Severe Pro*C / nested table error

Question

I am experiencing a nasty error which involves Pro*C accessing nested table columns. I am running Oracle 9.2.0.1.0 on Solaris 2.8 and using the GNU gcc compiler.
In my examples, I will use the sample data which comes with Pro*C ($ORACLE_HOME/precomp/demo/sql/coldemo1.sql).
This defines the following types and table (and also populates the table with sample data):
CREATE TYPE city_t AS OBJECT (name CHAR(30), population NUMBER);
CREATE TYPE citytbl_t AS TABLE OF city_t;
CREATE TABLE county_tbl (name CHAR(30), cities citytbl_t)
NESTED TABLE cities STORE AS citytbl_t_tbl;
I have no problems running the sample program ($ORACLE_HOME/precomp/demo/proc/coldemo1.pc). Note that you must follow the instructions included in the comments of coldemo1.pc (manually execute the Oracle Type Translator, precompile, compile and link).
The sample, however, merely retrieves and displays existing data. I would like my program to be able to insert new rows into county_tbl, including the nested table.
I have the following code (for simplicitys sake, everything is in the main procedure):
#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include <sqlca.h>                                /* SQL Communications Area */
#include <coldemo1.h>        /* OTT-generated header with C typedefs for the */
                                      /* database types city_t and citytbl_t */
#define CITY_NAME_LEN    30
#define COUNTY_NAME_LEN  30
#define MAXCITIES 5
void PrintSQLError()
  EXEC SQL WHENEVER SQLERROR CONTINUE;
  printf("SQL error occurred...\n");
  printf("%.*s\n", (int)sqlca.sqlerrm.sqlerrml,
         (CONST char *)sqlca.sqlerrm.sqlerrmc);
  EXEC SQL ROLLBACK RELEASE;
  exit(-1);
int main(int argc, char *argv[])
  char * uid = "scott/tiger";
  citytbl_t *cityTable;
  OCIInd cityTableInd;
  char county[COUNTY_NAME_LEN + 1];
  city_t *city[MAXCITIES];
  city_t_ind *cityInd[MAXCITIES];
  char cname[CITY_NAME_LEN + 1];
  int cpop;
  int numCities = MAXCITIES;
  int i, tableSize;
  EXEC SQL WHENEVER SQLERROR DO PrintSQLError();
  EXEC SQL CONNECT :uid;
  /* to null-terminate strings */
  EXEC ORACLE OPTION (char_map=string);
  /* Allocate the required descriptors */
  EXEC SQL ALLOCATE :cityTable;
  EXEC SQL ALLOCATE :city:cityInd;
  strcpy(county,"MYCOUNTY");
  /* Create an array of city objects */
  EXEC SQL FOR :numCities OBJECT CREATE :city:cityInd;
  for(i=0;i<numCities;i++)
    sprintf(cname,"city%d", i);
    cpop = (i * 100);
    EXEC SQL OBJECT SET NAME, POPULATION OF :city[i] TO :cname, :cpop;
    cityInd[i]->_atomic = cityInd[i]->NAME = cityInd[i]->POPULATION = OCI_IND_NOTNULL;
  cityTableInd = OCI_IND_NOTNULL;
  EXEC SQL COLLECTION APPEND :city:cityInd TO :cityTable:cityTableInd;
  EXEC SQL INSERT INTO county_tbl (NAME, CITIES) VALUES (:county, :cityTable:cityTableInd);
  EXEC SQL FREE :city;
  EXEC SQL FREE :cityTable;
  EXEC SQL COMMIT RELEASE;
  return(0);
}This works as expected, inserting one row into COUNTY_TBL, with name=MYCOUNTY, cities being a nested table containing five city elements.
My problem occurs when I create a trigger on COUNTY_TBL which attempts to modify elements of the nested table column. Heres the trigger (in this case, we are converting each NAME element of the cities table to upper case):
create or replace trigger biur_county
before insert or update on county_tbl
for each row
when (new.cities is not null)
BEGIN
IF(:NEW.cities.COUNT > 0) THEN
FOR i IN :NEW.cities.FIRST..:NEW.cities.LAST LOOP
:NEW.cities(i).NAME := UPPER( :NEW.cities(i).NAME );
END LOOP;
END IF;
END biur_county;
From SQL*Plus, this works as expected:
SQL> insert into county_tbl values ('ANOTHERCOUNTY',CITYTBL_T(CITY_T('city1',1),CITY_T('city2',2)));
1 row created.
SQL> select * from county_tbl where name='ANOTHERCOUNTY';
NAME
CITIES(NAME, POPULATION)
ANOTHERCOUNTY
CITYTBL_T(CITY_T('CITY1 ', 1), CITY_T('CITY2 ', 2))
1 row selected.
Note that the city names have been converted to upper case.
If, however, I attempt to run the Pro*C program again, with the trigger enabled, I get a severe error:
ORA-03113: end-of-file on communication channel
This also creates entries in the alert log:
Errors in file /space/oracle/product/9.2.0.1.0/admin/ATDEV/udump/atdev_ora_16790.trc:
ORA-07445: exception encountered: core dump [0000000100C8EF7C] [SIGBUS] [Invalid address alignment] [0x400000006] [] []
The trace file includes:
Exception signal: 10 (SIGBUS), code: 1 (Invalid address alignment), addr: 0x400000006, PC: [0x100c8e
f7c, 0000000100C8EF7C]
*** 2002-12-11 15:51:25.469
ksedmp: internal or fatal error
ORA-07445: exception encountered: core dump [0000000100C8EF7C] [SIGBUS] [Invalid address alignment]
[0x400000006] [] []
Current SQL statement for this session:
insert into county_tbl (NAME,CITIES) values (:b0,:b1:b2)
----- PL/SQL Call Stack -----
object line object
handle number name
38421f020 3 SCOTT.BIUR_COUNTY
If I subsequently disable the trigger (or alter the trigger so that the contents of the nested table are not modified), all is well again.
Has anyone else encountered a similar problem? I am assuming that the problem has something to do with the objects in Pro*C being transient and held in the object cache on the client. Perhaps I am doing something wrong in my program. Does anyone have any thoughts?
Certainly there are ways around this problem (remove the trigger, for example, or perform inserts through a view / instead of trigger, etc), but I would like to have an explanation as to why my method is not working.
Any insights would be greatly appreciated.

