Hi Mike,
It looks like this is due to a performance improvement put in 15.7 SP100 on little-endian platforms for indexes with large numbers of duplicate values. I don't find any sign anyone else has hit this issue yet during a downgrade, so there isn't an established workaround. We'll just have to work one out.
When an index is using this new feature, I believe sysindexes.status bit 0x100 would be on.
Does this query return any rows?
select object_name(id), name, status from sysindexes where status & 256 = 256
If so, does it make sense that some values of the keys would have a lot of duplicates?
If not, let me know; If so, I think we are on the right track and the thing to try would be dropping these indexes. At that point, you should be able to downgrade the database, dump it, and rebuild the indexes after you load it on the lower-versioned server
You could also drop the indexes, "set switch on 1558 with override" to disable the new feature when creating an index, and rebuild the index. You should then still be able to downgrade the database, and it would contain the indexes so they wouldn't have to be recreated later.
-bret
