Re: Enforcing (unsigned) long long ranges

Good catch. Yes, WebIDL apparently made the same mistake we had made at
first, not realizing why we need the narrower range. It should be changed
for the same reason we changed it in ES6.

For those who missed the earlier discussion, the "desirable properties wrt
equational reasoning" that Sigbjorn refers to are, for example:

var c = a + b;
if (isSafeInteger(a) && isSafeInteger(b) && isSafeInteger(c)) {
  // c represents the accurate sum of a and b
}

With the wider range, this guarantee is lost.



On Tue, Jan 7, 2014 at 11:44 PM, Sigbjorn Finne <sof@opera.com> wrote:

> Hi,
>
> currently, the ES binding for WebIDL uses the range of exactly
> representable integers [-2^53, 2^53] (*)
>
>   http://heycam.github.io/webidl/#es-long-long
>   http://heycam.github.io/webidl/#es-unsigned-long-long
>
> ES6 will provide support for testing the range of unambiguously
> representable integers [-2^53 + 1, 2^53 - 1]
>
>
> https://people.mozilla.org/~jorendorff/es6-draft.html#sec-
> number.issafeinteger
>
> a range with desirable properties wrt equational reasoning.
>
> So, with the above, if you have an attribute like
>
>   [EnforceRange] attribute long long tvisyn;
>
> an implementation of the ES binding will enforce the WebIDL "long long"
> range, but ES(6) code will have to perform a second check of the range if
> concerned about restricting this to the safe range.
>
> How about aligning the WebIDL IDL range with ES6 & unambiguously
> representable integers?
>
> (Apologies if this has been resolved already or there's an open issue on
> it; couldn't find any.)
>
> --sigbjorn
>
> * - a fairly recent change,
>
>
> http://dev.w3.org/cvsweb/2006/webapi/WebIDL/v1.xml.diff?r1=
> 1.74;r2=1.75;f=h
>
> (it was the "safe" range prior to that.)
>
>
>


-- 
    Cheers,
    --MarkM