/serapeum

Utilities beyond Alexandria

Primary LanguageCommon LispMIT LicenseMIT

Table of Contents

Overview

Build Status

Serapeum is a conservative library of Common Lisp utilities. It is a supplement, not a competitor, to Alexandria. That means it is safe to do:

(defpackage ... (:use #:cl #:alexandria #:serapeum),

without package conflicts.

There may already be too many utility libraries for Common Lisp. Releasing another has become something to apologize for, not celebrate. But I would rather make my apologies than have to maintain copy-pasted versions of the same utilities across a dozen systems. And, though Serapeum is justified even if only I ever use it, the best way to ensure its quality is to write it as if for general use.

Serapeum is conservative: its goal is to fill in gaps in Common Lisp, not to redesign it. But it is less conservative than Alexandria. Alexandria limits itself to utilities with a Common Lisp pedigree. Serapeum casts a wider net: other dialects of Lisp, and other languages in the functional and array families, have been drafted.

Alexandria is self-contained. It exists in splendid isolation, without depending on, or even acknowledging, other libraries. Serapeum tries to be a good citizen of the Quicklisp era: whenever possible, it avoids duplicating functionality that can be had elsewhere.

Many of the utilities in Serapeum are original ideas; many ideas are borrowed from other languages, or from other Lispers. I try to give credit in the docstrings, but sometimes I have forgotten where I got an idea or a name. I regard missing credits as bugs: please report them.

Serapeum is intended to be portable, but it is principally tested where it is developed, on SBCL and Clozure CL. Patches and bug reports for other Lisps are always welcome, however, including implementation-specific optimizations.

Commentary

One goal of Serapeum is to have excellent documentation. A utility library is a fork of its language; it deserves documentation of the same quality as a language reference. If a utility is not worth documenting, it is not worth having.

The full function reference will be found here. (It is in a separate file in deference to documentation browsers, which often print the README as a preamble to their own function reference).

Most utilities in Serapeum stand alone, but there are some families that deserve separate introduction.

A note about strings

Beginning Lispers often ask about Lisp: “Where are the string utilities?” Since strings in Lisp are a kind of vector, which is in turn a kind of sequence, the right place to look is under the documentation for sequences (and vectors).

This is also true for Serapeum: there are many sequence utilities, and a few vector utilities, that work equally well on strings. But there are also many string-specific utilities.

Dividing sequences

All recent functional programming languages share a family of useful sequence-related functions with terrible names. All of them are called something like “split”, “divide”, or “group”, more or less at random.

For each function, we ensure:

  • It is efficient.
  • It returns like sequences for like (lists for lists, strings for strings, &c.).
  • It accommodates generic sequences (list and vector are not necessarily an exhaustive partition of sequence).
  • It has a distinctive name which does not use any of the weasel words “split,” “divide,” or “group.”

The function that returns runs of like elements in a sequence is called runs:

(runs '(head tail head head tail))
=> '((head) (tail) (head head) (tail))

The function that returns a sequence in batches of a certain maximum size is called batches:

(batches (iota 11) 2)
=> ((0 1) (2 3) (4 5) (6 7) (8 9) (10))

The function which groups the like elements of a sequence is called assort (because it returns a sequence assorted by some property).

(assort (iota 10)
        :key (lambda (n) (mod n 3)))
=> '((0 3 6 9) (1 4 7) (2 5 8))

The function that takes a predicate and a sequence, and returns two sequences – one sequence of the elements for which the function returns true, and one sequence of the elements for which it returns false – is (still) called partition.

(partition #'oddp (iota 10))
=> (1 3 5 7 9), (0 2 4 6 8)

The generalized version of partition, which takes a number of functions and returns the items that satisfy each condition, is called partitions.

(partitions (list #'primep #'evenp) (iota 10))
=> ((2 3 5 7) (0 4 6 8)), (1 9)

Items that do not belong in any partition are returned as a second value.

Serapeum simply re-exports split-sequence, which seems to be firmly rooted under its present name.

Binding values in the function namespace

fbind, fbind*, fbindrec, and fbindrec* bind values in the function namespace.

fbind and fbindrec are like flet and labels, respectively.

(fbind ((fn (lambda ....))) ...)
≡ (flet ((fn ...)) ...)

