Re: [racket-users] [racket users] make-keyword-procedure follow-up

[Kevin Forchione]

> On Aug 30, 2019, at 1:19 AM, Philip McGrath  wrote:
> 
> Hi Kevin,
> 
> This is interesting! A number of people have wanted conveniences around 
> `keyword-apply` and accepting the same keywords as some other function. (The 
> trouble is, different people have different ideas of what those conveniences 
> should be.)
> 
> To start with, I had a few small suggestions about the code you sent: 
> `keyword-apply/filter` doesn't need to be a macro, and your macro should use 
> `fn` rather than `h`;
> the second result of `procedure-keywords` can be `#f` if the given procedure 
> accepts all keywords;
> `for/lists` might be more convenient than `for/fold`;
> I think some of the cases in `filter/kw` get the filtering backwards;
> keywords can be `quote`-ed as values, so you don't need so many 
> symbol-to-string-to-keyword conversions;
> using `set!` this way will cause lots of problems;
> there's no need for `list->vector` in `get-kw-val`; and
> in `def`, it might be better to use syntax parameters than to break hygiene.
> You may know this, but all keyword-accepting functions in Racket are 
> effectively implemented with `make-keyword-procedure`: the variants of 
> `lambda` and `define` that use `kw-formals` are macros that expand to more 
> primitive versions that don't know about keywords. The macros also do some 
> optimization where possible. For inspiration, you might be interested in the 
> implementation in 
> https://github.com/racket/racket/blob/master/racket/collects/racket/private/kw.rkt
>  
> 
> 
> For fun, I wrote a version that illustrates some of these suggestions and and 
> tries to do more work at compile-time. Be warned that it is not thoroughly 
> tested! I'm pasting it below, and I've also put it up as a Gist at 
> https://gist.github.com/LiberalArtist/292b6e99421bc76315110a59c0ce2b0d 
> 
> 
> -Philip

Hi Philip,
I’ve been using your kw-pass-through-lambda.rkt module quite happily for a 
while now. One issue I’ve run into though is that the local-keyword-apply 
appears to have an issue when the return is multiple values. For instance, in 
the test module, if you replace the:

  (define (h #:c c . x)
(list c x))


With:

  (define (h #:c c . x)
(values c x))

You’ll get an error: “Returned two values to single value return context”

Is there a way to fix this?

Kevin


-- 
You received this message because you are subscribed to the Google Groups 
"Racket Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to racket-users+unsubscr...@googlegroups.com.
To view this discussion on the web visit 
https://groups.google.com/d/msgid/racket-users/2A70A1D7-741E-4CDF-A29E-49BF1BF14EB7%40gmail.com.

Re: [racket-users] [racket users] make-keyword-procedure follow-up

[Jack Firth]

You might have some success using my arguments 
 package, which defines 
an arguments? data structure wrapping a bundle of positional and keyword 
args. It also provides some functions that are like make-keyword-procedure 
and keyword-apply, but with a nicer interface based on the arguments 
structure:

> (apply/arguments sort
   (arguments '("fo" "bar" "bazz") < #:key 
string-length))
'("bar" "bazz" "fo")
> (define/arguments (keywords-product args)
(for/product ([(k v) (in-hash (arguments-keyword args))])
  v))
> (keywords-product #:foo 2 #:bar 3)
6
> (keywords-product 'ignored #:baz 6 #:blah 4)
24

-- 
You received this message because you are subscribed to the Google Groups 
"Racket Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to racket-users+unsubscr...@googlegroups.com.
To view this discussion on the web visit 
https://groups.google.com/d/msgid/racket-users/83554d22-5863-4aa8-9d62-cbf6caef14c2%40googlegroups.com.

Re: [racket-users] [racket users] make-keyword-procedure follow-up

[Kevin Forchione]

