Bruce Momjian <pgman@candle.pha.pa.us> writes:
> Should we add it for 8.2 and see if we get any problem reports?
No. I don't believe this can work without a far more invasive patch
than this is. To point out just one problem, what of cached plans in
plpgsql functions? Those can't be carried across a genuine connection
reset (permissions and search path are two reasons why not). And the
protocol issues are not something you can just ignore, because the
command does break reasonable driver-level expectations.
regards, tom lane