Re: [HACKERS] Re: Regress tests reveal *serious* psql bug

Bruce Momjian <pgman@candle.pha.pa.us> writes:
> I think we can live with requiring a variable name to start with an
> alphabetic or underscore.

>     SELECT a[1:2]

> is clear and

>     SELECT a[1:myvar]

> expands to SELECT a[1].

No go --- SELECT a[1:b] where b is a field name is a valid query
currently.

I don't really see exactly what the benefit is of assuming that
:foo should expand to nothing rather than :foo, in the absence
of an actual definition for :foo.  My feeling is that a safe solution
would be (a) psql doesn't do variables at all without an explicitly enabling     command line switch; and (b) if psql
sees:foo where foo is not defined, it spits out an     error message.
 
Accepting undeclared variables went out with Basic and Fortran;
why are we intent on re-inventing a concept that's so obviously
dangerous?

> In fact, I think it should be an error to reference a variable that is
> not defined.  This will catch accidental references too.  If you
> reference a variable that does not exist like :myvar, it passes the
> literal :myvar to the backend.

That's two different answers, not one... but I could live with either
one of them...
        regards, tom lane