Jeff Davis <pgsql@j-davis.com> wrote:
> (aside: maybe we can implement ASSERT now using SSI; but it would
> require all transactions to be in SSI mode).
I see all kinds of cool things which could build on the SSI
technology or the predicate locks developed to support it, but I've
been trying to hold back on such discussions as much as possible
until 9.1 is stabilized. I just can't let this comment go without
observing that many of the cool uses would be a lot easier if one
could set a GUC which would require all transactions to be
serializable. Now isn't the time to discuss what it would be called
or what its settings would be (much less what corner cases might be
problems with such a restriction), but assuming it is possible does
facilitate daydreaming about such things as SQL ASSERT constraints.
;-)
-Kevin