(fbindrec ((fn (lambda ...))) ...)
≡ (labels ((fn ...)) ...)

fbind* and fbindrec* have no exact parallels: they bind functions in sequence, so that each can be used in the construction (not just the definition, as with fbindrec) of the next.

(fbind* ((flip2 (lambda (fn)
                 (lambda (x y)
                   (funcall fn y x))))
         (xcons (flip2 #'cons)))
  (xcons 2 1))
=> (1 . 2)

These are non-trivial implementations. In many cases, fbind can produce code that is more efficient than using funcall, and even eliminate the overhead of higher-order functions like compose and curry. And fbindrec, which builds on fbind, further implements the optimizing transformation from Waddell et. al., Fixing Letrec. (Note that the macroexpansion of fbind may sometimes appear simplistic, using macroexpand; this is the happy case when we can prove that the function is never used as a value.)

For binding values in the function namespace at the top level, Serapeum provides defalias:

(defalias xcons (flip #'cons))

This is equivalent to (setf (fdefinition ...)), but also gives the function a compile-time definition so compilers don’t complain about its being undefined.

Internal definitions

The local form lets you use top-level definition forms to create local bindings. You can use defun instead of labels, defmacro instead of macrolet, def (which is Serapeum’s macro for top-level lexical bindings) instead of let, and so forth.

This has three advantages:

  1. Given a set of variable, function, and macro bindings, you can leave it to the compiler to figure out how to nest them. (This could be because you are porting a function from a language that uses flat bindings, or just because you are writing a very complicated function.)

  2. You can use macro-defining macros (macros that expand into defmacro), as well as macros that expand into defun forms, to create local bindings.

  3. You can (using local* or block-compile) easily switch to block compilation of top-level functions.

Serapeum’s implementation of internal definitions is as complete as it can be while remaining portable. That means full support for variables, functions, and symbol macros, but restricted support for macros.

Example: macros that work locally and globally

For example, memoizing local functions is usually clumsy; given local you can define a single defmemo form that supports both defun and labels.

(defmacro defmemo (name params &body body)
  (with-gensyms (memo-table args result result?)
    `(let ((,memo-table (make-hash-table :test 'equal)))
       (defun ,name (&rest ,args)
         (multiple-value-bind (,result ,result?)
             (gethash ,args ,memo-table)
           (if ,result?
               ,result
               (setf (gethash ,args ,memo-table)
                     (apply (lambda ,params
                              ,@body)
                              ,args))))))))

At the top level, this expands into an example of “let over defun” (gensyms elided for readability):

;; This source form
(defmemo fibonacci (n)
    (if (<= n 1)
        1
        (+ (fibonacci (- n 1))
           (fibonacci (- n 2)))))
           
;; Expands into...
(let ((memo-table (make-hash-table :test 'equal)))
  (defun fibonacci (&rest args)
    (multiple-value-bind (result result?)
        (gethash args memo-table)
      (if result? result
          (setf (gethash args memo-table)
                (apply (lambda (n)
                         (if (<= n 1)
                             1
                             (+ (fibonacci (- n 1))
                                (fibonacci (- n 2)))))
                       args))))))

But within a local form, it expands differently. This nearly identical source form:

(local
  (defmemo fibonacci (n)
    (if (<= n 1)
        1
        (+ (fibonacci (- n 1))
           (fibonacci (- n 2)))))

  (fibonacci 100))

Expands into this very different code (simplified for readability):

(let (fn)
  (labels ((fibonacci (&rest args)
             (apply fn args)))
    (let ((memo-table (make-hash-table :test 'equal)))
      (setf fn
            (named-lambda fibonacci (&rest args)
              (multiple-value-bind (result result?)
                  (gethash args memo-table)
                (if result? result
                    (setf (gethash args memo-table)
                          (apply
                           (lambda (n)
                             (if (<= n 1) 1
                                 (+ (fibonacci (- n 1))
                                    (fibonacci (- n 2)))))
                           args))))))
      
      (fibonacci 100))))

Example: block compiling

The macro local* is almost the same as local, except that it leaves the last form in the body intact. This is useful for obtaining block compilation in Lisps that don’t have a syntax for it.

During development, you define functions at the top level inside a progn.

 (progn
   (defun aux-fn-1 ...)
   (defun aux-fn-2 ...)
   (defun entry-point ...))

Then, when you decide you want block compilation, simply switch the progn to a local*:

 (local*
   (defun aux-fn-1 ...)
   (defun aux-fn-2 ...)
   (defun entry-point ...))

