Comment 14 for bug 1251394

Bill, I do think display field changes will happen considerably more frequently and more commonly than index/browse/facet field changes, but it will probably still be not terribly often in the grand scheme. So, you are right, that alone probably doesn't justify a separate structure.

I guess the opportunity I am feeling is that I believe display extraction code can be done very simply if we allow the XSL to do the heavy lifting, especially if we place limits on the resulting transformed structure. If an experimental effort to write such code proved successful, it would not only provide a possible path for fast display field propagation, but also inform ways we might tweak the existing extraction process for more general speed gains.

As I said to start, I am just opening the discussion and "testing the waters" on this idea. If there is enough interest and/or no clear opposition, I can produce some code (which would probably clarify things and generate more discussion) and see if this has any real legs, or is just a dead-end. Thoughts?

P.S. Yes, the 1k/second was just a quick and dirty "on the fly" extraction test to STDOUT, no writing of rows. I know that writing rows will slow that to some degree :)