I am using a function from an external library returning a vector of four numbers and I want to access these values directly like it would be possible with destructuring-bind. See this pointless example:
(defun a-vector ()
(vector 1 2 3 4))
(defun a-list ()
(list 1 2 3 4))
(destructuring-bind (a b c d)
(a-list)
(format t "~D ~D ~D ~D~%" a b c d))
(destructuring-bind (a b c d)
(coerce (a-vector) 'list)
(format t "~D ~D ~D ~D~%" a b c d))
If I coerce the vector into a list it is possible and as performance isn't a problem here, it is maybe fine. But I was wondering if there is a more simple way?
You can bind variables to each cell as follows:
(defmacro with-aref ((&rest indices) array &body body)
(let ((a (gensym)))
`(let ((,a ,array))
(symbol-macrolet
,(loop
for n from 0
for i in indices
collect (list i `(aref ,a ,n)))
,#body))))
You would use it as follows:
(with-aref (w x y z) vec
(setf w (+ x y z)))
With a bit more work, you can also support indices and different categories of accessors. Let's say each binding is a triple (i n k) where i is an identifier, n a number (or nil) that represents the numerical index and k is either :place, :value or nil; :place binds the symbol with symbol-macrolet, :value just binds it with let.
First, let's help the user by providing shortcut notations:
x stands for (x nil nil)
(x o) either stands for (x o nil) or (x nil o), depending on whether option o is a number or a symbol (at macroexpansion time).
Besides, we may want to automatically ignore the nil identifier, the empty symbol || or symbols starting with an underscore (e.g. _, _var).
Here is the normalization function:
(defun normalize-index (index)
(flet ((ret (i n k)
(let ((ignored (or (null i)
(string= i "")
(char= #\_ (char (string i) 0)))))
(list (if ignored (gensym) i) n k ignored))))
(let ((index (alexandria:ensure-list index)))
(typecase index
(null (ret nil nil nil))
(cons (destructuring-bind (i &optional n (k nil kp)) index
(if kp
(ret i n k)
(etypecase n
(symbol (ret i nil n))
((integer 0) (ret i n nil))))))))))
We can apply this normalization to a list of indices, and keep track of ignored symbols:
(defun normalize (indices)
(loop
for i in indices
for norm = (normalize-index i)
for (index number kind ignore) = norm
collect norm into normalized
when ignore
collect index into ignored
finally (return (values normalized ignored))))
Then, we take care of nil numbers in normalized entries. We want the indices to increase from the last used index, or be given explicitly by the user:
(defun renumber (indices)
(loop
for (v n k) in indices
for next = nil then (1+ index)
for index = (or n next 0)
collect (list v index k)))
For example:
(renumber (normalize '(a b c)))
((A 0 NIL) (B 1 NIL) (C 2 NIL))
(renumber (normalize '((a 10) b c)))
((A 10 NIL) (B 11 NIL) (C 12 NIL))
(renumber (normalize '((a 10) (b 3) c)))
((A 10 NIL) (B 3 NIL) (C 4 NIL))
We do the same for the kind of variable we bind:
(defun rekind (indices)
(loop
for (v n k) in indices
for next = nil then kind
for kind = (or k next :place)
collect (list v n kind)))
For example:
(rekind (normalize '(a b c)))
((A NIL :PLACE) (B NIL :PLACE) (C NIL :PLACE))
(rekind (normalize '(a (b :value) c)))
((A NIL :PLACE) (B NIL :VALUE) (C NIL :VALUE))
Finally, all those steps are combined in parse-indices:
(defun parse-indices (indices)
(multiple-value-bind (normalized ignored) (normalize indices)
(values (rekind (renumber normalized))
ignored)))
Finally, the macro is as follows:
(defmacro with-aref ((&rest indices) array &body body)
