> On Sat, May 01, 2004 at 10:16:56PM -0000, Greg Sabino Mullane wrote:
>> I am very uneasy about this. Statements should stay invalidated, else
>> the prepared statement may no longer even do what was originally
>> intended when it was first created.
I think Greg's concern is overblown, and would result in an effectively
unusable facility if we did implement it that way. As an example,
adding an index to a table *should* result in flushing of cached plans
for that table (maybe the query can make good use of the new index),
but people would be quite unhappy if already-prepared statements for the
table just stopped working. More generally, the flush mechanism will
probably be written in a way that causes plan flushes to occur for
events that aren't even user-visible --- such as VACUUM FULL relocating
the catalog tuple that describes an object used by the plan.
If we re-parse from source then we will detect any changes that make the
query visibly incorrect. I don't really see that the user can have any
beef if he continues to use a prepared statement whose source text would
have a valid but incompatible meaning due to changes elsewhere.
regards, tom lane
