Re: On Registries

I think you guys are maybe talking about two different things here.
There's a difference between assigning a uniform name for something and
giving it status as real part of the API.

On the one hand, Mike is correct that identifiers are cheap.  Vendor X
saying "we're reserving VndX-Foo for our proprietary Foo algorithm" has no
cost to the web platform (especially if you can create a special vendor
ghetto for these identifiers).  On the other hand, Ryan is correct that
it's very expensive to actually "officially" add algorithms to the API, in
the sense that web developers can rely on them being present across the web.

To make a registry analogy, the TLS ciphersuite registry is divided into
two chunks, one more liberal than the other.
"""
   -  TLS Cipher Suite Registry: Future values with the first byte in
      the range 0-191 (decimal) inclusive are assigned via Standards
      Action [RFC2434].  Values with the first byte in the range 192-254
      (decimal) are assigned via Specification Required [RFC2434].
"""

I doubt there's much disagreement here that "official" algorithms need to
go through the whole WG process, since they need broad consensus to ensure
broad implementation.  (That's the equivalent of the Standards Action
requirement for "official" ciphersuite values).

So the main question is whether we provide any facility for non-official
algorithms to reserve / de-conflict their identifiers.  In registry
language, this would apply a First Come First Served policy, or
Specification Required -- no consensus process involved.  On the one hand,
this could help experimentation be more orderly, at not much cost to
implementors or developers.  On the other hand, implementors in the web
have had no problem coming up with ad-hoc deconfliction schemes (e.g.,
moz_* and webkit_*).  And it has proven difficult in other contexts to
maintain the separation between "official" and "non-official" values, e.g.:
http://tools.ietf.org/html/rfc6648

Net of those considerations, I'm inclined to not take any action for the
"non-official" algorithms.  There just doesn't seem to be a need for
coordinating action here; the community has a good history of organizing
itself.

But hopefully we can at least agree that "official" algorithms need to go
through a consensus process.

--Richard






On Fri, Aug 8, 2014 at 1:23 AM, Mike Jones <Michael.Jones@microsoft.com>
wrote:

