Su-Laine,

Looking at index-sort-as and the current design of <glossentry>, I think that the most direct solution would be:

1. Define a specialization of <data> named something like "sort-as" whose value or content specifies the sort string, with same semantic as index-sort-as (specifies the sort key and combines with the nominal sort key of the element to be sorted to form a distinct matching key such that two items with the same base key but different sort-as values would be treated as distinct items where they would otherwise be merged, if merging is applicable).

The notion of "nominal sort key" would need to be defined, i.e., as the string that would be used to sort the element if sort-as were not specified. This will of course be different for different element types and may be application-defined, but obvious sort keys would be topic titles, definition list terms, table cells within specific table rows, and so on. We would have to decide which of title, navtitle, and search title would be used as the base sort key for topics when two or more are specified for the same topic, but I would think navtitle would be the most appropriate choice.

The <sort-as> element is intended to be a direct child of the element that contains the sort key, e.g., <title>, <dt>, etc., as opposed to being used in topic metadata.

<sort-as> would be a separate domain within the base DITA vocabulary set.

2. Define a specialization of <sort-as>, <glossary-sort-as>, for specific use within <glossterm>, e.g.:

<glossentry id="entry-01"> <glossterm> <glossary-sort-as>eight</glossary-sort-as> <text>8-Mile</text> </glossterm> <glossdef> <p>A famous road in Detroit.</p> </glossentry>

<glossary-sort-as> would be an addition to the glossentry topic type.

This approach would satisfy both your direct requirement to make glossary sorting both possible and the markup as obvious as possible and satisfy my more general requirement for a general sorting control.

Cheers,