Re: [HACKERS] WITH RECURSIVE patches 0818 - Mailing list pgsql-patches

> I think I may have found another bug:
>
> WITH RECURSIVE t(i,j) AS (
>     VALUES (1,2)
> UNION ALL
>     SELECT t2.i, t.j
>     FROM (
>         SELECT 2 AS i
>     UNION ALL               /* Wrongly getting detected, I think */
>         SELECT 3 AS i
>     ) AS t2
>     JOIN
>         t
>         ON (t2.i = t.i)
> )
> SELECT * FROM t;
> ERROR:  attribute number 2 exceeds number of columns 1
>
> Is there some way to ensure that in the case of WITH RECURSIVE, the
> query to the right of UNION ALL follows only the SQL:2008 rules about
> not having outer JOINs, etc. in it, but otherwise make it opaque to
> the error-checking code?
>
> I know I didn't explain that well, but the above SQL should work and
> the error appears to stem from the parser's looking at the innermost
> UNION ALL instead of the outermost.

Here is new patches fixing the bug you pointed out (patches was
created by Yoshiyuki). Also I added your SQL to the regression test,
and now the patches is against CVS HEAD. For your convenience I also
include patches against the previous version.
--
Tatsuo Ishii
SRA OSS, Inc. Japan
*** pgsql/src/backend/executor/nodeAppend.c    2008-08-18 16:20:40.000000000 +0900
--- pgsql.patched/src/backend/executor/nodeAppend.c    2008-08-23 07:37:29.000000000 +0900
***************
*** 143,152 ****
--- 143,156 ----
      int            nplans;
      int            i;
      Plan       *initNode;
+     TupleDesc  save_tupledesc;

      /* check for unsupported flags */
      Assert(!(eflags & EXEC_FLAG_MARK));

+     /* Save tuple desc */
+     save_tupledesc = estate->es_rscan_tupledesc;
+
      /*
       * Set up empty vector of subplan states
       */
***************
*** 232,237 ****
--- 236,243 ----
      appendstate->as_whichplan = appendstate->as_firstplan;
      exec_append_initialize_next(appendstate);

+     /* Restore tuple desc */
+     estate->es_rscan_tupledesc = save_tupledesc;
+
      return appendstate;
  }