Re: [xml-dev] Interoperability

[Joe English <jenglish@flightlab.com>]

Tim Bray wrote:

> As to Joe's problems above, I tend to think that one of
> the best uses of a DTD is for documentation.  I find them
> laughably inadequate for some parts of business validation
> and real overkill for others - I just don't find much
> use for them at interchange time.

Same here, except that for most of my applications DTDs are
neither overkill nor inadequate; they're usually just right.
But for document processing and interchange, the value added
(mostly whitespace significance, default attribute values,
and indicating ID-ness) isn't worth the administrative overhead.

Earlier Sean quoted me as saying that I had given up
on DTDs.  I haven't at all; what I *have* given up on
is <!DOCTYPE ...> declarations.

> Having said all that,
> the spec is clear, you have to provide a SYSTEM identifier
> for your DTD, i.e. a URL... why is this hard?  -Tim

URLs can break in too many ways.  TMML users would be
rightfully upset if they couldn't process their documentation
because the SourceForge web site was down for maintenance,
or if their network connection got unplugged.  Document-relative
SYSTEM identifiers are a maintenance nightmare; you have to
keep too many copies of the public text around.  PUBLIC
identifier resolution would be great if it were universally
supported, but it isn't.  "SGML on the Web" is tremendously
useful, but so is XML *off* the Web, and too many XML parsers
fail to account for that.

Most importantly: it's very difficult to ensure that the URI
for a DTD will last as long as the documents that reference it.

Remember the time that Netscape reorganized their web site and
all the headline aggregators on the Web suddenly stopped working?
This sort of thing is an unacceptable risk.


--Joe English

  jenglish@flightlab.com