Thread: Issue on C function that reads int2[] (using "int2vector")

Issue on C function that reads int2[] (using "int2vector")

From
Rodrigo Hjort
Date:
Hello PG Hackers,


I created a custom C function with this signature:

CREATE FUNCTION calculate_hash(numbers int2[])
RETURNS int8
AS 'MODULE_PATHNAME', 'pg_calculate_hash'
LANGUAGE C
IMMUTABLE STRICT;


And here is the function source code (inspired in codes I found in src/backend/utils/adt/int.c):

PG_FUNCTION_INFO_V1(pg_calculate_hash);
Datum
pg_calculate_hash(PG_FUNCTION_ARGS)
{
  int2vector *int2Array = (int2vector *) PG_GETARG_POINTER(0);
  const int qtd = int2Array->dim1;

  elog(DEBUG1, "pg_calculate_hash(qtd=%d)", qtd);

  elog(DEBUG2, "  [ndim=%d, dataoffset=%d, elemtype=%d, dim1=%d, lbound1=%d]",
    int2Array->ndim, int2Array->dataoffset, int2Array->elemtype, int2Array->dim1, int2Array->lbound1);

  [...]
}


In order to test it against a table structure, I executed these instructions on psql:

db=# create table ss (s int2[]);
CREATE TABLE

db=# \d+ ss
                            Table "public.ss"
 Column |    Type    | Modifiers | Storage  | Stats target | Description 
--------+------------+-----------+----------+--------------+-------------
 s      | smallint[] |           | extended |              | 
Has OIDs: no

db=# insert into ss values ('[0:5]={58,17,15,36,59,54}');
INSERT 0 1

db=# select * from ss;
             s             
---------------------------
 [0:5]={58,17,15,36,59,54}
(1 row)


Then, whenever calling the function passing the int2[] column directly, strange values are read into the "int2vector" object:

db=# set client_min_messages to debug2;
SET

db=# select s, calculate_hash(s) from ss;
DEBUG:  pg_calculate_hash(qtd=0)
DEBUG:    [ndim=0, dataoffset=5376, elemtype=1536, dim1=0, lbound1=285227520]
             s             | calculate_hash 
---------------------------+---------------
 [0:5]={58,17,15,36,59,54} |             0
(1 row)


On the other hand, when I double-cast the int2[] column value, it works as expected (reading the proper "int2vector" structure):

db=# select s, calculate_hash(s::varchar::int2[]) from ss;
DEBUG:  pg_calculate_hash(qtd=6)
DEBUG:    [ndim=1, dataoffset=0, elemtype=21, dim1=6, lbound1=0]
             s             |   calculate_hash    
---------------------------+--------------------
 [0:5]={58,17,15,36,59,54} | 441352797842128896
(1 row)


Please, what is wrong with that function code?

Thanks in advance.

The whole project is on GitHub:


Best regards,

--
Rodrigo Hjort
www.hjort.co

Re: Issue on C function that reads int2[] (using "int2vector")

From
Tom Lane
Date:
Rodrigo Hjort <rodrigo.hjort@gmail.com> writes:
> I created a custom C function with this signature:

> CREATE FUNCTION calculate_hash(numbers int2[])
> RETURNS int8
> AS 'MODULE_PATHNAME', 'pg_calculate_hash'
> LANGUAGE C
> IMMUTABLE STRICT;

> And here is the function source code (inspired in codes I found in
> src/backend/utils/adt/int.c):

> PG_FUNCTION_INFO_V1(pg_calculate_hash);
> Datum
> pg_calculate_hash(PG_FUNCTION_ARGS)
> {
>   int2vector *int2Array = (int2vector *) PG_GETARG_POINTER(0);

Nope.  int2vector is not the same as int2[].  It might occasionally seem
to work, but in general it's not the same type.  And this particular
coding won't work at all on on-disk int2[] data, because it doesn't
account for toasting.
        regards, tom lane



Re: Issue on C function that reads int2[] (using "int2vector")

From
Rodrigo Hjort
Date:
2015-11-30 0:39 GMT-02:00 Tom Lane:
Rodrigo Hjort writes:
> I created a custom C function with this signature:

> CREATE FUNCTION calculate_hash(numbers int2[])
> RETURNS int8
> AS 'MODULE_PATHNAME', 'pg_calculate_hash'
> LANGUAGE C
> IMMUTABLE STRICT;

> And here is the function source code (inspired in codes I found in
> src/backend/utils/adt/int.c):

> PG_FUNCTION_INFO_V1(pg_calculate_hash);
> Datum
> pg_calculate_hash(PG_FUNCTION_ARGS)
> {
>   int2vector *int2Array = (int2vector *) PG_GETARG_POINTER(0);

Nope.  int2vector is not the same as int2[].  It might occasionally seem
to work, but in general it's not the same type.  And this particular
coding won't work at all on on-disk int2[] data, because it doesn't
account for toasting.

                        regards, tom lane


Thanks for the advice, Tom.

I ended up with the following code, which worked successfully:

#define ARRPTR16(x)   ((uint16 *) ARR_DATA_PTR(x))
#define ARRNELEMS(x)  ArrayGetNItems(ARR_NDIM(x), ARR_DIMS(x))
#define ARRISEMPTY(x) (ARRNELEMS(x) == 0)

PG_FUNCTION_INFO_V1(pg_calculate_hash);
Datum
pg_calculate_hash(PG_FUNCTION_ARGS)
{
  ArrayType *a = PG_GETARG_ARRAYTYPE_P_COPY(0);
  unsigned int i, qtd, tipo, nums[MAX_NUMEROS];
  uint16 *da;

  qtd = ARRNELEMS(a);
  tipo = ARR_ELEMTYPE(a);
  da = ARRPTR16(a);

  elog(DEBUG1, "pg_calculate_hash(qtd=%d, tipo=%d)", qtd, tipo);

  [...]

  pfree(a);
  PG_RETURN_INT64(hash);
}


Regards,

Rodrigo Hjort