General Programming Style Rules

• Write short functions, where each function provides a single, well-defined operation. Small functions are easier to read, write, test, debug, and understand.

• Use descriptive variable and function names. If it isn't clear from the name of a function or variable what its purpose is, document it with a documentation string and a comment. In fact, even if the purpose is evident from the name, it is still worth documenting your code.

• Don't write Pascal (or C) code in Lisp. Use the appropriate predefined functions -- look in the index to CLtL2, or use the APROPOS and DESCRIBE functions. Don't put a close parenthesis on a line by itself -- this can really irritate programmers who grew up on Lisp. Lisp-oriented text editors include tools for ensuring balanced parentheses and for moving across pairs of balanced parentheses. You don't need to stick comments on close parentheses to mark which expression they close.

• Use proper indentation -- you should be able to understand the structure of your definitions without noticing the parentheses. In general, the way one indents a form is controlled by the first symbol of the form. In DEFUNs, for example, one puts the symbol DEFUN, the function name, and the argument list all on the same line. If the argument list is too long, one can break it at one of the lambda keywords. Following the argument list, one inserts a carriage return and lists the expressions in the body of the definition, with each form starting on its own line indented three spaces relative to the open parenthesis of the parent (in this case the DEFUN). This general style -- of putting all the significant elements of a form on a single line, followed by a carriage return and the indented body -- holds for many Lisp constructs. There are, of course, variations, such as keeping the first clause on the same line as the COND or CASE symbol, and the rules are relaxed in different ways to keep line lengths to a manageable size. If you find yourself having trouble fitting everything in even with line breaking and relaxing the rules, either your function names are too long or your code isn't very modular. You should perceive this as a signal that you need to break up your big definitions into smaller chunks, each with a clearly defined purpose, and possibly replace long function names with concise but apt shorter ones.

• Use whitespace appropriately. Use whitespace to separate semantically distinct code segments, but don't use too much whitespace. For example,

      GOOD: 
         (defun foo (x y)
           (let ((z (+ x y 10)))
             (* z z)))
  
      BAD: 
         (defun foo(x y)(let((z(+ x y 10)))(* z z)))
  
         (defun foo ( x  y )
           (let ( ( z (+ x y 10) ) )
              ( * z z )
              )
           )    
  

  Although the Lisp reader and compiler don't care which you use, most experienced Lisp programmers find the first example much easier to read than the last two.

• Don't use line lengths greater than 80 characters. People who write code using Zmacs on Symbolics Lisp Machines are notoriously guilty of violating this rule, because the CPT6 font allows one to squeeze a tremendous amount of code on the display, especially if one spreads the code out horizontally. This makes it more difficult to read when printed out or read on an 80x24 xterm window. In fact, use a line length of 72 characters because it leaves a strip of white space at the edge of the window.

Often Misused Operators

• EVAL. Novices almost always misuse EVAL. When experts use EVAL, they often would be better off using APPLY, FUNCALL, or SYMBOL-VALUE. Use of EVAL when defining a macro should set off a warning bell -- macro definitions are already evaluated during expansion. See also the answer to question 3-12. The general rule of thumb about EVAL is: if you think you need to use EVAL, you're probably wrong.

• PROGV. PROGV binds dynamic variables and is often misused in conjunction with EVAL, which uses the dynamic environment. In general, avoid unnecessary use of special variables. PROGV is mainly for writing interpreters for languages embedded in Lisp. If you want to bind a list of values to a list of lexical variables, use

        (MULTIPLE-VALUE-BIND (..) (VALUES-LIST ..) ..)
  

  or

        (MULTIPLE-VALUE-SETQ (..) (VALUES-LIST ..))
  

  instead. Most decent compilers can optimize this expression. However, use of this idiom is not to be encouraged unless absolutely necessary.

• CATCH and THROW. Often a named BLOCK and RETURN-FROM are more appropriate. Use UNWIND-PROTECT when necessary.

• Destructive operations, such as NCONC, SORT, DELETE, RPLACA, and RPLACD, should be used carefully and sparingly. In general, trust the garbage collector: allocate new data structures when you need them.