>  You’re missing that user agents don’t have to implement registered
> algorithms.  They’re free to exercise their own best judgment.  If they
> weren’t, then you’d be on the side of having required algorithms, which I
> know you’re not.
>
>
>
> In the same way that registering new JWA algorithms doesn’t hurt anything,
> registering new WebCrypto algorithms doesn’t hurt anything.  But it can
> help.  There’s nothing to fear there.
>
>
>
> *From:* Ryan Sleevi [mailto:sleevi@google.com]
> *Sent:* Thursday, August 07, 2014 10:18 PM
>
> *To:* Mike Jones
> *Cc:* Mark Watson; public-webcrypto@w3.org
> *Subject:* RE: On Registries
>
>
>
> Mike,
>
> If we want to talk hyperbole, suggesting the W3C is akin to fearing
> openness is right up there.
>
> Every algorithm in WebCrypto represents a distinct capability for the Web.
> Thus is, by definition, an API, no different than Canvas.getContext is an
> API (
>
> http://www.whatwg.org/specs/web-apps/current-work/multipage/scripting.html#the-canvas-element
> )
>
> What scares me is individual UAs arbitrarily and unilaterally adding new
> capabilities to the web, without discussion and consensus of other UAs,
> which you have been so gung-ho on for so long and so clearly, and which I
> can only hope is an individual view and not current corporate thinking.
>
> That is the ONLY technical purpose a registry of allowing arbitrary
> organizations to register new algorithms serves. And expert review does NOT
> cut it.
>
> The registration of JWA/JWK affects no other implementations, except for
> those that developers try to exchange messages with. It is, by definition,
> safe.
>
> Exposing a new capability affects every User Agent that ever hopes to
> visit that site, and thus should absolutely be treated with fear and
> trembling, because the Web does not remove capabilities lightly. Suggesting
> such a site would "only affect/support IE" or "Only affect/support Chrome",
> which is what such arbitrary capability extensions mean, is the very
> essence of a return to browser wars and extinction through interoperability
> failure.
>
> On Aug 7, 2014 9:55 PM, "Mike Jones" <Michael.Jones@microsoft.com> wrote:
>
> So why, exactly, was it OK for you to add algorithms to JWA and JWK with a
> registry, when you would deny the same ability to others for WebCrypto?
> Why does openness and cooperation scare you so?
>
>
>
> *From:* Ryan Sleevi [mailto:sleevi@google.com]
> *Sent:* Thursday, August 07, 2014 9:48 PM
> *To:* Mike Jones
> *Cc:* Mark Watson; public-webcrypto@w3.org
> *Subject:* RE: On Registries
>
>
>
>
> On Aug 7, 2014 9:38 PM, "Mike Jones" <Michael.Jones@microsoft.com> wrote:
> >
> > We’re not talking about adding APIs.
>
> Right, I thought we had agreement that this was an API.
>
> Without that agreement, I suspect there is nothing further of fruit to
> come of this discussion. The points for why it is, unquestionably, an API
> have been laid out.
>
> > We’re talking about algorithms.  That’s a much more restricted extension
> space than the hypothetical one that you’re pontificating about.
> >
> >
> >
> > Cryptographers are more qualified to extend that space than the W3C, the
> WHATWG, ECMA, or you or me.  Let’s enable them do it, irrespective of the
> organization in which they write their spec.
> >
>
> Cryptographers, as an abstract group/concept, are the LEAST qualified of
> the groups you mentioned to write browser APIs, to understand both the
> limitations and the idioms of the platform, of the risks and the
> guarantees, and of how the Web and Standards work.
>
> With all due respect to all parties, it's like suggesting someone who
> specializes in Computer Aided Design to design a C API. Very different
> roles and functions.
>
> If they do so, its no different than me writing a spec for cold fusion and
> hiding it under my pillow, because its just a piece of paper.
>
> It means nothing without the involvement of UAs, to implement, to agree on
> things like IPR, etc. Any UA that went and implemented such extensions
> ad-hoc would rightly be called out as breaking the web, embracing,
> extending, and extinguishing.
>
> I can only encourage you to reach out to your counterparts on the IE team
> and in WebApps to understand that none of our organizations (and yeah, Moz,
> you too ;D) have their hands clean in these matters, and that's precisely
> why we are trying so hard to work together to ensure it does not happen.
>
> A spec that blesses or condones such API extensions, rather than
> condemning, is a step in the wrong direction for standards.
>
> >
> >
> > From: Ryan Sleevi [mailto:sleevi@google.com]
> > Sent: Thursday, August 07, 2014 9:33 PM
> >
> > To: Mike Jones
> > Cc: Mark Watson; public-webcrypto@w3.org
> > Subject: RE: On Registries
> >
> >
> >
> > Mike,
> >
