11 Structures

The Common Lisp structure mechanism provides a general way to define data types similar to C’s struct types. A structure is a Lisp object containing some number of slots, each of which can hold any Lisp data object. Functions are provided for accessing and setting the slots, creating or copying structure objects, and recognizing objects of a particular structure type.

In true Common Lisp, each structure type is a new type distinct from all existing Lisp types. Since the underlying Emacs Lisp system provides no way to create new distinct types, this package implements structures as vectors (or lists upon request) with a special “tag” symbol to identify them.

Macro: cl-defstruct name slots…

The cl-defstruct form defines a new structure type called name, with the specified slots. (The slots may begin with a string which documents the structure type.) In the simplest case, name and each of the slots are symbols. For example,

(cl-defstruct person first-name age sex)

defines a struct type called person that contains three slots. Given a person object p, you can access those slots by calling (person-first-name p), (person-age p), and (person-sex p). You can also change these slots by using setf on any of these place forms, for example:

(cl-incf (person-age birthday-boy))

You can create a new person by calling make-person, which takes keyword arguments :first-name, :age, and :sex to specify the initial values of these slots in the new object. (Omitting any of these arguments leaves the corresponding slot “undefined”, according to the Common Lisp standard; in Emacs Lisp, such uninitialized slots are filled with nil.)

Given a person, (copy-person p) makes a new object of the same type whose slots are eq to those of p.

Given any Lisp object x, (person-p x) returns true if x is a person, and false otherwise.

Accessors like person-first-name normally check their arguments (effectively using person-p) and signal an error if the argument is the wrong type. This check is affected by (optimize (safety …)) declarations. Safety level 1, the default, uses a somewhat optimized check that will detect all incorrect arguments, but may use an uninformative error message (e.g., “expected a vector” instead of “expected a person”). Safety level 0 omits all checks except as provided by the underlying aref call; safety levels 2 and 3 do rigorous checking that will always print a descriptive error message for incorrect inputs. See Declarations.

(setq dave (make-person :first-name "Dave" :sex 'male))
     ⇒ [cl-struct-person "Dave" nil male]
(setq other (copy-person dave))
     ⇒ [cl-struct-person "Dave" nil male]
(eq dave other)
     ⇒ nil
(eq (person-first-name dave) (person-first-name other))
     ⇒ t
(person-p dave)
     ⇒ t
(person-p [1 2 3 4])
     ⇒ nil
(person-p "Bogus")
     ⇒ nil
(person-p '[cl-struct-person counterfeit person object])
     ⇒ t

In general, name is either a name symbol or a list of a name symbol followed by any number of structure options; each slot is either a slot symbol or a list of the form ‘(slot-name default-value slot-options…)’. The default-value is a Lisp form that is evaluated any time an instance of the structure type is created without specifying that slot’s value.

(cl-defstruct person
     (first-name nil :read-only t)
     age
     (sex 'unknown))

slot-options is a list of keyword-value pairs, where the following keywords can be used:

:read-only

A non-nil value means the slot should not be setf-able; the slot’s value is determined when the object is created and does not change afterward.

:type

The expected type of the values held in this slot.

:documentation

A documentation string describing the slot.

Other slot options are currently ignored.

For obscure historical reasons, structure options take a different form than slot options. A structure option is either a keyword symbol, or a list beginning with a keyword symbol possibly followed by arguments. (By contrast, slot options are key-value pairs not enclosed in lists.)

(cl-defstruct (person (:constructor create-person)
                      (:type list)
                      :named)
     first-name age sex)

The following structure options are recognized.

:conc-name

The argument is a symbol whose print name is used as the prefix for the names of slot accessor functions. The default is the name of the struct type followed by a hyphen. The option (:conc-name p-) would change this prefix to p-. Specifying nil as an argument means no prefix, so that the slot names themselves are used to name the accessor functions.

:constructor

In the simple case, this option takes one argument which is an alternate name to use for the constructor function. The default is make-name, e.g., make-person. The above example changes this to create-person. Specifying nil as an argument means that no standard constructor should be generated at all.

In the full form of this option, the constructor name is followed by an arbitrary argument list. See Program Structure, for a description of the format of Common Lisp argument lists. All options, such as &rest and &key, are supported. The argument names should match the slot names; each slot is initialized from the corresponding argument. Slots whose names do not appear in the argument list are initialized based on the default-value in their slot descriptor. Also, &optional and &key arguments that don’t specify defaults take their defaults from the slot descriptor. It is valid to include arguments that don’t correspond to slot names; these are useful if they are referred to in the defaults for optional, keyword, or &aux arguments that do correspond to slots.

You can specify any number of full-format :constructor options on a structure. The default constructor is still generated as well unless you disable it with a simple-format :constructor option.

(cl-defstruct
    (person
     (:constructor nil)   ; no default constructor
     (:constructor new-person
                   (first-name sex &optional (age 0)))
     (:constructor new-hound (&key (first-name "Rover")
                                   (dog-years 0)
                              &aux (age (* 7 dog-years))
                                   (sex 'canine))))
    first-name age sex)

The first constructor here takes its arguments positionally rather than by keyword. (In official Common Lisp terminology, constructors that work By Order of Arguments instead of by keyword are called “BOA constructors”. No, I’m not making this up.) For example, (new-person "Jane" 'female) generates a person whose slots are "Jane", 0, and female, respectively.

The second constructor takes two keyword arguments, :name, which initializes the name slot and defaults to "Rover", and :dog-years, which does not itself correspond to a slot but which is used to initialize the age slot. The sex slot is forced to the symbol canine with no syntax for overriding it.

