- From: Tim Bray <tbray@textuality.com>
- Date: Thu, 26 Jun 1997 09:30:11 -0700
- To: w3c-sgml-wg@w3.org
The ERB met on June 25; everyone but Dave Hollander was present. There was considerable discussion of Parameter Entities, an unofficial summary follows: 1. The discussion in the WG makes it clear that XML's utility as an authoring environment would be severely compromised by omission of PE's, or even by constraining them much more than they are now. 2. It is generally agreed that implementation of the full suite of constraints on the placement of and replacement text for PE's is beyond what should be expected of a lightweight nonvalidating parser. 3. It would be dangerous to relax the constraints, i.e. say that PE references can go anywhere in the DTD, as this would tend to create a large legacy class of instances that would be well-formed but not 8879-conformant, and hard to make conformant. Thus it was unanimously agreed to have two sets of PE rules, one for the external and one for the internal subset. In the external subset, the rules will stand as they are now, although we'll try to improve the explanation in the spec. (Those who have said it could be explained more clearly are hereby invited to submit specific suggestions). In the internal subset, PEs must expand to match "markupdecl" (prod. 28 in the current draft), and references can only be placed where a markupdecl can be recognized. The feeling is that this level of recognition is well within the capabilities of the most modest parsers. In follow-on discussion, we realized that this highlights a weakness in the spec. Currently, it only dicusses validating and non-validating processors; the former are required to read the DTD. In fact, there are already non-validating parsers which *do* read and use the DTD, if only to extract default attributes and entity declarations. This seems like an obviously good thing to do. Yet, such processors are unlikely to want to fetch and retrieve the whole external subset, since if they're not validating they don't care about content models; it seems reasonable for there to be a common class of XML documents which group the markup declarations not required for validation, but useful to a processor, in the internal subset for efficient transfer down the wire. This is closely related to the question of the RMD; a conformant processor cannot refuse, at the moment, to read the external subset; should this be allowed in some class of nonvalidating parsers? And yet the phrase "some class" suggests the use of an option, something we have to date vigorously resisted introducing into XML. Solving this problem *may* be made easier by removing discussion of the processor entirely from the spec, as suggested by both Henry Thompson and Dan Conolly. We judged that this lack of clarity is not fatal to the progress toward a July 1st version of XML-lang (processors are de facto apparently doing what seems like the right thing) - but there will be a work item on our agenda later this year to address and clean up this area. Furthermore, the next release of the draft will contain an editorial note acknowledging the existence of this set of issues. Cheers, Tim Bray tbray@textuality.com http://www.textuality.com/ +1-604-708-9592
Received on Thursday, 26 June 1997 12:32:11 UTC