> > Can you name any organization or individual, outside the W3C, WHATWG,
> and ECMA qualified to extend the web by adding new APIs for UAs to
> implement?
> >
> > You can easily reach out to Travis L. and co on your side, or to your
> legal team, to better understand Microsoft's view against exactly that,
> with respect to participation and standardization that occurs in the
> WHATWG. Or you can look towards the involvement in TC39 to better
> understand that there is very much a split, in both organizations, as to
> which party is responsible for developing and standardizing which aspects.
> >
> > That the IETF, or any similar organization, would and can responsibly
> develop JavaScript APIs used by millions of developers, without the
> involvement of UAs, is not a reality. The core constituency is and has
> always been wherever the UAs are, since ultimately no standard, draft, or
> spec is ever meaningful until UAs sit down and implement it.
> >
> > So the only registry that matters is what UAs do. Everything else is a
> figment of a standards dream that doesn't exist, much like the debacle of
> XHTML that lead to the formation of the WHATWG to begin with.
> >
> > Put differently, interop is the only registry that matters. And the W3C
> (and WHATWG) is where interop happens. A registry doesn't make interop
> happen - it just documents the interop that already happened. So does a
> Wiki. Or a PubStatus. Or WebPlatform.org, MDN, MSDN, etc.
> >
> > On Aug 7, 2014 9:20 PM, "Mike Jones" <Michael.Jones@microsoft.com>
> wrote:
> >
> > Ryan, it seems that your primary motivation here is preventing things
> that could go wrong.  Mine is enabling things that can go right.
> >
> >
> >
> > I find it hypocritical that you’ll happily use the IANA Registry to
> extend JWK for use by WebCrypto in
> http://www.w3.org/TR/WebCryptoAPI/#iana-section but you’re unwilling to
> admit that enabling the IETF or others to similarly extend WebCrypto would
> likewise be a good thing.
> >
> >
> >
> > The IETF is happy to have others extend the Internet in a principled
> fashion.  Why are you so afraid to let the IETF or others extend the Web in
> a similarly principled fashion?
> >
> >
> >
> > JWK is stronger because WebCrypto can extend it.  It’s better for
> WebCrypto, better for JOSE, and better for the Internet.  WebCrypto and the
> Web would be similarly stronger if others outside the W3C could extend it.
> So let’s make it happen!
> >
> >
> >
> > I’ll be happy continue to advocate that it’s more important to WebCrypto
> and the W3C to enable responsible extensions by all, even though it may
> scare you, than to maintain a closed spec and closed process that only one
> organization can extend.  I believe that openness in this respect is “on
> the right side of history”.
> >
> >
> >
> >                                                             -- Mike
> >
> >
> >
> > From: Ryan Sleevi [mailto:sleevi@google.com]
> > Sent: Thursday, August 07, 2014 8:12 PM
> > To: Mike Jones
> > Cc: Mark Watson; public-webcrypto@w3.org
> > Subject: RE: On Registries
> >
> >
> >
> >
> > On Aug 7, 2014 7:56 PM, "Mike Jones" <Michael.Jones@microsoft.com>
> wrote:
> > >
> > > Replies about the W3C’s positive role in ensuring quality of algorithm
> registry entries inline at the end of this message…
> > >
> > >
> > >
> > > From: Ryan Sleevi [mailto:sleevi@google.com]
> > > Sent: Thursday, August 07, 2014 7:44 PM
> > > To: Mike Jones
> > > Cc: Mark Watson; public-webcrypto@w3.org
> > > Subject: RE: On Registries
> > >
> > >
> > >
> > > On Aug 7, 2014 7:31 PM, "Mike Jones" <Michael.Jones@microsoft.com>
> wrote:
> > > >
> > > > Thanks for your insightful reply, Mark.  A few comments inline below…
> > > >
> > > > From: Mark Watson [mailto:watsonm@netflix.com]
> > > > Sent: Thursday, August 07, 2014 6:02 PM
> > > > To: Mike Jones
> > > > Cc: Ryan Sleevi; public-webcrypto@w3.org
> > > > Subject: Re: On Registries
> > > >
> > > > On Thursday, August 7, 2014, Mike Jones <Michael.Jones@microsoft.com>
> wrote:
> > > >
> > > > Simple.  In the first case, the algorithm is a data value.  In the
> second case, it’s encoded in an API.  Data values are easily extensible.
> APIs are not.  That’s why extending the space of algorithms by registering
> new data values makes a world of sense.  Expending the algorithms by adding
> new APIs for each would be clunky, procedurally slow, and mostly unworkable.
> > > >
> > > > I think what Ryan is saying is that it should be no easier to add an
> algorithm than it is to add a new API (or, more strongly, that a new
> algorithm *is* a new API and _therefore_ should be no easier to add).
> > > >
> > > > I believe you’ve accurately identified the heart of the disagreement
> here, Mark.
> > > >
> > > > IF we decided that it should be easier than this to add new
