Author(s)Arnaud Le Hors, IBM, Joe Kesselman, IBM, Andy Clark, IBM, Ralf Pfeiffer, IBM
EntityReference models the XML &entityname; syntax, when used for
entities defined by the DOM. Entities hardcoded into XML, such as
character entities, should instead have been translated into text
by the code which generated the DOM tree.
An XML processor has the alternative of fully expanding Entities
into the normal document tree. If it does so, no EntityReference nodes
will appear.
Similarly, non-validating XML processors are not required to read
or process entity declarations made in the external subset or
declared in external parameter entities. Hence, some applications
may not make the replacement value available for Parsed Entities
of these types.
EntityReference behaves as a read-only node, and the children of
the EntityReference (which reflect those of the Entity, and should
also be read-only) give its replacement value, if any. They are
supposed to automagically stay in synch if the DocumentType is
updated with new values for the Entity.
The defined behavior makes efficient storage difficult for the DOM
implementor. We can't just look aside to the Entity's definition
in the DocumentType since those nodes have the wrong parent (unless
we can come up with a clever "imaginary parent" mechanism). We
must at least appear to clone those children... which raises the
issue of keeping the reference synchronized with its parent.
This leads me back to the "cached image of centrally defined data"
solution, much as I dislike it.
For now I have decided, since REC-DOM-Level-1-19980818 doesn't
cover this in much detail, that synchronization doesn't have to be
considered while the user is deep in the tree. That is, if you're
looking within one of the EntityReferennce's children and the Entity
changes, you won't be informed; instead, you will continue to access
the same object -- which may or may not still be part of the tree.
This is the same behavior that obtains elsewhere in the DOM if the
subtree you're looking at is deleted from its parent, so it's
acceptable here. (If it really bothers folks, we could set things
up so deleted subtrees are walked and marked invalid, but that's
not part of the DOM's defined behavior.)
As a result, only the EntityReference itself has to be aware of
changes in the Entity. And it can take advantage of the same
structure-change-monitoring code I implemented to support
DeepNodeList.
An XML processor has the alternative of fully expanding Entities into the normal document tree. If it does so, no EntityReference nodes will appear.
Similarly, non-validating XML processors are not required to read or process entity declarations made in the external subset or declared in external parameter entities. Hence, some applications may not make the replacement value available for Parsed Entities of these types.
EntityReference behaves as a read-only node, and the children of the EntityReference (which reflect those of the Entity, and should also be read-only) give its replacement value, if any. They are supposed to automagically stay in synch if the DocumentType is updated with new values for the Entity.
The defined behavior makes efficient storage difficult for the DOM implementor. We can't just look aside to the Entity's definition in the DocumentType since those nodes have the wrong parent (unless we can come up with a clever "imaginary parent" mechanism). We must at least appear to clone those children... which raises the issue of keeping the reference synchronized with its parent. This leads me back to the "cached image of centrally defined data" solution, much as I dislike it.
For now I have decided, since REC-DOM-Level-1-19980818 doesn't cover this in much detail, that synchronization doesn't have to be considered while the user is deep in the tree. That is, if you're looking within one of the EntityReferennce's children and the Entity changes, you won't be informed; instead, you will continue to access the same object -- which may or may not still be part of the tree. This is the same behavior that obtains elsewhere in the DOM if the subtree you're looking at is deleted from its parent, so it's acceptable here. (If it really bothers folks, we could set things up so deleted subtrees are walked and marked invalid, but that's not part of the DOM's defined behavior.)
As a result, only the EntityReference itself has to be aware of changes in the Entity. And it can take advantage of the same structure-change-monitoring code I implemented to support DeepNodeList.