Here is a quick review of Loving Common Lisp, by Mark Watson. I initially read an early draft version on Leanpub, but I found some time to (re)read the latest version available.
If you are new to Common Lisp, I would recommend to first try the excellent blog post written by Steve Losh, A Road to Common Lisp. Don’t miss it, it’s very well written, like many other posts by the same author. I would also recommend the Common Lisp Cookbook. There are plenty of additional resources to learn Lisp, either from a pragmatic viewpoint – Atabey Kaygun writes a lot of math/stat-related algorithm using CL on his Tumblr – or from a more formal perspective: O’Reilly has a lot of dedicated textbooks targeting the “modern Lisp”, aka Clojure, and I particularly love the one written by Carin Meier. If you want to stick with CL, strictly, then Practical Common Lisp is probably the best way to go.
Lisp is not just a language; it is also a programming environment and runtime environment.
That being said, let’s go back to this book, Loving Common Lisp. I must admit that it did not meet my expectations entirely, especially because I was expecting something more polished and/or developed, but this may well be due to the iterative publishing process that we commonly encounter on Leanpub. I am sure that the author knows what he is talking about, and I have read his blog posts and books for years. It also doesn’t mean that the book is not well written. There are still some proof-reading lacking here and there but overall it looks good. It is just that this is neither an introductory textbook, nor a practical cookbook: It’s just in-between, which leaves us with the feeling that this book is not completely finished. On the plus side, you get a complete set of Lisp programs to run from GitHub (part of those examples depend on clml).
The author starts by comparing Lisp with Java or even C/C++, especially regarding automatic memory management (allocation and garbage collection), fair enough. This is a well-known fact that GC is of concern, as well as JIT compilation or number representation for modern PLs. Lisp is an old beast, but still actively used in prod system (e.g., Grammarly). I have always been able to compile old Lisp code, and I like the fact that updating my quicklisp packages resume to updating almost nothing on my OS. Because when a package provides the expected functionalities and is not going to evolve anymore, it is done for good. Remember this guy?
I use this package every day, and have been doing so for years. It just works. At least, it works for all my use cases. And if it breaks somehow, I fix it. However, it has become painfully clear to me that I don’t have time to fix problems I don’t have. It’s been years since I could keep pace with the issues and pull requests. Whenever I try, I keep getting feedback that my fix isn’t good enough by some standard I don’t particularly care about.
Freezing a working package is a feature, not a lack of interest.
The following is based on the notes I took in an Org-file while reading the first part of the book. The author does not explain the differences between defvar, defparameter, setf and setq although they are used a lot interchangeably at the beginning of the book. Treatment of lists is, however, pretty standard and well exposed (car and cdr, cons and append, last and nth, etc.). An interesting example regarding shared structure in list processing is provided, by the way:
(setq x '(0 0 0 0))
(setq y (list x x x x))
(setf (nth 2 (nth 1 y)) 'x)
x
y
(setq z '((0 0 0 0) (0 0 0 0) (0 0 0 0)))
(setf z (nth 2 (nth 1 z)) 'x)
z
Beyond lists, vectors and arrays (make-array, or vector and make-sequence) are more efficient data structure when the number of elements is large. Beware that CL for scientific computing cannot be fast, portable, and convenient all at the same time. Notice that an array can “contain” any values, and thus mixing integers with float is allowed by the language. Operations on string (concatenate, search, subseq and string-*) and the fine distinction between eq, eql, and equal are also covered. For strings, we should prefer string. Instead of nth, we use char to extract a given character in a string. Hash tables are to be preferred when lists (coupled with assoc) are long. Main functions are gethash, make-hash-table, and maphash. Updating values in a hash table is done using remhash or clrhash. Note that these functions can modify their arguments, much like setf or setq, but the latter are macros and not functions. Finally, recall that read-only objects are inherently thread safe.
Functional programming means that we avoid maintaining state inside of functions and treat data as immutable.
Lisp functions get covered as well, with lots of examples relying on defun, keywords (&aux, &optional, &key), let special operator for local bindings, lambda and funcall. A closure is a function that references an outer lexically scoped variable, which typically happens when functions are defined inside let forms (see p. 47). The dotimes and dolist macros are close to Stata forvalues and foreach instructions, while the do macro is more general (and has nothing to do with its R dplyr counterpart, of course).
The rest of the book describes some applications of web and network programming using CLOS classes and various packages (drakma, hunchentoot). I also enjoyed reading the chapter on querying database since I am doing a lot of this stuff these days. This is where the book starts to be really interesting because it then becomes a real practical cookbook, while the preceding chapters were more of a discussion of what Lisp has to offer and why the author likes it.