Thread: Problem with using snprintf in C-function

Hi,

I have a problem with snprintf function which insert the variables to string with error.
This code I wrote for PostgreSQL 8.3.7 on Fedora 10 and it worked fine. But now I use CentOS 5.3 and PostgreSQL 8.3.8.

Function:
Datum calculateAccount(PG_FUNCTION_ARGS) {
    int64   accountId = PG_GETARG_INT64(0);
    DateADT startDate = PG_GETARG_DATEADT(1);
    DateADT endDate = PG_GETARG_DATEADT(2);

    char    command[QUERY_MAX_SIZE];
    char    startDateStr[MAXDATELEN + 3];
    char    endDateStr[MAXDATELEN + 3];

    snprintf(startDateStr, sizeof (startDateStr), "'%s'", DatumGetCString(DirectFunctionCall1(date_out, startDate)));
    snprintf(endDateStr, sizeof (endDateStr), "'%s'", DatumGetCString(DirectFunctionCall1(date_out, endDate)));
    elog(INFO, startDateStr);
    elog(INFO, endDateStr);
    snprintf(command, sizeof (command),
            "SELECT serviceId, periodStartDate, periodEndDate\
             FROM   accountServiceBaseView\
             WHERE  accountId = %ld AND\
                    periodStartDate <= %s AND\
                    periodEndDate >= %s;",
            accountId, startDateStr, endDateStr);
    elog(INFO, command);

    PG_RETURN_BOOL(true);
}

Result:
select calculateaccount(123, '01-01-2009', '01-02-2009');
INFO:  '2009-01-01'
INFO:  '2009-02-01'
INFO:  SELECT serviceId, periodStartDate, periodEndDate FROM accountServiceBaseView WHERE  accountId = 123 AND periodStartDate <= (null) AND periodEndDate >= '2009-01-01';


If I change the order of parameters to "periodStartDate <= %s AND periodEndDate >= %s AND accountId = %ld", all works fine. Unfortunately this method can not be applied to other functions of my module.


--
Best regards,
Ilya Urikh.

2009/10/17 Ilya Urikh <ilya.urikh@gmail.com>:
> Hi,
>
> I have a problem with snprintf function which insert the variables to string
> with error.
> This code I wrote for PostgreSQL 8.3.7 on Fedora 10 and it worked fine. But
> now I use CentOS 5.3 and PostgreSQL 8.3.8.
>
> Function:
> Datum calculateAccount(PG_FUNCTION_ARGS) {
>     int64   accountId = PG_GETARG_INT64(0);
>     DateADT startDate = PG_GETARG_DATEADT(1);
>     DateADT endDate = PG_GETARG_DATEADT(2);
>
>     char    command[QUERY_MAX_SIZE];
>     char    startDateStr[MAXDATELEN + 3];
>     char    endDateStr[MAXDATELEN + 3];
>
>     snprintf(startDateStr, sizeof (startDateStr), "'%s'",
> DatumGetCString(DirectFunctionCall1(date_out, startDate)));
>     snprintf(endDateStr, sizeof (endDateStr), "'%s'",
> DatumGetCString(DirectFunctionCall1(date_out, endDate)));
>     elog(INFO, startDateStr);
>     elog(INFO, endDateStr);
>     snprintf(command, sizeof (command),
>             "SELECT serviceId, periodStartDate, periodEndDate\
>              FROM   accountServiceBaseView\
>              WHERE  accountId = %ld AND\
>                     periodStartDate <= %s AND\
>                     periodEndDate >= %s;",
>             accountId, startDateStr, endDateStr);
>     elog(INFO, command);
>

Hello

my reply is little bit offtopic. Why you convert date values to
string? You can use placeholders and execute query with parameters in
native format. It's more simply and safe.

Regards
Pavel Stehule




>     PG_RETURN_BOOL(true);
> }
>
> Result:
> select calculateaccount(123, '01-01-2009', '01-02-2009');
> INFO:  '2009-01-01'
> INFO:  '2009-02-01'
> INFO:  SELECT serviceId, periodStartDate, periodEndDate FROM
> accountServiceBaseView WHERE  accountId = 123 AND periodStartDate <= (null)
> AND periodEndDate >= '2009-01-01';
>
>
> If I change the order of parameters to "periodStartDate <= %s AND
> periodEndDate >= %s AND accountId = %ld", all works fine. Unfortunately this
> method can not be applied to other functions of my module.
>
>
> --
> Best regards,
> Ilya Urikh.
>

Pavel,

Could you provide some example? I don't really understand which placeholder I can use.





--
Best regards,
Ilya Urikh.

