2017-02-03 11:18 GMT+01:00 Fabien COELHO <coe...@cri.ensmp.fr>:
> We can implement XA support for variables, ale I don't think so default
>> should be XA.
> I was answering your question, which is what you can do about the
> feedback: take the one hard/strong point into account in your proposal.
> You do not want to do that. Too bad.
> The argument that you keep on repeating about "other software do it like
> that so it is the good way" do not work because these software (Oracle,
> DB2, ...) have features unavailable to postgres which mitigate the issue
> I'm raising, and there is no such mitigation in postgres.
> Note that you can proceed and simply ignore my negative opinion, which
> will stay negative till these "secure" variables are transactional by
> default, or till nested/autonomous transactions are provided by postgres.
I'll work on my proposal in v11 time. Maybe in this time Postgres will
support autonomous transactions.
The variables syntax should be better integrated to core - it should be
implemented without getter/setter functions. I am not sure If statement SET
can be enhanced to allows the work with session variables without some
conflicts, but we will see.