Processing interoperability considerations

Darwin Information Typing Architecture (DITA) Version 1.2

(Non-normative) The DITA specification does not require processors to perform filtering, content reference resolution, key space construction, and other processing related to base DITA semantics in any particular order. This means that different conforming DITA processors may produce different results for the same initial data set and filtering conditions. DITA users and DITA implementers need to be aware of these potential differences in behavior when DITA content will be processed by different processors.

For conditional elements, an element is "applicable" if it is filtered in by active filtering specification and "inapplicable" if it is filtered out.

In general, in any situation in which two elements interact during processing, applying filtering before or after the processing is done will result in different results when either or both of the elements is conditional.

Filtering and content reference resolution

When two elements are merged as result of a content reference, the attributes of the two elements are combined. By default, the attributes of the referencing element take precedence over the referenced element. However, any attribute may specify the value "-dita-use-conref-target", which causes the referenced element's attribute to take precedence. This means that the effective value of the select attributes may reflect either the referencing element or the referenced element depending on how each attribute is configured on the referencing element. This in turn means that, in certain cases, filtering before resolving content references will produce a materially different result than when filtering is applied after resolving content references.

In two cases, the order in which filtering is applied results in either an element being in the effective result or an element not being in the effective result. There is a third case in which there will be either an empty element (and unresolvable content reference) or no element.

In the case where a referenced element is not applicable and the referencing element is explicitly applicable for the same condition (that is, both elements specify values for the same classification domain and the referencing element is applicable), if content references are resolved before filtering, the content reference is resolved and the effective value of the referencing element reflects the referenced element. If content referencing is resolved after filtering, the referenced element is filtered out and the content reference cannot be resolved, typically generating an error.

If the referencing element is not conditional and the referenced element is inapplicable, filtering applied before content reference resolution results in an unresolvable content reference. If filtering is applied after content resolution, the explicit condition on the referenced element becomes the effective value for that condition following content resolution and the result is then filtered out. The difference in these two cases is that in the first case the content reference cannot be resolved, resulting in a processing error and a potentially non-sensical element if the referencing element has required subelements (for example, a content reference from a topic to another topic, where the referencing topic must have a title subelement), but in the second case the element is filtered completely out.

Filtering and key space resolution

DITA allows processors to apply filtering before or after key space construction. Key space construction is the process of determining which key definitions are effective within a given map tree. Given two definitions of the same key with different applicabilities, the effective key space will be different if filtering is applied before key space construction than if it is applied after. In particular, if filtering is applied after key space construction, the key space is invariant, but if filtering is applied before key space construction, the effective key space will differ depending on the active conditions.

As an implementation detail for key-space-constructing processors, if filtering is applied before constructing the key space, then the set of effective key definitions is simply the first definition of each unique key name within the map tree. However, if filtering is applied after key space construction, and in particular, if a processor needs to allow dynamic resolution of keys based on different filtering specifications applied to the same constructed key space, then the set of effective key definitions is the first definition of each pair of unique key name and unique selection specification set. This second form of constructed key space would be needed by processors such as editors and content management systems that need to quickly provide different filtering-specific key bindings without reconstructing the entire key space for each new set of filtering conditions.

For example, given a map that contains two definitions for the key "topic-01", one with an @audience value of "expert" and one with an @audience value of "novice", a filter-first processor would only have at most one effective key definition for the key name "topic-01", whichever of the two definitions was filtered in by the active filter specification and was the first definition encountered (if both happen to be filtered in). In a processor that supports dynamic key definition filtering, there would be two effective definitions for the key name "topic-01", one for @audience of "expert" and one for @audience of "novice". The processor would also need to maintain knowledge of the definition order of the two key definitions in order to correctly handle the case when both "expert" and "novice" are applicable for a given key access request (in which case, whichever of the two definitions was first in the map tree would be used as the effective value of the key).

Link resolution

If a cross reference, link, or other linking element is resolved to its target before filtering and the target is subsequently filtered out, the link would be to a non-existent target but might reflect properties of the target (e.g., a cross reference link text might reflect the target's title). If the link is resolved after filtering is applied and the target is filtered out, the link is to a non-existent target, which will result in a different link text. The rendition effect for the navigation link will be the same: the link cannot be navigated because the target doesn't exist in the rendered result.

Map metadata propagation

Filtering applied before propagation can result in different effective values than if it is applied after. In particular, elements filtered before out will never contribute to propagation.

Topicref resolution

Resolution of topicrefs before filtering can result in use of topic-provided navigation titles or metadata that would not be used if the target topic was filtered out before resolution. In both cases, the topicref as rendered would be to a missing topic.

Chunking

A topicref subsequently filtered out that generates chunks would create chunks in the output if chunk processing is done before filtering but since the topicref would then be filtered out, the chunks would not be referenced.

Copy-to processing

If copy-to processing is done before filtering, two topicrefs, only one of which is applicable, could specify the same copy-to target, leading to a conflict and a potential ambiguity about which governs. If the topicrefs are filtered before copy-to processing, the conflict does not occur.