(multiple-value-bind (normalized ignored) (parse-indices indices)
(labels ((ignored (b) (remove-if-not #'ignoredp (mapcar #'car b)))
(ignoredp (s) (member s ignored)))
(loop
with a = (gensym)
for (i n k) in normalized
for binding = `(,i (aref ,a ,n))
when (eq k :value) collect binding into values
when (eq k :place) collect binding into places
finally (return
`(let ((,a ,array))
(let ,values
(declare (ignore ,#(ignored values)))
(symbol-macrolet ,places
(declare (ignore ,#(ignored places)))
,#body))))))))
For example:
(let ((vec (vector 0 1 2 3 4 5 6 7 8 9 10)))
(prog1 vec
(with-aref ((a 2) (b :value) c _ _ d (e 0) (f 1)) vec
(setf a (list a b c d e f)))))
The above is macroexpanded as:
(LET ((VEC (VECTOR 0 1 2 3 4 5 6 7 8 9 10)))
(LET ((#:G1898 VEC))
(LET ((#:G1901 VEC))
(LET ((B (AREF #:G1901 3))
(C (AREF #:G1901 4))
(#:G1899 (AREF #:G1901 5))
(#:G1900 (AREF #:G1901 6))
(D (AREF #:G1901 7))
(E (AREF #:G1901 0))
(F (AREF #:G1901 1)))
(DECLARE (IGNORE #:G1899 #:G1900))
(SYMBOL-MACROLET ((A (AREF #:G1901 2)))
(DECLARE (IGNORE))
(LET* ((#:G19011902 #:G1901)
(#:NEW1 (LIST (AREF #:G1901 2) B C D E F)))
(FUNCALL #'(SETF AREF) #:NEW1 #:G19011902 2)))))
#:G1898))
It produces the following result
#(0 1 (2 3 4 7 0 1) 3 4 5 6 7 8 9 10)
coredump's answer is lovely. This is a variant of it which binds variables rather than accessors, and also lets you optionally specify indices. So
(with-vector-elements ((a 3) b) x
...)
will bind a to the result of (aref x 3) and b to the result of (aref x 4), for instance.
This is really only useful over coredump's answer if you're intending to (a) not write back to the vector and (b) use the bindings a lot, so you want to avoid a lot of possible arefs (which I don't think compilers can generally optimize away without some fairly strong assumptions).
(defmacro with-vector-elements ((&rest indices) vector &body forms)
(let ((canonical-indices
(loop with i = 0
for index in indices
collect (etypecase index
(symbol
(prog1
`(,index ,i)
(incf i)))
(cons
(destructuring-bind (var idx) index
(assert (and (symbolp var)
(typep idx '(and fixnum (integer 0))))
(var idx) "Invalid index spec")
(prog1
index
(setf i (1+ idx))))))))
(vname (gensym "V")))
`(let ((,vname ,vector))
(let ,(loop for (var index) in canonical-indices
collect `(,var (aref ,vname ,index)))
,#forms))))
There is also a package called metabang-bind - with nickname bind - in which the function bind can handle much more destructuring situations:
(ql:quickload :metabang-bind)
(in-package :metabang-bind)
(bind ((#(a b c) #(1 2 3)))
(list a b c))
;; => (1 2 3)
If not using in-package, you can call the function as bind:bind.
The function bind you can think of roughly as a destructuring-let* (similar idea to clojure's let, however not so clean in syntax but understandable because it has also to handle structs and classes and also values).
All the other use cases it can handle are described here.
Is it possible, in scheme, to do something like the following?
(define (foo a b c d e)
(let ((recurse (foo a b c (bar d) e))
(if (= d e)
#f
(recurse))))
I have a function with a rather lengthy set of arguments and some conditions that recursively call it and it's rather unsightly, I'd like to be able to shorthand it.
I've tried this a few different ways but I either get an infinite loop (I presume let is eagerly evaluating) or I get bad let form errors.
You need a thunk:
(define (foo a b c d e)
(let ((recurse (lambda () (foo a b c (bar d) e))))
(if (= d e)
#f
(recurse))))
This then permits you to add arguments if some but not all of the parameters change.
