James Le Cuirot <chewi@aura-online.co.uk> writes:
> hubert depesz lubaczewski <depesz@gmail.com> wrote:
>> Perhaps you can explain what is the functionality you want to
>> achieve, as I, for one, don't understand. Do you want transactions?
>> Or not?
> I want an implicit transaction around the whole script if no explicit
> transactions are present in the script. This is what psql's -c option
> does and this is also what the pg gem does because both use PQexec.
Well, psql -c (and also multiple commands in PQexec) have more odd
behaviors than that. Notably, any error causes abandonment of the *whole*
string, even if it's got begin/end commands in it.
Most people think that psql -c's behavior is a legacy thing that we ought
to change someday; if backwards compatibility weren't such a big deal
around here, we might've done it already. You can find multiple bugs in
the pgsql-bugs archives that amount to complaints that psql -c ought to
work more like reading from stdin or a file. I don't think it's a good
idea for you to try to arrange things so that untold amounts of code is
depending on that behavior never changing.
In short, I think it would be good to push back on the way Chef is doing
things now, not perpetuate a dependency on a legacy behavior.
regards, tom lane