Which expands into something like:

(labels ((aux-fn-2 ...)
         (aux-fn-1 ...))
  (defun entry-point ...))

This has the slight disadvantage that calls to the entry points, including self calls, will still be compiled as global calls. If you want calls to the entry points to be compiled as local calls, you can use the block-compile macro instead.

Using block-compile, you can write:

(block-compile (:entry-points (entry-point))
  (defun aux-fn-1 ...)
  (defun aux-fn-2 ...)
  (defun entry-point ...))

And have it expand into something like:

(labels ((aux-fn-2 ...)
     (aux-fn-1 ...)
     (entry-point ...))
  (defalias entry-point #'entry-point))

Compile-time exhaustiveness checking

etypecase-of is just like etypecase, except that it takes an additional argument – the type to be matched against – and warns, at compile time, if the clauses in its body are not an exhaustive partition of that type.

ecase-of is a succint variant of etypecase with the same syntax as ecase.

typecase-of and case-of are etypecase-of and ecase-of, respectively, except that they expect, and enforce, the presence of an otherwise clause.

There are also continuable versions of these macros – ctypecase-of and ccase-of.

Example: enums

We may call a type defined using member an enumeration. Take an enumeration like this:

(deftype switch-state ()
  '(member :on :off :stuck :broken))

Now we can use ecase-of to take all the states of the switch into account.

(defun flick (switch)
  (ecase-of switch-state (state switch)
    (:on (switch-off switch))
    (:off (switch-on switch))))
=> Warning

(defun flick (switch)
  (ecase-of switch-state (state switch)
    (:on (switch-off switch))
    (:off (switch-on switch))
    ((:stuck :broken) (error "Sorry, can't flick ~a" switch))))
=> No warning

Even more usefully, we don’t have to worry about bugs caused by misspellings:

(defun flick (switch)
  (ecase-of switch-state (state switch)
    (:on (switch-off switch))
    (:offf (switch-on switch))          ;Gotcha!
    ((:stuck :broken) (error "Sorry, can't flick ~a" switch))))
=> Warning

Example: union types

(defun negative-integer? (n)
  (etypecase-of t n
    ((not integer) nil)
    ((integer * -1) t)
    ((integer 1 *) nil)))
=> Warning

(defun negative-integer? (n)
  (etypecase-of t n
    ((not integer) nil)
    ((integer * -1) t)
    ((integer 1 *) nil)
    ((integer 0) nil)))
=> No warning

CLOS

Serapeum includes some utilities for CLOS. These utilities do nothing earthshaking, but since the function reference does not include them, they should be documented somewhere.

Method combination: standard with context

Serapeum exports a method combination, serapeum:standard/context. You may recognize it as the wrapping-standard method combination due to Tim Bradshaw.

Generic functions defined with standard/context behave the same as ordinary generic functions, except that they allow an extra qualifier, :context. This extra qualifier works almost like :around, except instead of being run in most-specific-first order, like methods defined with :around, methods defined with :context are run in most-specific-last order. Furthermore, :context methods take priority over any other methods, including :around methods.

The big idea is that a class can use :context methods to make sure that any methods defined by subclasses – even :around methods – run in a certain dynamic context.

Metaclass: topmost-object-class

In most cases, when I write a metaclass, I want all of the classes defined using that metaclass to inherit from a specific class. Injecting a topmost class is not difficult to do, but it involves a certain amount of boilerplate.

To eliminate that boilerplate, Serapeum exports a metaclass, topmost-object-class, to use as a base class for your metaclasses. When you define a metaclass, all you have to do to ensure that classes defined using your metaclass inherit from a specific class is to supply the name of the class to inherit from in the definition of the metaclass. This is much better demonstrated than explained:

;;; The class to inherit from.
(defclass my-topmost-object ()
  ())

;;; The metaclass.
(defclass my-metaclass (serapeum:topmost-object-class)
  ()
  (:default-initargs
   :topmost-class 'my-topmost-object))

(defclass my-class ()
  ()
  (:metaclass my-metaclass))

(typep (make-instance 'my-class) 'my-topmost-object) => t

Note that, since the topmost object is usually a standard class, there is a validate-superclass method which allows an instance of topmost-object-class to inherit from a standard class.

Function reference

The complete reference is in a separate file.

(Note that the reference is generated from docstrings, and should not be edited by hand.)