7 of 7 people found the following review helpful
wide-ranging in scope, but heavy on C++ templates,
This review is from: Generative Programming: Methods, Tools, and Applications: Methods, Techniques and Applications (Paperback)
This book does cover lots of techniques that fall under the rubric of generative programming, but be warned that there is a strong emphasis on C++ template metaprogramming. If you don't know C++, or are a bit fuzzy on template syntax, you'll be missing a lot. Consider this a C++ template metaprogramming book, with some bonus chapters to put it all in context.
The opening chapters are a bit dull, containing a plethora of acronyms and jargon about various design methodologies, none of which seem to have made any particular impact in the last few years. I tentatively diagnose a mixture of thesis-itis and maybe translation-itis. If it's the former, I wouldn't be surprised to find that this made up part of the introduction or literature review chapter. However, the concept of feature diagrams is quite interesting, as it allows a graphical representation of a design specifying features and other properties (such as whether they're optional), without requiring any implementation (e.g. inheritance or parametric polymorphism), which is not possible with UML. How revelatory this is may depend on how seriously you take UML as a modelling tool, versus a convenient set of boxes and lines for representing class design.
There are also chapters on Aspect Oriented Programming, which is a pretty good survey of the field, and which provides useful motivation beyond logging. Additionally, there's a chapter on generators, which provide a convenient Domain Specific Language for specifying behaviour and performance of software components (list containers is the example in the book).
There's also a rather vacuous chapter on intentional programming, the brainchild of Charles Simonyi. It's an interesting enough idea, sitting somewhere between MDA, the Smalltalk class browser, and a souped-up IDE, but there's been absolutely no progress on it since the book was published, because Microsoft didn't release it, and Simonyi wasn't allowed to take any of the code with him when he left to set up Intentional Software. This chapter is ok on the big idea, albeit a bit breathless given it's not been shown to produce any useful software, but the worked example has all the allure of an Eclipse plugin tutorial.
The above material is sporadically interesting, but often a little pedestrian. However, the real action of the book takes place in the chapters on template metaprogramming in C++, which demonstrates how to generate related families of classes at compile time, using template instantiation and careful use of inlining to avoid inefficient virtual calls and indirection. It's very impressive. There are several examples, including a class hierarchy for a banking domain, and an in depth treatment of a matrix library, which successfully unifies all the types (sparse, full, banded etc.) under one library, while maintaining performance competitive with Fortran.
These C++ chapters demonstrate the ideas of the book in practice most clearly, although it also underlines the need for new tools and language features, given the outrageous ugliness of the compiler/template-abuse that is metaprogramming. On the other other hand, it is kind of cool to see colons, angle brackets and other bits of C++ coalesce into a strange new dialect using the compiler as an interpreter. If you're going to get anything at all out of these chapters, you do need to be comfortable with C++ templates. It would probably also help to have read something like Barton and Nackman's book, 'Scientific and Engineering C++', their unorthodox template designs providing a good warm up for what's in this book.
If you don't program in C++, or don't care for template metaprogramming, you may not find huge value for money in the other chapters. And admittedly, the writing style hardly sets the pulse racing. But if you're looking to do some intellectual stretching before taking on Andrei Alexandrescu's Modern C++ Design (and that's probably a very good idea), step right up!