add keyword-apply/dict to racket/dict

[AlexKnauth]

See #2548. The keyword-apply/dict function is a variation on keyword-apply that solves two problems:

• Keyword sorting
• Parallel-lists where keyword and value are associated only by having the "same index"

(keyword-apply/dict proc kw-dict pos-arg ... pos-args) -> any
      proc : procedure?
   kw-dict : (dict/c keyword? any/c)
   pos-arg : any/c
  pos-args : (listof any/c)

[sorawee]

In the same spirit as apply, can keyword-apply/dict support #:<kw> kw-arg ... too?

[jackfirth]

@sorawee That seems reasonable but also a little dangerous, since the static keywords could overlap with the dynamic ones. Example: (keyword-apply/dict some-keyword-function (hash '#:color "green") #:color "red" '()). I assume that would be an error?

@AlexKnauth The gen:dict interface allows duplicate keys (but leaves their behavior unspecified), and lists implement the dictionary interface by using assoc. What should this code do?

(keyword-apply/dict
  some-keyword-function
  '((#:color . "red") (#:color . "green"))
  '())

I think a naive implementation of keyword-apply/dict would ignore the second color keyword (because that's what dict-ref on an association list would do), but I'd much rather the implementation throw an error. If this happened in real code it almost certainly means there's a bug hidden somewhere.

[AlexKnauth]

I think that keyword-apply/dict should interpret all valid dicts as dict operations would interpret them. Either that's an invalid dict and dict? should return false, or keyword-apply/dict should interpret the dict as the dict operations do.

[AlexKnauth]

@sorawee It does support #:<kw> kw-arg ..., but I didn't document that. Maybe I should.

@jackfirth (keyword-apply/dict some-keyword-function (hash '#:color "green") #:color "red" '()) is an error just like (keyword-apply some-keyword-function '(#:color) '("green") #:color "red" '()) is, although the error message is about keyword-apply instead of keyword-apply/dict because it just delegates to keyword-apply of keyword-apply in that situation.

[AlexKnauth]

@sorawee I have now documented that keyword-apply/dict supports #:<kw> kw-arg .... The rendered docs look like

[AlexKnauth]

@jackfirth I have updated the error message, to be about keyword-apply/dict instead of keyword-apply.

> (keyword-apply/dict void (hash '#:color "green") #:color "red" '())
keyword-apply/dict: keyword duplicated in dict and direct keyword arguments: '#:color

[sorawee]

What is dict/c? I searched and only found one from mischief which I don't think is what you want.

[AlexKnauth]

Oh, I had assumed (dict/c keyword? any/c) was like (hash/c keyword? any/c), but for generic dicts and not just hash tables. But I suppose if it doesn't exist, maybe just dict? should go there? Although I would prefer it to say the keys must be keywords... would (and/c dict? (compose (listof keyword?) dict-keys)) work better? Kinda long though.

[AlexKnauth]

Okay, I've changed it to dict? for now

[AlexKnauth]

Is it okay if I try to copy the implementation of hash/c and modify it to define dict/c?

[samth]

I think that should be a separate PR. Maybe just use dict? here.

[samth]

This needs some tests, and a history note. Also, I'm not sure that the dict documentation is the right place for this. In general, since this seems like a broadly-useful feature having it presented as dict-specific seems wrong.

[sorawee]

Perhaps racket/function is a more appropriate place?

[AlexKnauth]

Currently racket/function doesn't depend on racket/dict or racket/private/dict. Is it okay to add that dependency?

If so then racket/function sounds like a good place.

[sorawee]

Yeah, that dependency will definitely slow down startup time a bit.

Following is the result of running for i in {1..10}; do time racket test.rkt; done on various programs:

#lang racket/base

#|
racket test.rkt  0.12s user 0.03s system 97% cpu 0.154 total
racket test.rkt  0.13s user 0.03s system 94% cpu 0.168 total
racket test.rkt  0.11s user 0.03s system 97% cpu 0.149 total
racket test.rkt  0.12s user 0.03s system 96% cpu 0.157 total
racket test.rkt  0.12s user 0.03s system 96% cpu 0.154 total
racket test.rkt  0.12s user 0.03s system 94% cpu 0.157 total
racket test.rkt  0.12s user 0.03s system 96% cpu 0.151 total
racket test.rkt  0.12s user 0.03s system 96% cpu 0.160 total
racket test.rkt  0.12s user 0.03s system 95% cpu 0.160 total
racket test.rkt  0.12s user 0.03s system 94% cpu 0.157 total
|#

#lang racket/base

(require racket/function)

