Tom Lane wrote:
I'm a bit dubious that the prepared-statement machinery actually results in a win compared to just issuing simple Query messages.
The driver has no infrastructure for issuing simple Query messages in v3 mode, and anyway we want to batch the BEGIN up with the query that follows it.
COMMIT/ROLLBACK might be of dubious value but I don't know if it's worth having a completely separate code path in the driver just for that one case.
-O