Folks,
I'd like to implement updateable cursors. I'll be working on just
getting updateable cursors working for relatively simple SELECT queries
(e.g. no joins, aggregates, grouping, user-defined function calls,
etc.). BTW, I believe that's all the SQL spec requires, but I need to
double check that. I'm also planning on only implementing only
INSENSITIVE cursors, and not allowing holdable cursors.
However, I'm a little unsure how tuple visibility should work with
updateable cursors. In particular:
- if the user updates a row X in the cursor, then rewinds the cursor and
fetches X again, should they see the new X or the old X?
- if the user updates a row X in the cursor, and then a query within the
cursor's transaction views X, should the query see new X or old X?
Any comments?
Neil