2009/10/17 Ilya Urikh <ilya.urikh@gmail.com>:
> Pavel,
>
> Could you provide some example? I don't really understand which placeholder
> I can use.
>

here is one fragment from orafce:
it's for delete statement, but SELECT statement is similar

        ItemPointer tid;
        Oid argtypes[1] = {TIDOID};
        char nulls[1] = {' '};
        Datum values[1];
        void *plan;

        tid = &rettuple->t_data->t_ctid;

        if (!(plan = SPI_prepare("DELETE FROM ora_alerts WHERE ctid = $1",
1, argtypes)))
            ereport(ERROR,
                (errcode(ERRCODE_TRIGGERED_ACTION_EXCEPTION),
                 errmsg("SPI_prepare failed")));

        values[0] = ItemPointerGetDatum(tid);

        if (SPI_OK_DELETE != SPI_execute_plan(plan, values, nulls, false, 1))
            ereport(ERROR,
                (errcode(ERRCODE_TRIGGERED_ACTION_EXCEPTION),
                errmsg("can't execute sql")));

        SPI_finish();
        return PointerGetDatum(rettuple);

I don't found better samples, but both functions are well documented.

your code should look like:

Datum params[3];
char nulls[3] = {' ',' ',' '} ;
Oid types[3] = {INT8OID, DATEOID, DATEOID};
void *plan;

  params[0] = PG_GETARG_DATUM(0);
  params[1] = PG_GETARG_DATUM(1);
  params[2] = PG_GETARG_DATUM(2);

  nulls[0] = PG_ARGISNULL(0) ? 'n' : ' ';
  ...

plan = SPI_prepare("SELECT ...",3, types);
if (plan == NULL)
  ... raise exception

result = SPI_execute_plan(plan, params, nulls, true, -1);
...

Regards
Pavel Stehule.


> On Sun, Oct 18, 2009 at 1:20 AM, Pavel Stehule <pavel.stehule@gmail.com>
> wrote:
>>
>> 2009/10/17 Ilya Urikh <ilya.urikh@gmail.com>:
>> > Hi,
>> >
>> > I have a problem with snprintf function which insert the variables to
>> > string
>> > with error.
>> > This code I wrote for PostgreSQL 8.3.7 on Fedora 10 and it worked fine.
>> > But
>> > now I use CentOS 5.3 and PostgreSQL 8.3.8.
>> >
>> > Function:
>> > Datum calculateAccount(PG_FUNCTION_ARGS) {
>> >     int64   accountId = PG_GETARG_INT64(0);
>> >     DateADT startDate = PG_GETARG_DATEADT(1);
>> >     DateADT endDate = PG_GETARG_DATEADT(2);
>> >
>> >     char    command[QUERY_MAX_SIZE];
>> >     char    startDateStr[MAXDATELEN + 3];
>> >     char    endDateStr[MAXDATELEN + 3];
>> >
>> >     snprintf(startDateStr, sizeof (startDateStr), "'%s'",
>> > DatumGetCString(DirectFunctionCall1(date_out, startDate)));
>> >     snprintf(endDateStr, sizeof (endDateStr), "'%s'",
>> > DatumGetCString(DirectFunctionCall1(date_out, endDate)));
>> >     elog(INFO, startDateStr);
>> >     elog(INFO, endDateStr);
>> >     snprintf(command, sizeof (command),
>> >             "SELECT serviceId, periodStartDate, periodEndDate\
>> >              FROM   accountServiceBaseView\
>> >              WHERE  accountId = %ld AND\
>> >                     periodStartDate <= %s AND\
>> >                     periodEndDate >= %s;",
>> >             accountId, startDateStr, endDateStr);
>> >     elog(INFO, command);
>> >
>>
>> Hello
>>
>> my reply is little bit offtopic. Why you convert date values to
>> string? You can use placeholders and execute query with parameters in
>> native format. It's more simply and safe.
>>
>> Regards
>> Pavel Stehule
>>
>>
>>
>>
>> >     PG_RETURN_BOOL(true);
>> > }
>> >
>> > Result:
>> > select calculateaccount(123, '01-01-2009', '01-02-2009');
>> > INFO:  '2009-01-01'
>> > INFO:  '2009-02-01'
>> > INFO:  SELECT serviceId, periodStartDate, periodEndDate FROM
>> > accountServiceBaseView WHERE  accountId = 123 AND periodStartDate <=
>> > (null)
>> > AND periodEndDate >= '2009-01-01';
>> >
>> >
>> > If I change the order of parameters to "periodStartDate <= %s AND
>> > periodEndDate >= %s AND accountId = %ld", all works fine. Unfortunately
>> > this
>> > method can not be applied to other functions of my module.
>> >
>> >
>> > --
>> > Best regards,
>> > Ilya Urikh.
>> >
>
>
>
> --
> Best regards,
> Ilya Urikh.
>

Ilya Urikh <ilya.urikh@gmail.com> writes:
> This code I wrote for PostgreSQL 8.3.7 on Fedora 10 and it worked fine. But
> now I use CentOS 5.3 and PostgreSQL 8.3.8.

It looks to me like you're assuming that %ld is the appropriate format
string for an int64 value.  This might be true on a 64-bit machine but
it would not be on 32-bit.  It's not obvious why that would lead to
exactly the results you show, but it could certainly account for
snprintf fetching the wrong bits for some of its arguments.

The portable way (within Postgres code) is to use INT64_FORMAT, which
will be set up correctly by configure.  So instead of

>              WHERE  accountId = %ld AND\

you need

               WHERE  accountId = " INT64_FORMAT " AND\

(relying on concatenation of literals in C).

BTW, if your compiler didn't warn you about this, you need to get a
new compiler, or at least start using the switches that would make it
warn you.  If you did get a warning, and ignored it, you need to
learn better habits ;-)

            regards, tom lane