Readability

• Don't use any C{A,D}R functions with more than two letters between the C and the R. When nested, they become hard to read. If you have complex data structures, you are often better off describing them with a DEFSTRUCT, even if the type is LIST. The data abstraction afforded by DEFSTRUCT makes the code much more readable and its purpose clearer. If you must use C{A,D}R, try to use DESTRUCTURING-BIND instead, or at least SECOND, THIRD, NTH, NTHCDR, etc.

• Use COND instead of IF and PROGN. In general, don't use PROGN if there is a way to write the code within an implicit PROGN. For example,

       (IF (FOO X)
           (PROGN (PRINT "hi there") 23)
           34)
  

  should be written using COND instead.

• Never use a 2-argument IF or a 3-argument IF with a second argument of NIL unless you want to emphasize the return value; use WHEN and UNLESS instead. You will want to emphasize the return value when the IF clause is embedded within a SETQ, such as (SETQ X (IF (EQ Y Z) 2 NIL)). If the second argument to IF is the same as the first, use OR instead: (OR P Q) rather than (IF P P Q). Use UNLESS instead of (WHEN (NOT ..) ..) but not instead of (WHEN (NULL ..) ..).

• Use COND instead of nested IF statements. Be sure to check for unreachable cases, and eliminate those cond-clauses.

• Use backquote, rather than explicit calls to LIST, CONS, and APPEND, whenever writing a form which produces a Lisp form, but not as a general substitute for LIST, CONS and APPEND. LIST, CONS and APPEND usually allocate new storage, but lists produced by backquote may involve destructive modification (e.g., ,.).

• Make the names of special (global) variables begin and end with an asterisk (*): (DEFVAR *GLOBAL-VARIABLE*) Some programmers will mark the beginning and end of an internal global variable with a percent (%) or a period (.). Make the names of constants begin and end with a plus (+): (DEFCONSTANT +E+ 2.7182818) This helps distinguish them from lexical variables. Some people prefer to use macros to define constants, since this avoids the problem of accidentally trying to bind a symbol declared with defconstant.

• If your program is built upon an underlying substrate which is implementation-dependent, consider naming those functions and macros in a way that visually identifies them, either by placing them in their own package, or prepending a character like a %, ., or ! to the function name. Note that many programmers use the $ as a macro character for slot access, so it should be avoided unless you're using it for that purpose.

• Don't use property lists. Instead, use an explicit hash table. This helps avoid problems caused by the symbol being in the wrong package, accidental reuse of property keys from other programs, and allows you to customize the structure of the table.

• Use the most specific construct that does the job. This lets readers of the code see what you intended when writing the code. For example, don't use SETF if SETQ will do (e.g., for lexical variables). Using SETQ will tell readers of your code that you aren't doing anything fancy. Likewise, don't use EQUAL where EQ will do. Use the most specific predicate to test your conditions.

• If you intend for a function to be a predicate, have it return T for true, not just non-NIL. If there is nothing worth returning from a function, returning T is conventional. But if a function is intended to be more than just a predicate, it is better to return a useful value. (For example, this is one of the differences between MEMBER and FIND.)

• When NIL is used as an empty list, use () in your code. When NIL is used as a boolean, use NIL. Similarly, use NULL to test for an empty list, NOT to test a logical value. Use ENDP to test for the end of a list, not NULL.

• Don't use the &AUX lambda-list keyword. It is always clearer to define local variables using LET or LET*.

• When using RETURN and RETURN-FROM to exit from a block, don't use (VALUES ..) when returning only one value, except if you are using it to suppress extra multiple values from the first argument.

• If you want a function to return no values (i.e., equivalent to VOID in C), use (VALUES) to return zero values. This signals to the reader that the function is used mainly for side-effects.

• (VALUES (VALUES 1 2 3)) returns only the first value, 1. You can use (VALUES (some-multiple-value-function ..)) to suppress the extra multiple values from the function. Use MULTIPLE-VALUE-PROG1 instead of PROG1 when the multiple values are significant.

