Re: [cssom-view] colorDepth/pixelDepth, match implementations or theoretical purity?

On Tue, Jun 18, 2013 at 6:25 PM, Simon Pieters <simonp@opera.com> wrote:

> In https://www.w3.org/Bugs/**Public/show_bug.cgi?id=17522<https://www.w3.org/Bugs/Public/show_bug.cgi?id=17522>we have a situation where all implementations and the spec agree on one
> thing, but that thing is wrong according to the industry standard terms. It
> is obvious that Glenn and I have different ideas about what the spec should
> say. In HTML, the policy in situations like these have almost always been
> to just match the implementations.
>

Not really. CSS3 and HTML5 specify hundreds of small changes that require
some implementation changes. The prior implemented support for pixelDepth
and colorDepth is nothing if not broken. We should fix it just like we are
fixing CSS Syntax and error recovery and myriads of other vaguely or
misspecified behavior.


>
> There can be reasonable exceptions where we would want to not match
> implementations despite interoperability, like if a security problem is
> identified or if a different definition would be vastly superior *and* the
> feature is not widely used in the Web such that changing it does not cause
> compat problems.
>
> In the case of colorDepth and pixelDepth, the proposed change does not
> address any security problem. It would start to expose the number of bits
> in the alpha channel (also see https://www.w3.org/Bugs/**
> Public/show_bug.cgi?id=14072<https://www.w3.org/Bugs/Public/show_bug.cgi?id=14072>
> ).
>
> The attributes can be used for fingerprinting. The proposed change does
> not change that, in fact it probably increases the fingerprinting. If we
> want to remove this fingerprinting vector, the attributes can be made to
> return a static value, probably 24. It seems at least some current
> implementations do not return a static value.
>
> I'd like to hear from other people what they think. Do implementers want
> to change these to match the industry terms? Or remove the fingerprinting
> vector? Or leave as is?
>
> --
> Simon Pieters
> Opera Software
>
>
>