> On Aug 30, 2019, at 1:19 AM, Philip McGrath  wrote:
> 
> Hi Kevin,
> 
> This is interesting! A number of people have wanted conveniences around 
> `keyword-apply` and accepting the same keywords as some other function. (The 
> trouble is, different people have different ideas of what those conveniences 
> should be.)
> 
> To start with, I had a few small suggestions about the code you sent: 
> `keyword-apply/filter` doesn't need to be a macro, and your macro should use 
> `fn` rather than `h`;
> the second result of `procedure-keywords` can be `#f` if the given procedure 
> accepts all keywords;
> `for/lists` might be more convenient than `for/fold`;
> I think some of the cases in `filter/kw` get the filtering backwards;
> keywords can be `quote`-ed as values, so you don't need so many 
> symbol-to-string-to-keyword conversions;
> using `set!` this way will cause lots of problems;
> there's no need for `list->vector` in `get-kw-val`; and
> in `def`, it might be better to use syntax parameters than to break hygiene.
> You may know this, but all keyword-accepting functions in Racket are 
> effectively implemented with `make-keyword-procedure`: the variants of 
> `lambda` and `define` that use `kw-formals` are macros that expand to more 
> primitive versions that don't know about keywords. The macros also do some 
> optimization where possible. For inspiration, you might be interested in the 
> implementation in 
> https://github.com/racket/racket/blob/master/racket/collects/racket/private/kw.rkt
>  
> 
> 
> For fun, I wrote a version that illustrates some of these suggestions and and 
> tries to do more work at compile-time. Be warned that it is not thoroughly 
> tested! I'm pasting it below, and I've also put it up as a Gist at 
> https://gist.github.com/LiberalArtist/292b6e99421bc76315110a59c0ce2b0d 
> 
> 
> -Philip
Thanks, Philip! Sorry for the bug in the keyword-apply/filter macro. I 
discovered it as well last night when I decided to convert the macro to a 
function. Glad others are working on this idea as well, perhaps something will 
find its way into the main package. As with much of my own code development  
No, I hadn’t realized make-keyword-procedure formed the basis of Racet keyword 
functions — I’m often being surprised like this — seems somewhere along the way 
I missed some fundamental primer of elementary concepts :) 

I’ve not had a chance to peruse the code and websites yet, but as clarification 
of what I have in mind: The problem I’m working on involves a function call f, 
in which some of the parameters are used to lookup functions in a table and 
then apply the remaining parameters in the application of that retrieved 
function g. All was going smoothly until I stored a function in the table that 
used keywords. So the filtering I’ve been working on would allow f to reference 
its own keywords without complaining about g’s and to “filter out” f’s keywords 
from those passed to g when they did not overlap (so to speak). Any additional 
keywords passed in the function call that were not defined by g would then be 
eared by g, which is why Ive been referring to f as a “pass-through”.  So if f 
defined #:foo and #:bar and g defined #:foo and #:baz f would bind any #:foo 
and #:bar values (or use its default values) for its own process, but pass g 
#:foo and #:baz as provided by the parameters passed in the call too f. Of 
course that lasts the flexibility that the #:foo value passed to f might be 
different from that of g — but that’s where I decided to draw the line :) 

Kevin

-- 
You received this message because you are subscribed to the Google Groups 
"Racket Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to racket-users+unsubscr...@googlegroups.com.
To view this discussion on the web visit 
https://groups.google.com/d/msgid/racket-users/D3F9EF44-D45D-4D02-A571-257F074CA499%40gmail.com.

Re: [racket-users] [racket users] make-keyword-procedure follow-up

[Philip McGrath]

Hi Kevin,

This is interesting! A number of people have wanted conveniences around
`keyword-apply` and accepting the same keywords as some other function.
(The trouble is, different people have different ideas of what those
conveniences should be.)

To start with, I had a few small suggestions about the code you sent:

   - `keyword-apply/filter` doesn't need to be a macro, and your macro
   should use `fn` rather than `h`;
   - the second result of `procedure-keywords` can be `#f` if the given
   procedure accepts all keywords;
   - `for/lists` might be more convenient than `for/fold`;
   - I think some of the cases in `filter/kw` get the filtering backwards;
   - keywords can be `quote`-ed as values, so you don't need so many
   symbol-to-string-to-keyword conversions;
   - using `set!` this way will cause lots of problems;
   - there's no need for `list->vector` in `get-kw-val`; and
   - in `def`, it might be better to use syntax parameters than to break
   hygiene.

You may know this, but all keyword-accepting functions in Racket are
effectively implemented with `make-keyword-procedure`: the variants of
`lambda` and `define` that use `kw-formals` are macros that expand to more
primitive versions that don't know about keywords. The macros also do some
optimization where possible. For inspiration, you might be interested in
the implementation in
https://github.com/racket/racket/blob/master/racket/collects/racket/private/kw.rkt

For fun, I wrote a version that illustrates some of these suggestions and
and tries to do more work at compile-time. Be warned that it is not
thoroughly tested! I'm pasting it below, and I've also put it up as a Gist
at https://gist.github.com/LiberalArtist/292b6e99421bc76315110a59c0ce2b0d

-Philip

#lang racket

;; License: Apache-2

(provide kw-pass-through-lambda
 local-keyword-apply
 local-kw-lst
 local-kw-val-lst
 (contract-out
  [keyword-apply/filter
   (-> procedure? (listof keyword?) list? list?
   any)]))