• When using MULTIPLE-VALUE-BIND and DESTRUCTURING-BIND, don't rely on the fact that NIL is used when values are missing. This is an error in some implementations of DESTRUCTURING-BIND. Instead, make sure that your function always returns the proper number of values.

• Type the name of external symbols, functions, and variables from the COMMON-LISP package in uppercase. This will allow your code to work properly in a case-sensitive version of Common Lisp, since the print-names of symbols in the COMMON-LISP package are uppercase internally. (However, not everybody feels that being nice to case-sensitive Lisps is a requirement, so this isn't an absolute style rule, just a suggestion.)

Lisp Idioms

• MAPCAN is used with a function to return a variable number of items to be included in an output list. When the function returns zero or one items, the function serves as a filter. For example, (mapcan #'(lambda (x) (when (and (numberp x) (evenp x)) (list x))) '(1 2 3 4 x 5 y 6 z 7))

Documentation

• Comment your code. Use three semicolons in the left margin before the definition for major explanations. Use two semicolons that float with the code to explain the routine that follows. Two semicolons may also be used to explain the following line when the comment is too long for the single semicolon treatment. Use a single semicolon to the right of the code to explain a particular line with a short comment. The number of semicolons used roughly corresponds with the length of the comment. Put at least one blank line before and after top-level expressions.

• Include documentation strings in your code. This lets users get help while running your program without having to resort to the source code or printed documentation.

Macros

• Never use a macro instead of a function for efficiency reasons. Declaim the function as inline -- for example,

       (DECLAIM (INLINE ..))
  

  This is not a magic bullet -- be forewarned that inline expansions can often increase the code size dramatically. INLINE should be used only for short functions where the tradeoff is likely to be worthwhile: inner loops, types that the compiler might do something smart with, and so on.

• When defining a macro that provides an implicit PROGN, use the &BODY lambda-list keyword instead of &REST.

• Use gensyms for bindings within a macro, unless the macro lets the user explicitly specify the variable. For example:

      (defmacro foo ((iter-var list) body-form
                     &body body)
         (let ((result (gensym "RESULT")))
           `(let ((,result nil))
              (dolist (,iter-var ,list ,result)
                (setq ,result ,body-form)
                (when ,result
                  ,@body)))))        
  

  This avoids errors caused by collisions during macro expansion between variable names used in the macro definition and in the supplied body.

• Use a DO- prefix in the name of a macro that does some kind of iteration, WITH- when the macro establishes bindings, and DEFINE- or DEF- when the macro creates some definitions. Don't use the prefix MAP- in macro names, only in function names.

• Don't create a new iteration macro when an existing function or macro will do.

• Don't define a macro where a function definition will work just as well -- remember, you can FUNCALL or MAPCAR a function but not a macro.

• The LOOP and SERIES macros generate efficient code. If you're writing a new iteration macro, consider learning to use one of them instead.

File Modularization

• If your program involves macros that are used in more than one file, it is generally a good idea to put such macros in a separate file that gets loaded before the other files. The same things applies to primitive functions. If a macro is complicated, the code that defines the macro should be put into a file by itself. In general, if a set of definitions form a cohesive and "independent" whole, they should be put in a file by themselves, and maybe even in their own package. It isn't unusual for a large Lisp program to have files named "site-dependent-code", "primitives.lisp", and "macros.lisp". If a file contains primarily macros, put "-macros" in the name of the file.

Stylistic Preferences

• Use (SETF (CAR ..) ..) and (SETF (CDR ..) ..) in preference to RPLACA and RPLACD. Likewise (SETF (GET ..) ..) instead of PUT.

• Use INCF, DECF, PUSH and POP instead instead of the corresponding SETF forms.

• Many programmers religiously avoid using CATCH, THROW, BLOCK, PROG, GO and TAGBODY. Tags and go-forms should only be necessary to create extremely unusual and complicated iteration constructs. In almost every circumstance, a ready-made iteration construct or recursive implementation is more appropriate.

