previous   next   contents   index

25 February, 2000

3. Constraints imposed upon DOM

Table of contents

We have found that the DOM has utility in a number of scenarios, and that these scenarios have differing requirements and constraints. In particular, we find that editing application scenarios require specific support that the browser or runtime environment typically does not. We have identified the following requirements that are directly associated with support for editing application scenarios as distinct from runtime or playback scenarios.

3.1. Document modality

Due to the time-varying behavior of SMIL and SMIL-integrated document types, we need to be able to impose different constraints upon the model depending upon whether the environment is editing or browsing/playing back. As such, we need to introduce the notion of modality to the DOM (and perhaps more generally to XML documents). We need a means of defining modes, of associating a mode with a document, and of querying the current document mode.

We are still considering the details, but it has been proposed to specify an active mode that is most commonly associated with browsers, and a non-active or editing mode that would be associated with an editing tool when the author is manipulating the document structure.

3.2. Node locking

Associated with the requirement for modality is a need to represent a lock or read-only qualification on various elements and attributes, dependent upon the current document mode.

For an example that illustrates this need within the SMIL DOM: To simplify runtime engines, we want to disallow certain changes to the timing structure in an active document mode (e.g. to preclude certain structural changes or to make some properties read-only). However when editing the document, we do not want to impose these restrictions. It is a natural requirement of editing that the document structure and properties be mutable. We would like to represent this explicitly in the DOM specification.

There is currently some precedent for this in HTML browsers. E.g. within Microsoft Internet Explorer, some element structures (such as tables) cannot be manipulated while they are being parsed. Also, many script authors implicitly define a "loading" modality by associating script with the document.onLoad event. While this mechanism serves authors well, it nevertheless underscores the need for a generalized model for document modality.

(ED: The node locking could be currently supported with the DOMException NO_MODIFICATION_ALLOWED_ERR. )

3.3. Grouped, atomic changes

A related requirement to modality support is the need for a simplified transaction model for the DOM. This would allow us to make a set of logically grouped manipulations to the DOM, deferring all mutation events and related notification until the atomic group is completed. We specifically do not foresee the need for a DBMS-style transaction model that includes rollback and advanced transaction functionality. We are prepared to specify a simplified model for the atomic changes. For example, if any error occurs at a step in an atomic change group, the atomicity can be broken at that point.

As an example of our related requirements, we will require support to optimize the propagation of changes to the time-graph modeled by the DOM. A typical operation when editing a timeline shortens one element of a timeline by trimming material from the beginning of the element. The associated changes to the DOM require two steps:

Typically, a timing engine will maintain a cache of the global begin and end times for the elements in the timeline. These caches are updated when a time that they depend on changes. In the above scenario, if the timeline represents a long sequence of elements, the first change will propagate to the whole chain of time-dependents and recalculate the cache times for all these elements. The second change will then propagate, recalculating the cache times again, and restoring them to the previous value. If the two operations could be grouped as an atomic change, deferring the change notice, the cache mechanism will see no effective change to the end time of the original element, and so no cache update will be required. This can have a significant impact on the performance of an application.

When manipulating the DOM for a timed multimedia presentation, the efficiency and robustness of the model will be greatly enhanced if there is a means of grouping related changes and the resulting event propagation into an atomic change.


previous   next   contents   index