On Sep 3, 2005, at 03:32, Robin Berjon wrote:

> Henri Sivonen wrote:
>> I think specifying character entities in a DTD and expecting them to 
>> be supported is a serious spec but in XHTML 1.x. Carrying this over 
>> to XHTML 2.0 means that the HTML WG is in denial about the 
>> relationship of DTDs and interactive user agents and wants to wish 
>> the consequences of the XML spec away.
>
> To be fair, there's nothing forbidding a browser from recognizing the 
> public id used in the doctype and mapping that to a local copyy of the 
> DTD,

Right--if it is the full DTD. But that's a problem with 
backwards-compatible changes to the language without retroactive 
catalog changes to the UAs. Anyway, loading the full DTD even from a 
local catalog has a performance impact considering the sheer size of 
the DTDs (which is why Mozilla uses a short bogo-DTD).

> or even better to internal application knowledge of that DTD which may 
> be needed (such as entity definitions). A lot of SVG 1.1 
> implementations do that (for some defaulted attributes, not for 
> entities, but it's all the same) with no impact whatsoever on 
> performance.

I think pretending that DTD-based infoset augmentation happens when the 
DTD is not actually processed is worse than honestly not reading the 
DTD and then codifying the default behavior in the absence of 
particular attributes on the app layer (which you have to do anyway in 
order to support DTDless docs).

-- 
Henri Sivonen
hsivonen@i...
http://hsivonen.iki.fi/