Current implementation uses a single sequence "rails_sequence" to generate the id values for all tables. Comments in the code indicate this was done because the author couldn't determine a way to get table-specific sequences.
I cloned what's done in the postgres adapter, to achieve what's generally used as a standard in the Oracle world, which is to name the sequence after the table.
NOTE: this change is NOT backwards-compatible. However, it's easy to make work; just create a sequence for each table.
On the mailing list John Higgins had an idea about a more flexible approach, in which a regexp pattern is used to describe the name of the sequences. But that's beyond me at this point, and I think this patch is at least an improvement.
