Re: [xml-dev] XML is text-only ... why?

["Stephen D. Williams" <sdw@lig.net>]

To be clearer than I was:
Stephen D. Williams wrote:
> Fast Infoset is not what W3C Efficient XML Interchange (EXI) is based 
> on, although it is one of the candidates considered in detail, along 
> with my Efficiency Structured XML 1.0 (ESXML).
>
> Here is our first public working draft of the EXI format:
> http://www.w3.org/TR/2007/WD-exi-20070716/
>
> I feel it is an incomplete baseline in a number of ways, however it 
> includes most of the base structure and encoding method and is a 
> significant milestone.
The key considered candidate turned out to be "Efficient XML" from 
AgileDelta which, along with contributions from at least one other 
format (which doesn't seem to be publicly noted at the moment), is the 
basis for the EXI format specification.  I think of this format 
specification as a baseline that can be improved, but it may be that it 
is complete enough and further changes aren't worth the complexity or 
other costs. 

"AgileDelta's Efficient XML" (EFX) product, and therefore "W3C's 
Efficient XML Interchange" (EXI) specification, has a clever 
reformulation of the kinds of ideas used in both older bit-wise encoding 
formats (MPEG4's hand-optimized, but static, scene description encoding 
comes to mind, along others) and a number of dynamic encoding methods.  
This improved reformulation directly addresses the flexibility and 
architecture of XML while closely taking into account exactly what 
information is known while encoding and decoding.  Additionally, some 
interesting scalar encoding methods are used when typed values are 
enabled.  I think it can be improved, and should be improved for 
something that is likely to be used very widely.  Already though it 
leads as a solution to a number of related general problems compared to 
anything else.  Other candidates overlapped in some ways and 
investigated other aspects of the problem set to a greater degree, which 
may lead to additional useful capabilities.  The first thing to get 
right is the structure and basic methods; this is what the current 
specification does.
> I initiated the OpenEXI open source implementation and Santiago 
> Pericas-Geertsen and others quickly joined.  OpenEXI is hosted as an 
> open source project by Sun, although I don't believe the project 
> codebase is public quite yet.
OpenEXI is either a project independent of the W3C EXI working group or 
only loosely associated with it, depending on how you look at it.  The 
chartered output of the EXI working group is the specification and 
supporting documents.  OpenEXI is an independent open implementation of 
the specification and of experimental additions to it.  There are a 
number of large and small participant companies and individuals in 
OpenEXI and much progress has been made.  Suggesting it was inevitable 
by someone in the group and it is separate from the W3C XML Binary 
Characterization (XBC) and EXI working groups, except for the overlap of 
participants and the benefit to the work of the group in having an open 
implementation. 

We have as a group put in several years of hard work diligently 
formulating and answering the need for a nearly universal and flexibly 
efficient representation of mostly arbitrary data that is 
feature-compatible with XML and competitive with nearly all other 
solutions.  The path is difficult and it is a sign of health that there 
are differences of opinion in areas we haven't reached complete 
consensus.  It is important to avoid adding endless features which 
create complexity, code size, and interoperability issues, but at the 
same time it is just as important to avoid freezing a solution at the 
first sign of success when additional performance and coverage of 
additional use cases can be had for incremental work.

Another big part of the benefit of the working groups has been a 
crystallization of related ideas, including use cases, requirements, 
measurement methodology, and ways of comparing and analyzing different 
methods and degrees of encoding data.  While the documentation may be 
large enough to be hard to digest, some ideas should become widely used 
such as thinking of how much is "externalized" in a particular 
representation and how.

sdw

-- 
swilliams@hpti.com http://www.hpti.com Per: sdw@lig.net http://sdw.st
Stephen D. Williams 703-371-9362C 703-995-0407Fax 20147 AIM: sdw