brev-separate
- brev-separate
- Making macros
- Making procedures
- define-closure
- match-define
- call-table, call-table*, call-vector, call-string, call-list, and call-record
- call-key*
- ct, ctq, ct*, ctq*
- define-some
- define-parameters
- define-curry
- c a.k.a. {{π}} a.k.a. @>
- fn
- over
- as-list
- as-string
- memoize and memoize!
- make-tree-accessor
- make-sloppy-tree-accessor
- newline1 and skip1
- Making values
- ref*
- Doing stuff
- Source code
This is brev-separate, a miscellaneous hodge-podge of macros and procedures that all have the shared aim of brevity. Sort of my take on the clojurian and miscmacros and (chicken base) genre.
It's called brev-separate since the full brev egg also imports and reexports a bunch of other eggs, including the aforementioned clojurian and miscmacros.
Making macros
Chicken has syntax-rules macros and ir-macros. Let's shave off some of the boiler plate so that it's easier to make macros.
define-syntax-rules
There is already the wonderful define-syntax-rule in miscmacros but here is also define-syntax-rules:
(define-syntax-rules foo () ((foo bar) (+ bar 3)) ((foo bar baz) (* bar baz)))
It's just the ultra lazy person's shorthand for:
(define-syntax foo (syntax-rules () ((foo bar) (+ bar 3)) ((foo bar baz) (* bar baz))))
define-ir-syntax*
define-ir-syntax* is more interesting.
(define-ir-syntax* name (pattern body) ...)
It uses matchable to dispatch between different call signatures (kinda similar to how syntax-rules work) while also allowing you to inject, compare, strip-syntax and syntax inside, as per usual with ir-macros.
Here's an example:
(define-ir-syntax* (aif test yes no) `(let ((,(inject 'it) ,test)) (if ,(inject 'it) ,yes ,no))) (aif (... expensive test ...) (car it) (print "oh! no!"))
When you have multiple call-signatures, wrap pattern / body set with parens.
(define-ir-syntax* ((aif #f yes no) no) ((aif test yes no) `(let ((,(inject 'it) ,test)) (if ,(inject 'it) ,yes ,no))))
define-ir-syntax
Sometimes pattern matching is overkill or you have something else in mind.
define-ir-syntax macros are just
(define-ir-syntax name body)
where the body has access to body, inject, compare, strip-syntax and syntax, as in the following example:
(define-ir-syntax comp-prod (apply * body)) (comp-prod 2 3 4)
β 24
As a rule of thumb, if you are deliberately injecting new names into the namespace that's when you are using ir-macros, and when you want to avoid doing that, use syntax-rules.
Making procedures
define-closure
(define-closure bindings head body ...)
This works like your normal
(define head body ...)
except that bindings are lexically closed over body.
(define-closure (x 0) (counter) (inc! x)) (counter) (counter) (counter)
β 1 2 3
The pairs of bindings aren't individual paren-wrapped, just alternating between name and expression. The set of bindings as a whole has parens.
(define-closure (x 0 y 10) (jolly) (list (inc! x) (dec! y))) (jolly) (jolly) (jolly)
β (1 9) (2 8) (3 7)
match-define
Deprecatedβ₯
Calling match-define directly is deprecated in favor of using the macro from match-generics, which can expand to it or to normal define as needed. (But match-define is still used internally to avoid circular dependencies.)
call-table, call-table*, call-vector, call-string, call-list, and call-record
A closure interface to hash-tables.
(define arity (call-table)) (define color (call-table)) (arity cons 2) (color cons 'blue) ((over (x cons)) (list arity color))
β (2 blue)
call-table takes two optional keyword argument, default:, to set the default response for unknown keys, and seed: which can be a hash-table or an alist, and defaults to empty.
There is also call-table* which by default cons its values to a list instead of replacing them. It takes four keyword arguments. proc: which defaults to cons, initial which defaults to '(), and unary which defaults to #f, and seed: as above.
