Обсуждение: Projecting attributes of function return values
Hi all
I found 2 BUGs ? caused by ParseComplexProject() function in parse_func.c.
The result of the following cases are different from the design
of ParseComplexProject() in parse_func.c ?
1. Projecting attributes of function return values never execute right operations. create table a (elem int4);
createfunction ax2(int4) returns a as 'select $1*2;' language 'sql'; select elem(ax2(1)); the result is elem ----
4 (1 row) It's wrong. PostgreSQL executor evaluates functions twice. After applying patch the result is the
following elem ---- 2 (1 row)
2.PostgreSQL parser rejects functions unnecessarily create function elemout(a) returns int4 as 'select $1.elem;'
language'sql'; select elemout(ax2(1)); ERROR: Function 'elemout' has bad return type 315169
This rejection by elog() is necessary ? I removed it and the result is elemout ------- 2 (1 row)
I posted a patch for these cases (Subject:Projecting attributes
of function return values) a few days ago.
The same patch is at the end of this posting.
If those cases are BUGs,please confirm my patch.
Thanks
Hiroshi Inoue
Inoue@tpf.co.jp
*** backend/parser/parse_func.c.orig Thu Dec 3 18:25:31 1998
--- backend/parser/parse_func.c Thu Dec 3 18:39:18 1998
***************
*** 1352,1358 **** newexpr->typeOid = funcnode->functype;
newexpr->opType= FUNC_EXPR; newexpr->oper = (Node *) funcnode;
! newexpr->args = lcons(first_arg, NIL); return (Node *) newexpr;
}
--- 1352,1358 ---- newexpr->typeOid = funcnode->functype;
newexpr->opType= FUNC_EXPR; newexpr->oper = (Node *) funcnode;
! newexpr->args = expr->args; return (Node *) newexpr;
}
***************
*** 1359,1366 **** }
- elog(ERROR, "Function '%s' has bad return type %d",
- funcname, argtype); break; } case T_Param:
--- 1359,1364 ----
I have applied this to the CURRENT tree, so the fix will appear in 6.5. I have not applied it to the RELEASE tree, so it will not be in 6.4.1. Not sure there is going to be enough testing for that. Thanks for the patch. > > Hi all > > I have a simple patch about the treatment of functions. > But it may be self-satisfied. > Please check my patch at the end of this posting. > > Case 1. executor evaluates functions twice > > create table a (elem int4); > create function ax2(int4) returns a as > 'select $1*2;' language 'sql'; > select elem(ax2(1)); > > the result is > > elem > ---- > 4 > (1 row) > > it's wrong ? > it must be the following ? > > elem > ---- > 2 > (1 row) > > Case 2.parser rejects functions > > create function elemout(a) returns int4 as > 'select $1.elem;' language 'sql'; > select elemout(ax2(1)); > > ERROR: Function 'elemout' has bad return type 315169 > > this elog() is necessary ? > In my patch I removed it and the result is > > elemout > ------- > 2 > (1 row) > > Thanks > > Hiroshi Inoue > Inoue@tpf.co.jp > > *** backend/parser/parse_func.c.orig Thu Dec 3 18:25:31 1998 > --- backend/parser/parse_func.c Thu Dec 3 18:39:18 1998 > *************** > *** 1352,1358 **** > newexpr->typeOid = funcnode->functype; > newexpr->opType = FUNC_EXPR; > newexpr->oper = (Node *) funcnode; > ! newexpr->args = lcons(first_arg, NIL); > > return (Node *) newexpr; > } > --- 1352,1358 ---- > newexpr->typeOid = funcnode->functype; > newexpr->opType = FUNC_EXPR; > newexpr->oper = (Node *) funcnode; > ! newexpr->args = expr->args; > > return (Node *) newexpr; > } > *************** > *** 1359,1366 **** > > } > > - elog(ERROR, "Function '%s' has bad return type %d", > - funcname, argtype); > break; > } > case T_Param: > --- 1359,1364 ---- > > -- Bruce Momjian | http://www.op.net/~candle maillist@candle.pha.pa.us | (610) 853-3000+ If your life is a hard drive, | 830 Blythe Avenue + Christ can be your backup. | Drexel Hill, Pennsylvania19026
>
> I have applied this to the CURRENT tree, so the fix will appear in 6.5.
> I have not applied it to the RELEASE tree, so it will not be in 6.4.1.
> Not sure there is going to be enough testing for that.
>
> Thanks for the patch.
>
> >
> > Hi all
> >
> > I have a simple patch about the treatment of functions.
> > But it may be self-satisfied.
> > Please check my patch at the end of this posting.
> >
> > Case 1. executor evaluates functions twice
> >
> > create table a (elem int4);
> > create function ax2(int4) returns a as
> > 'select $1*2;' language 'sql';
It addresses a problem I've mentioned some time ago.
Functions returning sets or complex types have targetlists,
but these aren't handled correctly in the executor.
Me and Thomas G. Lockhard where discussing another rangetable
entry type lately, which is in fact a subselect. That might
cover this problem more completely in the future.
So pls let it out of the REL6_4 tree for now.
Jan
--
#======================================================================#
# It's easier to get forgiveness for being wrong than for being right. #
# Let's break this rule - forgive me. #
#======================================== jwieck@debis.com (Jan Wieck) #
> -----Original Message----- > From: Jan Wieck [mailto:jwieck@debis.com] > Sent: Monday, December 14, 1998 6:38 PM > To: Bruce Momjian > Cc: Inoue@tpf.co.jp; pgsql-hackers@postgreSQL.org > Subject: Re: [HACKERS] Projecting attributes of function return values > > > > > > I have applied this to the CURRENT tree, so the fix will appear in 6.5. > > I have not applied it to the RELEASE tree, so it will not be in 6.4.1. > > Not sure there is going to be enough testing for that. > > > > Thanks for the patch. > > > > > > > > Hi all > > > > > > I have a simple patch about the treatment of functions. > > > But it may be self-satisfied. > > > Please check my patch at the end of this posting. > > > > > > Case 1. executor evaluates functions twice > > > > > > create table a (elem int4); > > > create function ax2(int4) returns a as > > > 'select $1*2;' language 'sql'; > > It addresses a problem I've mentioned some time ago. > Functions returning sets or complex types have targetlists, > but these aren't handled correctly in the executor. > It includes the fact that func_tlist of Func node is not used to project attributes of function return values except when functions are written in 'sql' language ? and the fact that PL/pgSQL functions return different type(HeapTuple) from 'sql' functions which return the pointer to TupleTableSlot node ? My samples are all written in 'sql' language. Seems the cases I mentioned are simple bugs. Thanks. Hiroshi Inoue Inoue@tpf.co.jp