. Think of it as an engineering tradeoff.

I do have a modest proposal for any readers who are adamantly opposed to flipping between the two sections: buy two copies. Leave one open to the code part, and one open to the corresponding commentary part. The fact that I get royalties has nothing to do with this suggestion, honest. ;-)