Both versions of call-table lets you access the underlying hash-table by calling them with no arguments, and to set them by calling them with the keyword argument update:.
(color update: my-other-hash-table)
Full documentation for call-tables Full documentation for callable arrays
call-key*
Sometimes you think call-table is convenient but you only need one key.
For call-key, just use make-parameter.
But call-key* is awesome since it accumulates its values.
It has the same proc, unary, and initial keyword arguments as call-table*. It doesn't have seed because the idea is that you just use inititial. The generated procedure has update (which takes a new list as argument) and get (which you only need for unary call-keys).
(define horses (call-key*)) (horses 'ruby) (horses 'kind-girl) (horses 'tornado) (horses)
β (tornado kind-girl ruby)
ct, ctq, ct*, ctq*
This is sugar for creating call-tables with some values already filled.
The q variants are implicitly quasiquoted while the non-q variants aren't.
I.e.
(let ((banana-color 'yellow)) (ctq banana ,banana-color apple red))
is equivalent to
(let ((banana-color 'yellow)) (call-table seed: `((banana . ,banana-color) (apple . red))))
and
(let ((banana-color 'yellow)) (ct 'banana banana-color 'apple 'red))
The * variants create call-table* instances instead, and splice one level of lists:
(ctq* banana yellow apple (green red))
These call-tables aren't closed, you can add more keys and values to them.
define-some
This is for making functions that implicitly returns '() on an empty? first argument. In other words, it defines a body for patterns with some non-empty value as first argument, hence the name define-some.
For example,
(define-some (descseq num) (cons num (descseq (sub1 num))))
is shorthand for
(define (descseq num) (if (empty? num) '() (cons num (descseq (sub1 num)))))
so
(descseq 5)
β (5 4 3 2 1)
define-parameters
(define-parameters foo 0 bar #t baz '() quux 'banana)
is shorthand for
(define foo (make-parameter 0)) (define bar (make-parameter #t)) (define baz (make-parameter '())) (define quux (make-parameter 'banana))
define-curry
It's nice that you can make specific curries with the SRFI-219 style define heads (which is implemented per default in Chicken).
That's nice if you know exactly how many stragglers and how many immediate args you have, but sometimes you need the currying itself to be arbitrary arity.
Let's say you already have something like:
(define (foo bar baz bax) (print baz) (+ bar baz bax))
but you realize you need arbitrary-arity currying.
Just change it to use define-curry instead of define:
(define-curry (foo bar baz bax) (print baz) (+ bar baz bax)) (= (foo 100 20 3) ((foo 100) 20 3) ((foo 100 20) 3) ((foo) 100 20 3) (((foo) 100) 20 3) (((foo 100) 20) 3) ((((foo) 100) 20) 3))
Prints seven 20 and returns #t.
It only works when foo otherwise would have fixed arity.
c a.k.a. {{π}} a.k.a. @>
This isn't the traditional c-combinator from mockingbirds and such. It's just a one-letter spelling of "curry". It's a function combinator.
((c + 1 20 300) 4000 50000)
β 54321
I also exported it using the name π for those who find emoji names more comfortable to use due to namespace issues.
I later found out that @> from the holes egg is same the combinator as this. Then, a few months later than that, I found out that partial from Clojure is the same combinator as this.
It has arbitrary arity and can work on arbitrary arity functions, but isn't recursive to multiple levels.
fn
(fn body ...)
is shorthand for
(lambda some-basic-bindings body ...)
where some-basic-bindings is one of
- args
- (x . rest)
- (x y . rest)
- (x y z . rest)
and the fn macro automatically figures out which of those four you mean.
The function can refer to itself as f.
(In other words, if the body refers to f, the expansion gets wrapped in (letrec ((f β¦)) f).)
over
(over body ...)
is shorthand for
(c map (lambda some-basic-bindings body ...))
except that the map can take any number of lists (or other sequences, like strings) and that i is also anaphorically bound to the list index in body.