(module+ test
  (require rackunit)
  (define (h #:c c . x)
(list c x))
  (define g
(kw-pass-through-lambda (#:c [c 0] . args)
  (list c (local-keyword-apply h args
  (define f
(kw-pass-through-lambda (n p #:a [a 0] #:b [b 0] . ns)
  (list local-kw-lst local-kw-val-lst
a b n p ns
(local-keyword-apply g ns
  (check-equal? (f 2 3 4 5 #:a 42 #:c 52)
'((#:a #:c) (42 52) 42 0 2 3 (4 5) (52 (52 (4 5)
  ;; My implementation of "filtering" keywords has a different result,
  ;; but maybe I don't understand what you were trying to do.
  ;; Your version did this:
  ;;   (check-exn #rx"procedure: h\n  given keyword: #:z"
  ;;  (λ () (f 2 3 4 5 #:z 42 #:c 52)))
  ;; Mine does this instead:
  (check-equal? (f 2 3 4 5 #:z 42 #:c 52)
'((#:c #:z) (52 42) 0 0 2 3 (4 5) (52 (52 (4 5))

;; potential further extensions:
;;  - make keyword-apply/filter and local-keyword-apply
;;accept extra keyword and by-position args like keyword-apply
;;  - implement a define version of kw-pass-through-lambda
;;  - various performance optimizations

(require syntax/parse/define
 racket/stxparam
 (for-syntax syntax/parse/lib/function-header
 racket/list
 racket/match
 racket/sequence
 syntax/transformer))



;; runtime support

(define (keyword-apply/filter proc kw-lst kw-val-lst by-pos-args)
  ;; like keyword-apply, but skips keywords that aren't allowed
  (define-values [required-kws allowed-kws]
(procedure-keywords proc))
  (match allowed-kws
[#f ;; accepts all keywords
 (keyword-apply proc kw-lst kw-val-lst by-pos-args)]
['() ;; accepts no keywords
 (apply proc by-pos-args)]
[_
 (for/lists [kw-lst*
 kw-val-lst*
 #:result (keyword-apply proc
 kw-lst*
 kw-val-lst*
 by-pos-args)]
([kw (in-list kw-lst)]
 [val (in-list kw-val-lst)]
 #:when (memq kw allowed-kws))
   (values kw val))]))

(define (kw-arg-ref kw kw-lst kw-val-lst
[fail-thunk
 ;; we'll use procedure-reduce-keyword-arity
 ;; to avoid getting here when required kws are missing
 (λ () (error 'kw-arg-ref "shouldn't get here"))])
  (or (for/first ([this-kw (in-list kw-lst)]
  [val (in-list kw-val-lst)]
  #:break (keywordlist #'(arg.name ... (~? rest-arg-name
   "duplicate argument name"
   #:fail-when (check-duplicates (syntax->list #'((~? arg.kw) ...))
 #:key 

[racket-users] racket users] make-keyword-procedure follow-up

[Kevin Forchione]

> On Aug 29, 2019, at 1:24 PM, Kevin Forchione  wrote:
> 
> Hi guys,
> I’ve been working for a little while with the idea of being able to pass 
> keyword arguments through a function that doesn’t define them. Additionally I 
> wanted to allow the “pass-through” function to define its own keywords. 
> Additionally didn’t want to have to pre-specify what function might be on the 
> receiving end of the call. But finally, if both pass-through and called 
> functions define the same keywords I didn’t want to have to differentiate 
> between them. 
> 
> - This seems to involve some combination of make-keyword-procedure and 
> keyword-apply.
> -  Since make-keyword-procedure expects a “vanilla” function (one without 
> keywords specified) I decided to define a macro that would wrap the function 
> in a let  that with default bindings for each keyword defined by the 
> pass-through. Inside the function I would then assign any values provided by 
> the function call to those variables. 
> - Additionally I would build a parameterized list of keywords defined by the 
> pass-through chain. These would be used in conjunction with the keyword list 
> produced by procedure-keywords and the keywords/values captured by the 
> function call to “filter” the lists used by keyword-apply. The idea being to 
> eliminate any keyword/values supplied to the pass-through and defined by the 
> pass-through that were not defined by the  called function. This would allow 
> keywords not defined by either to be error by the called function. 
> 
> As you can see, it’s a convoluted approach and I’m not sure how robust it 
> actually. I’m presenting working code (for my test cases…) but also wondering 
> if someone hasn’t already crated that wheel. :) 
A little syntactic sugar makes it appear more palatable with the addition of:

(define-syntax (keyword-apply/filter stx)
  (syntax-parse stx
[(_ fn kw kv args)
 #'(let ()
 (define-values (rkw akw) (procedure-keywords h))
 (define-values (Δkw Δkv) (filter/kw (current-caller-kw) akw kw kv))
 (keyword-apply fn
  Δkw
  Δkv
  args))]))

And now the definitions are somewhat clearer:

(define (h #:c c . x)
  (list c x))
(def (g ((c 0)) . args)
  (list c (keyword-apply/filter h kw kv args)))
(def (f ((a 0)(b 0)) n p . ns)
  (list kw kv a b n p ns (keyword-apply/filter g kw kv ns)))

Kevin

-- 
You received this message because you are subscribed to the Google Groups 
"Racket Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to racket-users+unsubscr...@googlegroups.com.
To view this discussion on the web visit 
https://groups.google.com/d/msgid/racket-users/751B8754-6244-4A44-B779-ECD52DA070B5%40gmail.com.

