Well ... we don't run with 9i yet (should that be on our must-have list for 4.7?).
The idea behind not loading the queries was that Oracle and/or PG (mostly PG) might screw us by breaking backwards compatibility, meaning we'd need to provide separate queries for the two different versions.
In practice we've ended up dropping support for older PG versions other than the most-recently-obsolete one and haven't run into backwards compatibilies that have required us make use of the query dispatcher's support of versioned queries.
That was the thinking.
I'm curious to see what others think. What if we threw an error if a query was actually executed that has blank SQL? That would catch cases you're not going to catch at query file load time (i.e. when there's no query at all because someone forgot to add their query files to the CVS tree ...)
