Re: [XHTML2] CITELANG, TITLELANG attributes

> Specifications should define what UAs should do in _any_ scenario. The
> CSS, XML, and SVG specs are quite well defined in that regard. The HTML
> specs have traditionally been quite vague in that area.

There is a perfectly valid case for specifications to leave things
implemenation dependent, although they must be clear when things are
implementation dependent.  This allows innovation and optimisation
in areas that are not fundamental to the standard.

HTML, for example, should leave most, if not all, rendering issues
implementation dependent.  The demand for it to do otherwise often result
from a misunderstanding that HTML includes CSS, W3C and proprietary DOMs,
HTTP and scripting languages and is a total specification for a WYSIWYG
viewing platform.

Other examples of deliberate underspecification are that both the deflate
compression algorithm and MPEG specify how to perform the decode process,
but leave encoding implementation dependent, with only the constraint
that the result of applying encoding and then decoding steop, in the
first case, should be the same as the original, and in the second case,
should be close to the original.

There is also a long tradition of leaving certain features of programming
languages deliberately underspecified, so that they can easily be 
implemented on multiple types of hardware and permit extensive optimisation.
The old FORTRAN specification even had a second level definitions section
which was all about what you couldn't assume.

Received on Friday, 30 July 2004 01:49:01 UTC