:copier

The argument is an alternate name for the copier function for this type. The default is copy-name. nil means not to generate a copier function. (In this implementation, all copier functions are simply synonyms for copy-sequence.)

:predicate

The argument is an alternate name for the predicate that recognizes objects of this type. The default is name-p. nil means not to generate a predicate function. (If the :type option is used without the :named option, no predicate is ever generated.)

In true Common Lisp, typep is always able to recognize a structure object even if :predicate was used. In this package, cl-typep simply looks for a function called typename-p, so it will work for structure types only if they used the default predicate name.

:include

This option implements a very limited form of C++-style inheritance. The argument is the name of another structure type previously created with cl-defstruct. The effect is to cause the new structure type to inherit all of the included structure’s slots (plus, of course, any new slots described by this struct’s slot descriptors). The new structure is considered a “specialization” of the included one. In fact, the predicate and slot accessors for the included type will also accept objects of the new type.

If there are extra arguments to the :include option after the included-structure name, these options are treated as replacement slot descriptors for slots in the included structure, possibly with modified default values. Borrowing an example from Steele:

(cl-defstruct person first-name (age 0) sex)
        ⇒ person
(cl-defstruct (astronaut (:include person (age 45)))
     helmet-size
     (favorite-beverage 'tang))
        ⇒ astronaut

(setq joe (make-person :first-name "Joe"))
     ⇒ [cl-struct-person "Joe" 0 nil]
(setq buzz (make-astronaut :first-name "Buzz"))
     ⇒ [cl-struct-astronaut "Buzz" 45 nil nil tang]

(list (person-p joe) (person-p buzz))
     ⇒ (t t)
(list (astronaut-p joe) (astronaut-p buzz))
     ⇒ (nil t)

(person-first-name buzz)
     ⇒ "Buzz"
(astronaut-first-name joe)
     ⇒ error: "astronaut-first-name accessing a non-astronaut"

Thus, if astronaut is a specialization of person, then every astronaut is also a person (but not the other way around). Every astronaut includes all the slots of a person, plus extra slots that are specific to astronauts. Operations that work on people (like person-first-name) work on astronauts just like other people.

:noinline

If this option is present, this structure’s functions will not be inlined, even functions that normally would.

:print-function

In full Common Lisp, this option allows you to specify a function that is called to print an instance of the structure type. The Emacs Lisp system offers no hooks into the Lisp printer which would allow for such a feature, so this package simply ignores :print-function.

:type

The argument should be one of the symbols vector or list. This tells which underlying Lisp data type should be used to implement the new structure type. Records are used by default, but (:type vector) will cause structure objects to be stored as vectors and (:type list) lists instead.

The record and vector representations for structure objects have the advantage that all structure slots can be accessed quickly, although creating them are a bit slower in Emacs Lisp. Lists are easier to create, but take a relatively long time accessing the later slots.

:named

This option, which takes no arguments, causes a characteristic “tag” symbol to be stored at the front of the structure object. Using :type without also using :named will result in a structure type stored as plain vectors or lists with no identifying features.

The default, if you don’t specify :type explicitly, is to use records, which are always tagged. Therefore, :named is only useful in conjunction with :type.

(cl-defstruct (person1) first-name age sex)
(cl-defstruct (person2 (:type list) :named) first-name age sex)
(cl-defstruct (person3 (:type list)) first-name age sex)
(cl-defstruct (person4 (:type vector)) first-name age sex)

(setq p1 (make-person1))
     ⇒ #s(person1 nil nil nil)
(setq p2 (make-person2))
     ⇒ (person2 nil nil nil)
(setq p3 (make-person3))
     ⇒ (nil nil nil)
(setq p4 (make-person4))
     ⇒ [nil nil nil]

(person1-p p1)
     ⇒ t
(person2-p p2)
     ⇒ t
(person3-p p3)
     ⇒ error: function person3-p undefined

Since unnamed structures don’t have tags, cl-defstruct is not able to make a useful predicate for recognizing them. Also, accessors like person3-first-name will be generated but they will not be able to do any type checking. The person3-first-name function, for example, will simply be a synonym for car in this case. By contrast, person2-first-name is able to verify that its argument is indeed a person2 object before proceeding.

:initial-offset

The argument must be a nonnegative integer. It specifies a number of slots to be left “empty” at the front of the structure. If the structure is named, the tag appears at the specified position in the list or vector; otherwise, the first slot appears at that position. Earlier positions are filled with nil by the constructors and ignored otherwise. If the type :includes another type, then :initial-offset specifies a number of slots to be skipped between the last slot of the included type and the first new slot.

Except as noted, the cl-defstruct facility of this package is entirely compatible with that of Common Lisp.

The cl-defstruct package also provides a few structure introspection functions.

Function: cl-struct-sequence-type struct-type

This function returns the underlying data structure for struct-type, which is a symbol. It returns record, vector or list, or nil if struct-type is not actually a structure.

Function: cl-struct-slot-info struct-type

This function returns a list of slot descriptors for structure struct-type. Each entry in the list is (name . opts), where name is the name of the slot and opts is the list of slot options given to defstruct. Dummy entries represent the slots used for the struct name and that are skipped to implement :initial-offset.

Function: cl-struct-slot-offset struct-type slot-name

Return the offset of slot slot-name in struct-type. The returned zero-based slot index is relative to the start of the structure data type and is adjusted for any structure name and :initial-offset slots. Signal error if struct struct-type does not contain slot-name.

Function: cl-struct-slot-value struct-type slot-name inst

Return the value of slot slot-name in inst of struct-type. struct and slot-name are symbols. inst is a structure instance. This routine is also a setf place. Can signal the same errors as cl-struct-slot-offset.