[Openstreetmap] Re: [Openstreetmap-dev] OSM's Schema - moving it forwards.

[Ben Gimpert]

Cantrell and I were joking out-of-band about the infinite variations of
this thread, and so maybe this message will be my last set of (tired)
quips?

On Tue, Nov 29, 2005 at 09:31:02PM +0100, Lars Aronsson wrote:
> Tom Carden wrote:
> 
> > Nah, the main advantage of XML is that everyone says "oh, OK", 
> > but if you didn't use it they'd be saying "why aren't we using 
> > XML"?
> 
> Let's remember what the alternatives are. If a simple text file or 
> CSV (comma separated values) is enough, that is good.

But for most problems CSV *is* enough! Since the underlying data layer
is usually an RDBMS, why shoe-horn your data into into a tree if it's
gonna end up as tables and columns? You could preach to me about layers
of abstraction atop the data model, but in reality "column-y" data
models tend to trickle up through the whole set of application layers.

So should we get rid of RDBMS systems for something else? Should we free
the world of the overpriced, crufty, bug-ridden 800 gorrila that is
Oracle? Ahh, now we're talking. ;) Go Subversion.

> For more 
> complex data structures, if the alternative is to invent something 
> like the TIFF file format or the ISO 2709 used by libraries to 
> transfer bibliographic records, then XML is an enormous advantage.  
> For these old formats, you would either be restricted to use some 
> vendor's proprietary subroutine library (only available in Turbo 
> Pascal) or (if you're lucky) you could buy the format spec for 48 
> CHF from ISO.  Not to talk about all semi-proprietary 
> communication protocols we had before everything became TCP/IP.

I think you're confusing two distinct issues: Good standards must be
open, *and* incrementally developed.

Closed (proprietary) standards suck for obvious reasons. Even if they're
technically snazzy (SWF), they suck because of lock-in.

But in addition, good standards must be incrementally developed! This is
the aspect the XML marketroids want to ignore: For the same reasons you
can't design software ahead-of-time, waterfall style, you can't design a
data model ahead of time. Instead you write the simplest thing that
works for your software, then revamp your standard only when necessary.
Make this simple standard, simply-adapted transparent and open and
public so partners / clients / emus using your standard can adapt on
their side.

A sad axiom? Incrementally developed *closed* standards will always beat
waterfall developed *open* standards. See SWF, DOC, PDF, etc. Every
time.

> I think the current shift towards XML + Unicode is as important as 
> the shift towards ASCII in the 1960s.  According to Wikipedia, 
> ASCII was first launched in 1963.  And lower case letters were 
> added in the 1967 edition of the standard.  XML 1.0 is from 1998.

ASCII vs. EBCDIC was about internationalization (character sets). XML is
really about data models, with i18n grafted on after the fact, and
badly. Does anyone else find it sad and funny that every non-English XML
document starts with the same few ASCII English characters?

XML is almost a decade old, and most serious academics and hackers avoid
it like the plague. Now I admit this is very anecdotal, but it shouldn't
be ignored. Seriously: Why didn't the Apache group use proper XML for
their v2 configuration files? Because XML is a technical failure, and a
marketing success.

Sigh.

		Ben