Re: [dita] Indexterm: page ranges

[Erik Hennum <ehennum@us.ibm.com>]

Hi, Chris and TC:

As is probably no surprise, I have some resistance to this statement:

> On the other hand, index entries can reflect completely orthogonal organizations.

In principle, I see the rationale for an indexable semantic unit that's a subset of a sequence -- for instance, a range of table rows or list items (as with the step ranges from Chris's spark plug example).

I don't, however, see the rationale for overlap of an indexable semantic unit across partial peer elements within the topic body.

An index term doesn't indicate the presence of a word or two at a location but, instead, a thought expressed by a complete phrase, paragraph, list, section, topic, or group of topics.

If a writer has a need to indicate that a thought starts in the middle of one section and ends in the middle of another section but doesn't include the whole sections, the content has a problem. A topic-oriented architecture requires clear units of meaning with streamlined discourse and straightforward organization. So, we would want to ask ourselves whether those sections have been defined cleanly and, if not, whether DITA ought to provide special features for content with structural problems.

For printed output, a writer would need to indicate a range only when the indexed unit of content spans more than two pages (if I recall the convention correctly). How often will we have a range of contiguous peer elements that meet the following criteria?

• The indexed subset of a list or table spans more than two pages
• The full list or table is so much longer than the indexed subset that the index term doesn't apply to the list or table taken as a whole.
• Breaking up the list or table into separate, semantically distinct lists or tables isn't possible.

Finally, I'd note that the range issue for an index term resembles the scope issue for a metadata property such as audience or platform.  Some markup languages or tools have used markers to indicate the start and end of a metadata property. That approach supports overlaps across semantic units and so on, providing lots of flexibility but resulting in spaghetti text.  By applying the basic approach of metadata scope to indexing ranges, we would get the same simplicity and also consistency.


That's my take, anyway,


Erik Hennum
ehennum@us.ibm.com