Hi,
I'm using libpq PQexecParams ability to send query parameters as binary dat=
a,=20
more specific - a binary representation of an empty array.
I have the problem:=20
the code sending empty binary array works on 8.3 and 8.4=20
but stopped working on postgres 9.0 beta2/3/4, it fails with '22003','integ=
er out of range error'.
Here is the test code:
char *command=3D"INSERT INTO test.arraytypes(nullarr) VALUES($1)";
// ndims hassnull typeid ((dim+lbound)*ndims)
const int empty_array_length =3D 4 + 4 + 4 + (8*1);
// constructing empty array representation:
char * buf =3D (char *)malloc(empty_array_length);
memset(buf,0,empty_array_length);
char * out =3D buf;
// ndims
*((int*)out) =3D htonl(1);
out+=3D4;
// hassnull
*((int*)out) =3D 0;
out+=3D4;
// typeid 'int4'
*((int*)out) =3D htonl(23);
out+=3D4;
const Oid oids[] =3D {1007};// _int4 oid
const int paramFormats[]=3D{1};
const int lens[] =3D {empty_array_length};
const char * const * vals =3D {&buf};
PGresult* re =3D PQexecParams(conn, command, 1, oids,
(const char *const * ) vals, lens, paramFormats, 1);
char *err1=3DPQresultErrorMessage(re); // ERROR: integer out of range for =
9.0 beta
// sql creation code:
//CREATE TABLE test.arraytypes( nullarr int[])
I've traced the error down to array_recv function, and found this overflow =
check to be offending.
/src/backend/utils/adt/arrayfuncs.c: Line 1214
for (i =3D 0; i < ndim; i++)
{
int ub;
dim[i] =3D pq_getmsgint(buf, 4);
lBound[i] =3D pq_getmsgint(buf, 4);
ub =3D lBound[i] + dim[i] - 1;
/* overflow? */
if (lBound[i] > ub)
ereport(ERROR,
(errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
errmsg("integer out of range")));
}
In the empty array case, the ndim=3D=3D1, and dim[0]=3D=3D0, so lBound[0] >=
ub[0]
Seems that the overflow check was introduced fairly recently, here is the d=
iscussion of it:
http://archives.postgresql.org/pgsql-hackers/2009-08/msg02073.php
The array with dim[i] =3D 0 seems legitimate, since this situation is handl=
ed by the code below:
if (nitems =3D=3D 0)
{
/* Return empty array ... but not till we've validated element_type */
PG_RETURN_ARRAYTYPE_P(construct_empty_array(element_type));
}
So, is it really a bug in 9.0? Or maybe the array representation rules chan=
ged somehow?
Vladimir Shakhov | Developer
www.alliedtesting.com