[racket-users] [racket users] make-keyword-procedure follow-up

[Kevin Forchione]

Hi guys,
I’ve been working for a little while with the idea of being able to pass 
keyword arguments through a function that doesn’t define them. Additionally I 
wanted to allow the “pass-through” function to define its own keywords. 
Additionally didn’t want to have to pre-specify what function might be on the 
receiving end of the call. But finally, if both pass-through and called 
functions define the same keywords I didn’t want to have to differentiate 
between them. 

- This seems to involve some combination of make-keyword-procedure and 
keyword-apply.
-  Since make-keyword-procedure expects a “vanilla” function (one without 
keywords specified) I decided to define a macro that would wrap the function in 
a let  that with default bindings for each keyword defined by the pass-through. 
Inside the function I would then assign any values provided by the function 
call to those variables. 
- Additionally I would build a parameterized list of keywords defined by the 
pass-through chain. These would be used in conjunction with the keyword list 
produced by procedure-keywords and the keywords/values captured by the function 
call to “filter” the lists used by keyword-apply. The idea being to eliminate 
any keyword/values supplied to the pass-through and defined by the pass-through 
that were not defined by the  called function. This would allow keywords not 
defined by either to be error by the called function. 

As you can see, it’s a convoluted approach and I’m not sure how robust it 
actually. I’m presenting working code (for my test cases…) but also wondering 
if someone hasn’t already crated that wheel. :) 

#lang racket

(require (for-syntax syntax/parse
 racket/syntax))

(define current-caller-kw (make-parameter '()))

(define (get-kw-val w v kw kv)
  (define key (string->keyword (symbol->string w)))
  (define kws (list->vector kw))
  (define idx (vector-member key kws))
  (cond
[(false? idx) v]
[else (define kvs (list->vector kv))
  (vector-ref kvs idx)]))

(define-syntax (def stx)
  (syntax-parse stx
[(_ (f ((w v) ...) k ... . ks) body0 body ...)
 (with-syntax ([kw (format-id #'f "kw")]
   [kv (format-id #'f "kv")])
   #'(define f
   (let ([w v] ...)
 (make-keyword-procedure
  (λ (kw kv k ... . ks)
(parameterize ([current-caller-kw
(append (current-caller-kw)
(map (λ (x) (string->keyword 
(symbol->string x)))
 (list 'w ...)))])
  (set! w (get-kw-val 'w w kw kv)) ...
body0 body ...))]))

(define (filter/kw ckw fkw kw kv)
  (cond
[(empty? ckw) (values kw kv)]
[(empty? (remove* fkw ckw)) (values kw kv)]
[else
 (define diff (remove* fkw ckw))
 (define vkw (list->vector kw))
 (define vkv (list->vector kv))
 (for/fold ([wacc '()] [vacc '()])
   ([v kv]
[k kw] #:unless (member k diff))
   (values (append wacc (list k)) (append vacc (list v]))

(define (h #:c c . x) (list c x))

(def (g ((c 0)) . args)
  (define-values (rkw akw) (procedure-keywords h))
  (define-values (Δkw Δkv) (filter/kw (current-caller-kw) akw kw kv))
  (list c (keyword-apply h
 Δkw
 Δkv
 args)))
(def (f ((a 0)(b 0)) n p . ns) (list kw kv a b n p ns (keyword-apply g
 kw
 kv
 ns)))


;=> '((#:a #:c) (42 52) 42 0 2 3 (4 5) (52 (52 (4 5
(f 2 3 4 5 #:a 42 #:c 52)
;=> application: procedure does not expect an argument with given keyword
;  procedure: h
;  given keyword: #:z
;  arguments...:
(f 2 3 4 5 #:z 42 #:c 52)

Kevin

-- 
You received this message because you are subscribed to the Google Groups 
"Racket Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to racket-users+unsubscr...@googlegroups.com.
To view this discussion on the web visit 
https://groups.google.com/d/msgid/racket-users/27D5D96D-F9D9-4ECB-9AE0-92FD1EB065C8%40gmail.com.