<div dir="ltr">Do refinement types work for what you want?<div><br></div><div><a href="http://docs.racket-lang.org/ts-reference/Experimental_Features.html?q=refinement#(form._((lib._typed-racket/base-env/prims..rkt)._declare-refinement))">http://docs.racket-lang.org/ts-reference/Experimental_Features.html?q=refinement#(form._((lib._typed-racket/base-env/prims..rkt)._declare-refinement))</a><br>
</div><div><br></div><div style><div>#lang typed/racket</div><div>(declare-refinement even?)</div><div>(: two (Refinement even?))</div><div>(define two</div><div> (let ((x 2))</div><div> (if (even? x) x (error 'bad))))</div>
<div><br></div><div>There are a couple of issues with them, mostly that they are not sound when dealing with mutable objects or non pure functions, which allows you to break the soundness of TR.<br></div></div><div style>
<a href="http://bugs.racket-lang.org/query/?cmd=view+audit-trail&pr=13059">http://bugs.racket-lang.org/query/?cmd=view+audit-trail&pr=13059</a></div></div><div class="gmail_extra"><br><br><div class="gmail_quote">
On Fri, Dec 28, 2012 at 2:17 PM, Ray Racine <span dir="ltr"><<a href="mailto:ray.racine@gmail.com" target="_blank">ray.racine@gmail.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div dir="ltr">Any plans something along the lines of Scala's proposed Value Types. <div><br></div><div><div>A path:</div><div><br></div><div>Allow for "special" struct: decls (vstruct: ?) where the parent is a limited subset of non structure parent types (base value types such as String, Number). </div>
<div><br></div><div>These special structures MUST contain one and only one value of the parent "special" type or it is a compile error. </div><div>The structure constructor does not construct the wrapping structure but instead passes through the wrapped value, but *always* invokes the validator during pass-thru.</div>
<div>TR treats the type as a subtype of the base value type.</div><div><br></div><div>e.g.</div><div><br></div><div>(struct: Identifier String ([value : String])</div><div> #:methods gen:validator (lambda (thing) ...) ;; is of type (Any -> Boolean : Identifier))</div>
<div><br></div><div>(define id (Identifier "myidentifier")) ;; validator invoked, no structure was allocated, `id' is just a String value, is a subtype of String.</div><div><br></div><div>
(define uc-id (Identifer (string-upcase id))) ;; validator invoked, as id is a just a string no unboxing in (string-upcase id), in fact no allocations here at all.</div><div><br></div><div>Under the covers the Identifier constructor never creates the structure, it acts as a pass through id : (String -> String) function. i.e. the runtime representation of `id' is always as a String so any struct <-> value boxing / unboxing is avoided. I think there is enough machinery in place to get pretty close to this.</div>
<div><br></div><div>What is gained?</div><div><br></div><div>What is done internally in TR defining Natural, Index, Exact-Positive-Integer can now be done without special internal defining, just another constrained base type. One can start to define things like Age [1 .. 120]. </div>
<div>Another is IMHO a HUGE source of program error is just not enough time to do proper validation at IO boundries where entering data is of the form Strings and Bytes and it needs to be lifted.</div><br>Consider the following typical use case from Amazon's AWS API, a Tasklist parameter.<br>
<br>Parameter - Tasklist : String[1-256]<br><br>Specifies the task list to poll for activity tasks.<br><br>The specified string must not start or end with whitespace. It must not contain a : (colon), / (slash), | (vertical bar), or any control characters (\u0000-\u001f | \u007f - \u009f). Also, it must not contain the literal string "arn".<br>
<br>Most likely, I'll punt.</div><div><br></div><div>(: call-it (String ... -> ...))</div><div>(define (call-it task-list ...)</div><div><br></div><div>If I'm ambitious today.</div><div>
<br></div><div>;; would prefer (define-new-type Tasklist String)</div><div>(define-type Tasklist String) ;; tighten things later down the road, <sigh> none type generative</div><div><br></div><div>
<div>(: call-it (Tasklist ... -> ...))</div><div>(define (call-it task-list ...)</div></div><div><br></div><div>What I'd like to do.</div><div><br></div><div>(define-value-type Tasklist String [1 .. 256] (lambda (this) ....)) ;; mad use of regexp in validator fn (Any -> Boolean : Tasklist)</div>
<div><br></div><div>(call-it (Tasklist "mytasklist") ...)</div><div><br></div><div>(call-it (Tasklist "arn:bad/tasklist") ...)</div><div><br></div><div>(define-value-type Age Integer [1 .. 120]) ;; no special validation beyond bounds check.</div>
<div><br></div><div> </div><div><br></div><div><br></div><div><br></div></div>
<br>____________________<br>
Racket Users list:<br>
<a href="http://lists.racket-lang.org/users" target="_blank">http://lists.racket-lang.org/users</a><br>
<br></blockquote></div><br></div>