Since most of the arguments don't change you should consider using named let:
(define (foo a b c d e)
(let recurse ((d d))
(if (= d e)
#f
(recurse (cdr d)))))
This is basically the same as:
(define (foo a b c d e)
(define (recurse d)
(if (= d e)
#f
(recurse (cdr d))))
// start recursion
(recurse d))
All the other variables will be available through the closure.
So I've been trying to solve this problem:
Given an element E and a list L insert E into every position in the list L (so the result is a list of lists). Example:
(insert-everywhere 'a '(b c)) would give ((a b c) (b a c) (b c a))
This is easy enough but there is one other condition in my problem that is making it difficult for me - if an element of L is a list itself then the element must also be inserted into every position in the sublist. For example:
(insert-everywhere 'd '(a (b c))) would return: ((d a (b c)) (a d (b c)) (a (d b c)) (a (b d c)) (a (b c d)) (a (b c) d)).
This is the code I have so far (which I mostly lifted from here):
#lang racket
(define (insert-at pos elmt lst)
(if (empty? lst) (list elmt)
(if (list? (car lst)) (insert-everywhere elmt (car lst))
(if (= 1 pos)
(cons elmt lst)
(cons (first lst)
(insert-at (- pos 1) elmt (rest lst)))))))
(define (insert-everywhere sym lst)
(remove-duplicates
(map (lambda (i)
(insert-at i sym lst))
(range 1 (+ 2 (length lst))))))
where this line: (if (list? (car lst)) (insert-everywhere elmt (car lst)) is supposed to handle the sublists but it isn't working. (If I run (insert-everywhere 'd '(a (b c))) with the above code I get ((d a (b c)) (a (d b c) (b d c) (b c d))))
If anyone has suggestions on how to approach this differently I'd be happy to hear.
I wouldn't do indexing as it is very inefficient. Rather reverse the input list and build the list from end to beginning making the results in reverse order. You have a current list that you add elements to with cons that you use to add new additions to the results and each level each result that existed gets the one element added too.
As parameters you have state. When i made a reference I used result and cur and typically my iteration did like this for (insert-everywhere 'd '(a b c)):
lst cur results
(c b a) () ((d))
(b a) (c) ((d c) (c d))
(a) (b c) ((d b c) (b d c) (b c d))
() (a b c) ((d a b c) (a d b c) (a b d c) (a b c d)))
Now adding support for sublists are just doing the same with them and then do a map such that you create one result per sublist in the result, adding cur in addition to adding it as an element.
Notice all new results are just cur with an added inserted element and all th erest gets a new element in fron which is the first element of the input. cur will grow and it is shared so only the elements up to the inserted element will be unique to that sub result.
I have a working implementation, but it's no fun getting the solution prematurely. Have fun.
I came across Pascal Bourguignon's solutions of the 99 Lisp problems and was wondering if his recursive solution of problem 27 using a nested mapcan-mapcar-construct could also be written using nested loops.
His solution is definitely very elegant:
(defun group (set sizes)
(cond
((endp sizes)
(error "Not enough sizes given."))
((endp (rest sizes))
(if (= (first sizes) (length set))
(list (list set))
(error "Cardinal mismatch |set| = ~A ; required ~A"
(length set) (first sizes))))
(t
(mapcan (lambda (combi)
(mapcar (lambda (group) (cons combi group))
(group (set-difference set combi) (rest sizes))))
(combinations (first sizes) set)))))
The function combinations is defined here as:
(defun combinations (count list)
(cond
((zerop count) '(())) ; one combination of zero element.
((endp list) '()) ; no combination from no element.
(t (nconc (mapcar (let ((item (first list)))
(lambda (combi) (cons item combi)))
(combinations (1- count) (rest list)))
(combinations count (rest list))))))
I started with a simple approach:
(defun group-iter (set sizes)
(loop :with size = (first sizes)
:for subgroup :in (combination size set)
:for remaining = (set-difference set subgroup)
:collect (list subgroup remaining) :into result
:finally (return result)))
which results in:
> (group-iter '(a b c d e f) '(2 2 2))
(((A B) (F E D C)) ((A C) (F E D B)) ((A D) (F E C B)) ((A E) (F D C B))
((A F) (E D C B)) ((B C) (F E D A)) ((B D) (F E C A)) ((B E) (F D C A))
((B F) (E D C A)) ((C D) (F E B A)) ((C E) (F D B A)) ((C F) (E D B A))
((D E) (F C B A)) ((D F) (E C B A)) ((E F) (D C B A)))
But now I am totally failing to implement the nesting which takes care of the further processing of remaining. As far as I understood there is always a way to express a recursion with a iteration but how does it look like here?
