Re: [xml-dev] Does the W3C allow "reference=?UTF-8?Q?implementations=22=

[rjelliffe <rjelliffe@allette.com.au>]

 On Wed, 24 Nov 2010 14:11:05 +0000, David Carlisle <davidc@nag.co.uk> 
 wrote:
> On 24/11/2010 13:29, Costello, Roger L. wrote:

>> What is the rationale for not allowing reference implementations?

> Of course that happens anyway with pdf implementations mimicking
> adobe acrobat reader, even where it doesn't follow the spec, and html
> implementations mimicking IE. Declaring acrobat (or IE) a "reference"
> implementation would just encourage such reverse engineering.

 The larger the technology is, the more chance it is based on some 
 existing deployed systems. So the purpose of standardization there is to 
 expose, consolidate and modernize probably in that order. Adopting a 
 reference implementation for a large standard means privileging a 
 particular product, which you don't want to do if there are competitors 
 and you don't want to do if there is a market-dominator either.

 In those cases you have, in effect, reference *sources* (one or many) 
 rather than implementations. An implementation of a standard may be a 
 primary reference source, but also a bad implementation if the standard 
 has cleaned up some area.

 Take the value of the ISO Linux ABI: some versions of Linux, under the 
 influence of the UNIX ABI standards, are the reference *source* of the 
 ISO standard. And if the standardization kept true to the source, Linux 
 would conform to the ISO standard, of course.

 To contrast, consider PDF. Adobe's products were the source, but 
 because ISO PDF is a series of profiles for different markets, you 
 wouldn't want Acrobat's capabilities to be the reference implementation, 
 for example. The point of the standardization exercise is to get more 
 voices in on selecting the profile and what information is necessary in 
 order to judge and possibly use the technology.

 On the issue of whether ISO/IEC allows reference implementations, it is 
 enough to point out that there are no procedures in place to support 
 reference implementations in the ISO/IEC directives. (There is slim 
 material about what I call reference sources, such as how to handle 
 product names in standards etc.)  So it is quite possible that some 
 committee uses some software as its primary source, and keeps it up to 
 date, and tests its tests suites against it, and even speaks of a 
 'reference implementation." However, AFAICS this has no official status 
 at ISO: there is just no slot for it.

 Nowadays, most important standards are co-branded with other consortia: 
 this gives plenty of scope for the standards committee wearing their 
 consortium hats to put out a reference implementation (if allowed).

 However, most standards bodies do not have the resources to maintain a 
 reference implementation in the long term. Just as most ongoing FOSS 
 projects don't have the resources to work with standardization groups. 
 Rather than thinking "do standards need reference implementations" it 
 might be better to consider "do FOSS developers need extra support to 
 help them take advantage of standards bodies?"

 Finally, I think that reference implementations are most useful when 
 the technology is quite small and layered: IETF RFC-sized in particular.

 Cheers
 Rick