On Sunday 19 July 2009 14:39:33 Martijn van Oosterhout wrote:
> On Sun, Jul 19, 2009 at 03:15:38AM +0200, Andres Freund wrote:
> > Hi Robert, Hi All,
> >
> > Patch applies with some offset changes, code changes look sensible, I
> > personally like the new syntax and the features it may allow in future.
> > One, possibly big, gripe remains though:
> > The formerly valid statement which cannot be written without the
> > parentheses and stay semantically equivalent:
> > EXPLAIN (SELECT 1 ORDER BY 1) UNION ALL (SELECT 2 ORDER BY 1);
> > is now not valid anymore (The added %prec UMINUS causes the first '(' to
> > be recognize as start of the option list as intended).
> > This currently can only be resolved by using an option list like:
> > EXPLAIN (VERBOSE OFF) ...
> > Its also currently impossible to use an empty set of parentheses to
> > resolve this - this could easily be changed though.
> >
> > I have to admit I don't see a nice solution here except living with the
> > incompatibility... Perhaps somebody has a better idea?
>
> I think another possibility might be to allow the syntax:
>
> EXPLAIN VERBOSE ANALYSE (options) SELECt ...;
>
> Sure, it's a bit ugly, but in the grammer you could then do:
> > ExplainStmt: EXPLAIN opt_analyze opt_verbose ExplainableStmt
> >
> > | EXPLAIN opt_analyze opt_verbose '(' explain_option_list ')'
> > | ExplainableStmt
>
> Which means that (I think) bison can use the token *after* the '(' to
> disambiguate, and since SELECT is a reserved word I think the problem
> may be solved.
I think that does not work since explain_option_name has to include keywords
to be able to use ANALYZE and VERBOSE.
Its solvable by not allowing all keywords there but only ANALYZE and VERBOSE.
Involves some duplication though...
Patch attached.
Andres