When I try this code on Emacs SLIME, the apply function gives a different result. Isn't it supposed to give the same result? Why does it give a different result? Thanks.
CL-USER> (apply #'(lambda (n)
(cons n '(b a))) '(c))
(C B A)
CL-USER> (cons '(c) '(b a))
((C) B A)
cons takes an element and a list as arguments. So (cons 'x '(a b c d)) will return (x a b c d).
apply takes a function and a list of arguments -- but the arguments will not be passed to the function as a list! They will be split and passed individually:
(apply #'+ '(1 2 3))
6
(actually, it takes one function, several arguments, of which the last must be a list -- this list will be split and treated as "the rest of the arguments to the function". try, for example, (apply #'+ 5 1 '(1 2 3)), which will return 12)
Now to your code:
The last argument you passed to the apply function is '(c), a list with one element, c. Apply will treat it as a list of arguments, so the first argument you passed to your lambda-form is c.
In the second call, you passed '(c) as first argument to cons. This is a list, which was correctly included in the first place of the resulting list: ( (c) b a).
The second call would be equivalent to the first if you did
(cons 'c '(b a))
(c b a)
And the first call would be equivalent to the second if you did
(apply #'(lambda (n) (cons n '(b a))) '((c)))
((c) b a)
CL-USER 51 > (cons '(c) '(b a))
((C) B A)
CL-USER 52 > (apply #'(lambda (n)
(cons n '(b a)))
'(c))
(C B A)
Let's use FUNCALL:
CL-USER 53 > (funcall #'(lambda (n)
(cons n '(b a)))
'(c))
((C) B A)
See also what happens when we apply a two element list:
CL-USER 54 > (apply #'(lambda (n)
(cons n '(b a)))
'(c d))
Error: #<anonymous interpreted function 40600008E4> got 2 args, wanted 1.
There is a symmetry between &rest arguments in functions and apply.
(defun function-with-rest (arg1 &rest argn)
(list arg1 argn))
(function-with-rest 1) ; ==> (1 ())
(function-with-rest 1 2) ; ==> (1 (2))
(function-with-rest 1 2 3 4 5) ; ==> (1 (2 3 4 5))
Imagine we want to take arg1 and argn and use it the same way with a function of our choice in the same manner as function-with-rest. We double the first argument and sum the rest.
(defun double-first-and-sum (arg1 &rest argn)
(apply #'+ (* arg1 2) argn))
(double-first-and-sum 1 1) ; ==> 3
(double-first-and-sum 4 5 6 7) ; ==> 26
The arguments between the function and the list of "rest" arguments are additional arguments that are always first:
(apply #'+ 1 '(2 3 4)) ; ==> (+ 1 2 3 4)
(apply #'+ 1 2 3 '(4)) ; ==> (+ 1 2 3 4)
This is very handy since often we want to add more arguments than we are passed (or else we could just have used the function apply is using in the first place. Here is something called zip:
(defun zip (&rest args)
(apply #'mapcar #'list args))
So what happens when you call it like this: (zip '(a b c) '(1 2 3))? Well args will be ((a b c) (1 2 3)) and the apply will make it become (mapcar #'list '(a b c) '(1 2 3)) which will result in ((a 1) (b 2) (c 3)). Do you see the symmetry?
Thus you could in your example you could have done this:
(apply #'(lambda (&rest n)
(cons n '(b a))) '(c))
;==> ((c) b a)
(apply #'(lambda (&rest n)
(cons n '(b a))) '(c d e))
;==> ((c d e) b a)