• Don't use LET* where LET will do. Don't use LABELS where FLET will do. Don't use DO* where DO will do.

• Don't use DO where DOTIMES or DOLIST will do.

• If you like using MAPCAR instead of DO/DOLIST, use MAPC when no result is needed -- it's more efficient, since it doesn't cons up a list. If a single cumulative value is required, use REDUCE. If you are seeking a particular element, use FIND, POSITION, or MEMBER.

• If using REMOVE and DELETE to filter a sequence, don't use the :test-not keyword or the REMOVE-IF-NOT or DELETE-IF-NOT functions. Use COMPLEMENT to complement the predicate and the REMOVE-IF or DELETE-IF functions instead.

• Use complex numbers to represent points in a plane.

• Don't use lists where vectors are more appropriate. Accessing the nth element of a vector is faster than finding the nth element of a list, since the latter requires pointer chasing while the former requires simple addition. Vectors also take up less space than lists. Use adjustable vectors with fill-pointers to implement a stack, instead of a list -- using a list continually conses and then throws away the conses.

• When adding an entry to an association list, use ACONS, not two calls to CONS. This makes it clear that you're using an alist.

• If your association list has more than about 10 entries in it, consider using a hash table. Hash tables are often more efficient. (See also [2-2].)

• When you don't need the full power of CLOS, consider using structures instead. They are often faster, take up less space, and easier to use.

• Use PRINT-UNREADABLE-OBJECT when writing a print-function.

• Use WITH-OPEN-FILE instead of OPEN and CLOSE.

• When a HANDLER-CASE clause is executed, the stack has already unwound, so dynamic bindings that existed when the error occured may no longer exist when the handler is run. Use HANDLER-BIND if you need this.

• When using CASE and TYPECASE forms, if you intend for the form to return NIL when all cases fail, include an explicit OTHERWISE clause. If it would be an error to return NIL when all cases fail, use ECASE, CCASE, ETYPECASE or CTYPECASE instead.

• Use local variables in preference to global variables whenever possible. Do not use global variables in lieu of parameter passing. Global variables can be used in the following circumstances:

  • When one function needs to affect the operation of another, but the second function isn't called by the first. (For example, *load-pathname* and *break-on-warnings*.)

  • When a called function needs to affect the current or future operation of the caller, but it doesn't make sense to accomplish this by returning multiple values.

  • To provide hooks into the mechanisms of the program. (For example, *evalhook*, *, /, and +.)

  • Parameters which, when their value is changed, represent a major change to the program. (For example, *print-level* and *print-readably*.)

  • For state that persists between invocations of the program. Also, for state which is used by more than one major program. (For example, *package*, *readtable*, *gensym-counter*.)

  • To provide convenient information to the user. (For example, *version* and *features*.)

  • To provide customizable defaults. (For example, *default-pathname-defaults*.)

  • When a value affects major portions of a program, and passing this value around would be extremely awkward. (The example here is output and input streams for a program. Even when the program passes the stream around as an argument, if you want to redirect all output from the program to a different stream, it is much easier to just rebind the global variable.)

• Beginning students, especially ones accustomed to programming in C, Pascal, or Fortran, tend to use global variables to hold or pass information in their programs. This style is considered ugly by experienced Lisp programmers. Although assignment statements can't always be avoided in production code, good programmers take advantage of Lisp's functional programming style before resorting to SETF and SETQ. For example, they will nest function calls instead of using a temporary variable and use the stack to pass multiple values. When first learning to program in Lisp, try to avoid SETF/SETQ and their cousins as much as possible. And if a temporary variable is necessary, bind it to its first value in a LET statement, instead of letting it become a global variable by default. (If you see lots of compiler warnings about declaring variables to be special, you're probably making this mistake. If you intend a variable to be global, it should be defined with a DEFVAR or DEFPARAMETER statement, not left to the compiler to fix.)

Correctness and Efficiency Issues

• In CLtL2, IN-PACKAGE does not evaluate its argument. Use defpackage to define a package and declare the external (exported) symbols from the package.