> algorithms and especially if we decided that groups other than W3C Working
> Groups should be able to do so, then a registry makes sense as a mechanism
> to coordinate that.
> > > >
> > > > Agreed.
> > > >
> > > > Otherwise (which is where we are now), then the definitive list of
> algorithms is to be found in the sum total of the output of the W3C
> WebCrypto Working Group and nowhere else.
> > > >
> > > > If we decide that he definitive list of algorithms is only to be
> produced by the W3C WebCrypto Working Group, I believe that would be a
> significant missed opportunity.  The WebCrypto API is an exercise in
> packaging algorithms developed by cryptographers for use by Web developers,
> just like JOSE is.  Neither working group’s primary expertise is
> cryptography.  Cryptographers should be the ones to write the extensions
> specs defining new algorithms – not us.  Some of those may occur in the W3C
> but some may occur in the IETF and some may be individual drafts by people
> such as Dan Bernstein, David McGrew, and Brian LaMacchia.
> > > >
> > > > We would be doing the WebCrypto API and the Web a significant
> disservice if we don’t enable people other than us to define and register
> new algorithms for use with WebCrypto.  We should be humble enough to
> recognize that defining new crypto algorithms is not our expertise and let
> those who are experts define them for use with our spec, no matter where
> they choose to do the work.
> > > >
> > >
> > > I agree with the sentiment that anyone should be able to write
> definitions for algorithms, and am excited to see Trevor's Curve25519 draft.
> > >
> > > I disagree with the sentiment that it should happen outside the W3C.
> To do so is to return to the browser wars, where both Microsoft and
> Mozilla, though well motivated, wrecked great harm through "embrace,
> extend, extinguish" and the introduction at large of new vendor-specific
> APIs, often without specs (or without free licensing, or with great patent
> encumbrance, or through active hostility towards other UAs efforts to
> interop)
> > >
> > >
> > >
> > > The W3C (and the WHATWG) exist to help prevent that terrible harm from
> ever happening again. The way to do that is by having multiple UAs
> coordinate and ship features responsibly, to agree on specifications, and
> to avoid vendor lock-in.
> > >
> > > Regardless of this group's cryptographic expertise, which i agree is
> unfortunately lacking, we are filled with UA implementors, the sole
> entities with the power to make - or break - the web; For developers, for
> other UA implementors, and most importantly, for users, for this generation
> of the web and those to come. For that, there can and should be no
> alternative - we must agree, as UAs, and the W3C exists precisely to
> support and guide that agreement.
> > >
> > >
> > >
> > >
> > >
> > > Suggesting that using a responsibly managed registry would be a return
> to the “browser wars” or that features would be shipped without
> specifications is hyperbole.  I’m only advocating a specification-required
> registry with expert review.  The W3C would appoint the appropriate experts
> to ensure that the specifications registering algorithms are clear and
> well-written and meet any other criteria decided by the W3C.  The W3C can
> ensure the quality of registered algorithms without having to write all the
> drafts itself.  It’s unnecessary and detrimental hubris to think that we’re
> the only ones qualified to do so.
> > >
> > >
> > >
> > >                                                             -- Mike
> >
> > Expert Review is exactly the detrimental sort of stuff that got us into
> this.
> >
> > The W3C REC track is exactly that. A path for _anyone_ to write specs,
> submit them to the WG, for UAs and users to agree in interest, and to
> develop and publish such specifications with clear views on patents,
> interoperability, and applicability to the web.
> >
> > No single expert is qualified to reflect consensus of a
> multi-stakeholder UA community, which is a fundamental and nonnegotiable
> portion of any Web API.
> >
> > So if we talk a panel of experts, who are they? Well, we probably want
> some UAs, those are essential. We probably need some cryptographers, since
> security is key. We should hear from developers, since this matters to
> them. And we should probably have a few representatives from the public.
> >
> > And suddenly, you have an Expert Panel that is indistinguishable in form
> or function from a WG. And without having to reinvent a process for that
> review.
> >
> > Any solution that fails to go through the W3C/WHATWG process is, in
> spirit and effect, a return to the browser wars - including the designation
> of sole experts (which the W3C also has readily available if this WG should
> ever need - the TAG and AB)
>