Here is an example:
((over (+ x x y i)) '(10 20 40) '(3 6 9))
β (23 47 91)
Like fn, the function can refer to itself as f.
((over (display x) (eif x (f (sub1 x)) x)) '(2 4 3))
displays 210432103210 and returns (0 0 0).
as-list
Here is a functional combinator for Scheme that lets its arguments treat their arguments as if they were lists.
((as-list (c filter odd?)) 130752)
β 1375
((as-list cdr reverse) 23311358)
β 5311332
((as-list delete-duplicates) 23311358)
β 23158
(define (vowel? l) ((as-list (c member l)) "aeiou")) ((as-list (c filter vowel?)) "magnetic mountaintop")
β βaeiouaioβ
Together with over:
((over (if (vowel? x) x (char-upcase x))) "fleet foxes")
β βFLeeT FoXeSβ
as-string
Sinilar to as-list but for when you wanna use string procedures on lists of characters, or on symbols.
memoize and memoize!
Similar to what's in other libraries, brev-separate ships a memoize, to cache function results for args. (memoize! proc) is a macro that expands to (define proc (memoize proc)), which is good for recursive functions.
make-tree-accessor
Sometimes you just need an arbitrarily long tree dereferencer.
(make-tree-accessor cddadadaddddar)
makes cddadadaddddar real. Works for any sequence of a's and d's.
make-sloppy-tree-accessor
As above, but uses scar and scdr instead of car and cdr.
newline1 and skip1
newline1 is a version of Scheme's newline that does nothing the first time it's called.
(for-each (fn (newline1) (display x)) '(a b c d e))
prints
a b c d e
skip1 is a general combinator that makes these:
(define tab1 (skip1 (fn (newline) (display " ")))) (for-each (fn (tab1) (display x)) '(ol li li li))
prints
ol li li li
To reset them (so they'll skip their next invocation), call them with a single argument, 'reset.
(newline1 'reset) (tab1 'reset)
skip1 can take any number of procedures with any number of arguments, it runs ((compose proc1 proc2 β¦) arg1 arg2 β¦).
For example,
(define conv1 (skip1 print add1 *)) (for-each conv1 '(1 2 3) '(10 20 30))
prints
41 91
Making values
with-result
This is something that is sometimes cozy:
(with-result (print 1 2 (save 3) 4 5 6) (print 7 8))
Prints 123456 and 78, returns 3
aif-with-result
(aif-with-result (odd? (save 3)) (+ it 4) (+ it 1000))
β 7
That tests the entire expression but only stores the saved part into it.
Combining aif and with-result would do the opposite:
(aif (with-result (pred? (save 27))) it #f)
That tests 27 and stores 27 into it, while the pred? call is thrown away.
keep
(keep proc argsβ¦) runs proc for its sideeffects and returns the last arg.
For example, (+ 3 (keep print 'hi 4)) prints hi4 and returns 7.
It is curried on the procedure so you can ((keep proc) argsβ¦).
A shorthand for with-result when you only have one expression and only care about the last (or only) arg.
empty?
This is a generic predicate to see if a string is "", a list is '(), a number is 0 etc.
eif, econd, ewhen, eor and eand
eif is a version of if (or, to be precise, of aif since it is anaphoric) that treats empty things as falsy.
(eif "" it 'no)
β no
econd, similarly, is an "empty is falsy" version of acond. There's also ewhen, eor and eand.
(eor 0 (add1 it))
β 1
like?
like? is a unary curried version of equal?
is?
is? is a unary curried version of eq?
scdr, and scar
A scar is like a car but returns '() if the pair has no car. A scdr is like a cdr but returns '() if the pair has no cdr.
normalize-absolute-pathname
(normalize-absolute-pathname filename)
Uses current-directory to try to figure out an absolute path for filename.
slice
Here is an generic slice multimethod for Scheme.