• The ARRAY-TOTAL-SIZE-LIMIT may be as small as 1024, and the CALL-ARGUMENTS-LIMIT may be as small as 50.

• Novices often mistakenly quote the conditions of a CASE form. For example, (case x ('a 3) ..) is incorrect. It would return 3 if x were the symbol QUOTE. Use (case x (a 3) ..) instead.

• Avoid using APPLY to flatten lists. Although

      (apply #'append list-of-lists)
  

  may look like a call with only two arguments, it becomes a function call to APPEND, with the LIST-OF-LISTS spread into actual arguments. As a result it will have as many arguments as there are elements in LIST-OF-LISTS, and hence may run into problems with the CALL-ARGUMENTS-LIMIT. Use REDUCE or MAPCAN instead:

       (reduce #'append list-of-lists :from-end t)
       (mapcan #'copy-list list-of-lists)
  

  The second will often be more efficient (see note below about choosing the right algorithm). Beware of calls like (apply f (mapcar ..)).

• NTH must cdr down the list to reach the elements you are interested in. If you don't need the structural flexibility of lists, try using vectors and the ELT function instead.

• CASE statements can be vectorized if the keys are consecutive numbers. Such CASE statements can still have OTHERWISE clauses. To take advantage of this without losing readability, use #. with symbolic constants:

       (eval-when (:compile-toplevel :load-toplevel
                   :execute)
          (defconstant RED 1)
          (defconstant GREEN 2)
          (defconstant BLUE 3))
  
       (case color
          (#.RED   ...)
          (#.GREEN ...)
          (#.BLUE  ...)
          ...)
  

• Don't use quoted constants where you might later destructively modify them. For example, instead of writing '(c d) in

      (defun foo ()
        (let ((var '(c d)))
  	 ...))
  

  write (list 'c 'd) instead. Using a quote here can lead to unexpected results later. If you later destructively modify the value of var, this is self-modifying code! Some Lisp compilers will complain about this, since they like to make constants read-only. Modifying constants has undefined results in ANSI CL. See also the answer to question [3-13].

  Similarly, beware of shared list structure arising from the use of backquote. Any sublist in a backquoted expression that doesn't contain any commas can share with the original source structure.

• Don't proclaim unsafe optimizations, such as

       (proclaim '(optimize (safety 0) (speed 3)
                            (space 1))) 
  

  since this yields a global effect. Instead, add the optimizations as local declarations to small pieces of well-tested, performance-critical code:

      (defun well-tested-function ()
         (declare (optimize (safety 0) (speed 3)
                  (space 1)))
         ...)
  

  Such optimizations can remove run-time type-checking; type-checking is necessary unless you've very carefully checked your code and added all the appropriate type declarations.

• Some programmers feel that you shouldn't add declarations to code until it is fully debugged, because incorrect declarations can be an annoying source of errors. They recommend using CHECK-TYPE liberally instead while you are developing the code. On the other hand, if you add declarations to tell the compiler what you think your code is doing, the compiler can then tell you when your assumptions are incorrect. Declarations also make it easier for another programmer to read your code.

• Declaring the type of variables to be FIXNUM does not necessarily mean that the results of arithmetic involving the fixnums will be a fixnum; it could be a BIGNUM. For example,

      (declare (type fixnum x y))
      (setq z (+ (* x x) (* y y)))
  

  could result in z being a BIGNUM. If you know the limits of your numbers, use a declaration like

      (declare (type (integer 0 100) x y))
  

  instead, since most compilers can then do the appropriate type inference, leading to much faster code.

• Don't change the compiler optimization with an OPTIMIZE proclamation or declaration until the code is fully debugged and profiled. When first writing code you should say (declare (optimize (safety 3))) regardless of the speed setting.

• Depending on the optimization level of the compiler, type declarations are interpreted either as (1) a guarantee from you that the variable is always bound to values of that type, or (2) a desire that the compiler check that the variable is always bound to values of that type. Use CHECK-TYPE if (2) is your intention.