#| 
racket test.rkt  0.12s user 0.03s system 95% cpu 0.161 total
racket test.rkt  0.14s user 0.04s system 89% cpu 0.196 total
racket test.rkt  0.12s user 0.03s system 97% cpu 0.158 total
racket test.rkt  0.12s user 0.03s system 96% cpu 0.159 total
racket test.rkt  0.12s user 0.03s system 95% cpu 0.155 total
racket test.rkt  0.12s user 0.03s system 97% cpu 0.154 total
racket test.rkt  0.13s user 0.03s system 95% cpu 0.169 total
racket test.rkt  0.12s user 0.03s system 95% cpu 0.158 total
racket test.rkt  0.12s user 0.03s system 98% cpu 0.156 total
racket test.rkt  0.12s user 0.03s system 97% cpu 0.151 total
|#

#lang racket/base

(require racket/dict)

racket test.rkt  0.20s user 0.05s system 96% cpu 0.256 total
racket test.rkt  0.20s user 0.05s system 97% cpu 0.253 total
racket test.rkt  0.20s user 0.05s system 96% cpu 0.251 total
racket test.rkt  0.20s user 0.05s system 93% cpu 0.272 total
racket test.rkt  0.19s user 0.05s system 96% cpu 0.251 total
racket test.rkt  0.20s user 0.05s system 95% cpu 0.257 total
racket test.rkt  0.21s user 0.05s system 95% cpu 0.273 total
racket test.rkt  0.20s user 0.05s system 95% cpu 0.260 total
racket test.rkt  0.20s user 0.05s system 95% cpu 0.269 total
racket test.rkt  0.19s user 0.05s system 97% cpu 0.246 total

And here are ones that use raco make (raco make test.rkt; for i in {1..10}; do time racket test.rkt; done)

#lang racket/base

#|
racket test.rkt  0.08s user 0.02s system 96% cpu 0.100 total
racket test.rkt  0.08s user 0.03s system 82% cpu 0.132 total
racket test.rkt  0.09s user 0.03s system 96% cpu 0.115 total
racket test.rkt  0.08s user 0.02s system 93% cpu 0.111 total
racket test.rkt  0.08s user 0.02s system 97% cpu 0.103 total
racket test.rkt  0.08s user 0.02s system 95% cpu 0.112 total
racket test.rkt  0.08s user 0.02s system 95% cpu 0.103 total
racket test.rkt  0.08s user 0.03s system 95% cpu 0.112 total
racket test.rkt  0.08s user 0.02s system 94% cpu 0.109 total
racket test.rkt  0.08s user 0.03s system 93% cpu 0.116 total
|#

#lang racket/base

(require racket/function)

#| 
racket test.rkt  0.08s user 0.02s system 95% cpu 0.108 total
racket test.rkt  0.09s user 0.03s system 94% cpu 0.123 total
racket test.rkt  0.09s user 0.03s system 95% cpu 0.119 total
racket test.rkt  0.08s user 0.02s system 95% cpu 0.108 total
racket test.rkt  0.08s user 0.02s system 94% cpu 0.111 total
racket test.rkt  0.08s user 0.03s system 93% cpu 0.116 total
racket test.rkt  0.08s user 0.03s system 96% cpu 0.109 total
racket test.rkt  0.08s user 0.02s system 95% cpu 0.108 total
racket test.rkt  0.08s user 0.02s system 96% cpu 0.106 total
racket test.rkt  0.08s user 0.02s system 97% cpu 0.104 total
|#

#lang racket/base

(require racket/dict)

racket test.rkt  0.15s user 0.04s system 96% cpu 0.202 total
racket test.rkt  0.15s user 0.04s system 95% cpu 0.196 total
racket test.rkt  0.14s user 0.04s system 95% cpu 0.183 total
racket test.rkt  0.14s user 0.04s system 95% cpu 0.188 total
racket test.rkt  0.15s user 0.04s system 93% cpu 0.199 total
racket test.rkt  0.14s user 0.04s system 95% cpu 0.186 total
racket test.rkt  0.15s user 0.04s system 91% cpu 0.210 total
racket test.rkt  0.14s user 0.04s system 96% cpu 0.188 total
racket test.rkt  0.14s user 0.04s system 97% cpu 0.178 total
racket test.rkt  0.15s user 0.04s system 94% cpu 0.195 total

[samth]

I'm fine with leaving it in racket/dict given that. The other option would be to skip generics entirely, and use hashes.

[Aside: the hyperfine tool is very useful for that kind of measurement.]

[AlexKnauth]

I have added some tests in pkgs/racket-test/tests/racket/keyword-apply-dict.rkt.

[sorawee]

Can this be merged by 7.9 release?

[AlexKnauth]

The last review mentioned 3 things: tests, a history note, and placement. I have added tests and a history note, and we agreed on placement in racket/dict. If there are no objections, can I merge at the end of the day today?