(slice '(hello now there you are) 1 3)
β (now there)
(slice "so this is where you are hiding" 3 7)
β (#\t #\h #\i #\s)
(let ((str "so this is where you are hiding")) (set! (slice str 3 7) "that") str)
β "so that is where you are hiding"
(slice 1243153 -3 -0)
β (1 5 3)
Because of Scheme's call-by-value semantics, set! doesn't work on numbers.β₯
descend
Descend is sort of like a named let except it does three magic things.
First of all, the let tag is always desc. I almost named the macro itself desc but that would've been bad since then you couldn't nest them.
Second of all, if the binding is to a value of the same name e.g. (lis lis) you can just put the name there. You can mix these shorthand bindings with normal bindings.
Now, if the first binding starts of as empty?, the third magic thing (which I'll get into shortly) is disabled and you can go on your merry way only using the above two magics.
(descend ((sum 0) (nums '(1 2 3 4))) (if (null? nums) sum (desc (+ sum (car nums)) (cdr nums))))
β 10
Otherwise, if it does start out non-empty...
(descend ((nums '(1 2 3 4))) (+ (car nums) (desc (cdr nums))))
β 10
That's right. It only recurs the value is non-empty.
?->
?-> is a combinator that takes a predicate test and a transformer, and returns a unary procedure that transforms its argument if the predicate applies.
That's a mouthful. Let's break it down with some examples:prcx
You can make a car that only cars if the list is not null:
(map (?-> (o not null?) car) '((a b c) () (1 2 3) () ())) β (a () 1 () ())
Or a reverse that only reverses if it receives a list:
(map (?-> list? reverse) '(a (l u f r e d n o w) "summer's" (y a d))) β (a (w o n d e r f u l) "summer's" (d a y))
The default when the predicate doesn't apply is to just pass through the argument, but you can set a different default with a keyword argument:
(map (?-> (o not null?) car #:default #f) '((a b c) () (1 2 3) () ())) β (a #f 1 #f #f)
with and this
Sort of like a one-branch aif or an always-true awhen.
(with (+ 13 14) (print it) (+ it it))
Prints 27 and returns 54.
There's also this that uses the word that as the anaphor instead, since the anaphor it is so often already taken in anaphoric programming.
string->read
Shorthand for (fn (with-input-from-string x read)).
string->dwim
Like string->read, but leave it as a string if it wouldn't make sense. (For example, contains spaces, periods, or parantheses.)
ref*
Ref* is a universal version of list-ref, hash-table-ref etc.
It can handle lists, alists, hash-tables, strings, vectors, records, and all callable procedures (like call-tables).
You can also pass multiple arguments to deref*erence recursively. Designed by Chris Brannon.
The order is left to right, like Java's dots or Clojure's arrow, unlike compose or nested calls. That is to say, the left-most is the outermost.
Here's an extended example.
(define foo (make-hash-table)) (set! (hash-table-ref* foo 'bar) '((x . #(1 2 3)) (y . #(4 5 6)))) (ref* foo 'bar) β ((x . #(1 2 3)) (y . #(4 5 6))) (ref* foo 'bar 'y) β #(4 5 6) (ref* foo 'bar 'y 0) β 4
There's also the refx partially applied variant: it takes only the indices and returns a procedure of one argument.
For example, you could make a car that worked on strings and vectors too, like this: (refx 0) and you could make a caddar like this: (refx 2 0).
Doing stuff
for-each-line
(for-each-line filename body ...)
body is called for its sideeffects once per line of text in filename with the variable line anaphorically bound to that line.
for-each-stdin
(for-each-stdin body ...)
body is called for its sideeffects once per line of text in standard input (a.k.a. (current-input-port)) with the variable line anaphorically bound to that line.
For example:
(for-each-line "/tmp/foo" (print line))
niy
(niy)
NIY stands for "not implemented yet".
Errors out if called with no arguments or if any of its arguments are true. Sort of like a living FIXME.
die
(die "Write your message here")
Like print, but writes to error port and calls (exit 1).
Source code
git clone https://idiomdrottning.org/brev-separate