Re: Undefined psql variables

Hello Tom,

> I'm inclined to suggest that we should require all extensions beyond the
> boolean-literal case to be set up as a keyword followed by appropriate
> argument(s); that seems like it's enough to prevent syntax conflicts from
> future additions. So you could imagine
>
> \if defined varname
> \if sql boolean expression to send to server
> \if compare value operator value
>
> It would be easy to allow "not" in front of any one of these, but
> it's less clear how to do AND or OR combinations. You can always
> fake AND with nested \if's, but OR is a bit more of a problem.
> Maybe we don't need it.
>
> Other ideas about how to design this?

My 0.02 €:

I have convinced myself that, unlike pgbench, psql does not really need an
advanced client-side-implemented language, so the smaller the better. What
I mean by this is that from psql point of view it is ok that the actual
expression evaluation is performed server-side. From a user experience
point of view it would look similar to pgbench, just the evaluator does
not need to be client-side.

So I would suggest something close but maybe simpler than what you suggest
above. If there is just one thing, it is true or false, checked client
side, well, this is already implemented:-).

\if something

If there are more than one argument, or maybe if previous true/false
evaluation failed, then:

\if sql expression to be evaluated server side

Then the result is checked for true or false client-side. It would be
equivalent to:

SELECT sql expression to be evaluted server side AS is_ok \gset
\if :is_ok

Finally I would suggest that client to server would only communicate by
variable substitution, as the backtick patch with external processes.

For checking variable definition, I would suggest to extend the variable
access syntax so that there is no exception to the one thing rule between
client side and server side evaluation:

\if :?variable

the :?... is subsituted by true or false depending on whether the variable
exists.

\if NOT :?variable

would work by executing "NOT ..." on the server. No need for "defined"
which would not look like SQL function calls anyway, no need for any
operator client side or clumsy rules.

--
Fabien.