On 18/09/2019 08:58, Peter Bex wrote:
> Perhaps we should treat databases that don't support this as second-class
> and raise an exception when you try to roll back a nested transaction?
> That way, the "happy path" will always work.  Not sure this is great
> design, though.  Raising an exception when trying to make a nested
> transaction could also work, but that means some library code is
> completely unusable in certain databases (how many databases don't have
> these, anyway?).

I should have caught up on the thread before replying :-D

--
Alaric Snell-Pym   (M7KIT)
http://www.snell-pym.org.uk/alaric/