• If you get warnings about unused variables, add IGNORE declarations if appropriate or fix the problem. Letting such warnings stand is a sloppy coding practice.

• choose the right algorithm. For example, consider seven possible implementations of COPY-LIST:

      (defun copy-list (list)
        (let ((result nil))
          (dolist (item list result)
            (setf result (append result (list item))))))
  
      (defun copy-list (list)
        (let ((result nil))
          (dolist (item list (nreverse result))
            (push item result))))
  
      (defun copy-list (list)
        (mapcar #'identity list))
  
      (defun copy-list (list)
        (let ((result (make-list (length list))))
          (do ((original list (cdr original))
               (new result (cdr new)))
              ((null original) result)
            (setf (car new) (car original)))))
  
      (defun copy-list (list)
         (when list
            (let* ((result (list (car list)))
                   (tail-ptr result))
              (dolist (item (cdr list) result)
                (setf (cdr tail-ptr) (list item))
                (setf tail-ptr (cdr tail-ptr))))))
          
      (defun copy-list (list)
         (loop for item in list collect item))
  
      (defun copy-list (list)
         (if (consp list) 
             (cons (car list)
                   (copy-list (cdr list)))
             list))
  

  The first uses APPEND to tack the elements onto the end of the list. Since APPEND must traverse the entire partial list at each step, this yields a quadratic running time for the algorithm. The second implementation improves on this by iterating down the list twice; once to build up the list in reverse order, and the second time to reverse it. The efficiency of the third depends on the Lisp implementation, but it is usually similar to the second, as is the fourth. The fifth algorithm, however, iterates down the list only once. It avoids the extra work by keeping a pointer (reference) to the last cons of the list and RPLACDing onto the end of that. Use of the fifth algorithm may yield a speedup. Note that this contradicts the earlier dictum to avoid destructive functions. To make more efficient code one might selectively introduce destructive operations in critical sections of code. Nevertheless, the fifth implementation may be less efficient in Lisps with cdr-coding, since it is more expensive to RPLACD cdr-coded lists. Depending on the implementation of nreverse, however, the fifth and second implementations may be doing the same amount of work. The sixth example uses the Loop macro, which usually expands into code similar to the third. The seventh example copies dotted lists, and runs in linear time, but isn't tail-recursive.

  There is a long-running discussion of whether pushing items onto a list and then applying NREVERSE to the result is faster or slower than the alternatives. According to Richard C. Waters (Lisp Pointers VI(4):27-34, October-December 1993), the NREVERSE strategy is slightly faster in most Lisp implementations. But the speed difference either way isn't much, so he argues that one should pursue the option that yields the clearest and simplest code, namely using NREVERSE. Here's code for a possible implementation of NREVERSE. As is evident, most of the alternatives to using NREVERSE involve essentially the same code, just reorganized.

    (defun nreverse (list)
       ;; REVERSED is the partially reversed list, 
       ;; CURRENT is the current cons cell, which will be
       ;;   reused, and
       ;; REMAINING are the cons cells which have not yet
       ;;   been reversed.
       (do* ((reversed nil)		
  	   (current list remaining)
  	   (remaining (cdr current) (cdr current)))
  	 ((null current)
  	  reversed)
         ;; Reuse the cons cell at the head of the list:
         ;;    reversed := ((car remaining) . reversed)
         (setf (cdr current) reversed)
         (setf reversed current)))
  

• use type declarations liberally in time-critical code, but only if you are a seasoned Lisp programmer. Appropriate type declarations help the compiler generate more specific and optimized code. It also lets the reader know what assumptions were made. For example, if you only use fixnum arithmetic, adding declarations can lead to a significant speedup. If you are a novice Lisp programmer, you should use type declarations sparingly, as there may be no checking to see if the declarations are correct, and optimized code can be harder to debug. Wrong declarations can lead to errors in otherwise correct code, and can limit the reuse of code in other contexts. Depending on the Lisp compiler, it may also be necessary to declare the type of results using THE, since some compilers don't deduce the result type from the inputs.

• check the code produced by the compiler by using the disassemble function