Answer 1

HELLO!!!
This problem reveals ONLY in DEBUG configuration (in RELEASE - all ok, but DEBUG is very important feature ;)) and operator
new in my case have not an influence on this situation.
I found a problem in other place when calling boxes=box.ptr();
The boxes oracle class have a nested table. In C++ :
class CBoxes : public oracle::occi::PObject {
private:
OCCI_STD_NAMESPACE::string BOXNAME;
oracle::occi::Number BOXNUM;
OCCI_STD_NAMESPACE::vector<CWidget*> WIDGETS;
ptr() function calling readSQL method of CBoxes class:
BOXNAME = streamOCCI_.getString();
BOXNUM = streamOCCI_.getNumber();
oracle::occi::getVector(streamOCCI_, WIDGETS);
and getVector is:
//cutted from occiObjects.h------------------------------
template <class T>
void getVector(const AnyData &any, OCCI_STD_NAMESPACE::vector<T> &vect)
{OCCI_STD_NAMESPACE::vector< PObject *> vec_pobj;      
getVectorOfPObjects( any,vec_pobj);//!!!!!!! :[  ]
vect.clear();
int size= vec_pobj.size();
for( int i=0; i< size; i++)
vect.push_back( (T)vec_pobj[i] );
}//<- the destructor of vec_pobj is being called and if vec_pobj not an empty vector, the program crashes.
I suppose that getVectorOfPObjects( any,vec_pobj) can't correctly allocate memory for vec_pobj elements in DEBUG mode.