FreeTDS Development Group

["James K. Lowden" <jklowden AT schemamania.org>]

On Wed, 31 Mar 2004, "Michael Sims" <michaels AT crye-leike.com> wrote:
> > Based on what you found, and on my knowledge of our db-lib
> > implementation, it's very hard to see how (or if) this is caused by
> > FreeTDS.  The --with-mssql extension resolves to the very same db-lib
> > call, and works correctly.  Perhaps the older with-sybase handling
> > expects dbresults() to behave a little differently?
...
> <quote php_sybase_db.c's sybase_query()>
> /* The following is more or less the equivalent of mysql_store_result().
>  * fetch all rows from the server into the row buffer, thus:
>  * 1)  Being able to fire up another query without explicitly reading
>  all
> rows
>  * 2)  Having numrows accessible
>  */
> 
> retvalue=dbnextrow(sybase_ptr->link);
> 
> if (retvalue==FAIL) {
>   RETURN_FALSE;
> }
> 
> num_fields = dbnumcols(sybase_ptr->link);
> if (num_fields<=0) {
>   RETURN_TRUE;
> }
> </quote>

Well, well.  That's a little bit weird.  

On the PHP side, usually one calls dbnumcols() before dbnextrow().  I
don't think it's wrong to do it in the other order (since they're setting
up a do-while loop).  But it's unconventional.  

(The idea is that dbresults() sets up the metadata, so you can ask about
the columns count, their names, datatypes.  You bind your columns and ask
if there are any rows to bother fething.  Then use dbnextrow() to fetch
them.)

Next up is "num_fields<=0".  That's never supposed to happen, according to
Sybase or Microsoft (and, hence, according to FreeTDS).  dbnumcols()
should return 0 or more columns.  So, that test should never evaluate to
true.  

http://msdn.microsoft.com/library/default.asp?url=/library/en-us/dblibc/dbc_pdc04d_3c6r.asp

I don't have time to look into it further right now, but here's our
implementation of dbnumcols():

int
dbnumcols(DBPROCESS * dbproc)
{
        TDSRESULTINFO *resinfo;
        TDSSOCKET *tds;

        tds = (TDSSOCKET *) dbproc->tds_socket;
        resinfo = tds->res_info;
        if (resinfo)
                return resinfo->num_cols;
        return 0;
}

IOW, we return whatever's in dbproc->tds_socket->res_info->num_cols.  It's
possible that the lower layers set that puppy to -1 at some point, and
this code doesn't clear it.  For now, you could try changing it to:

int
dbnumcols(DBPROCESS * dbproc)
{
        TDSRESULTINFO *resinfo;
        TDSSOCKET *tds;

        tds = (TDSSOCKET *) dbproc->tds_socket;
        resinfo = tds->res_info;
        if (resinfo)
                return (resinfo->num_cols < 0)? 0 : resinfo->num_cols;
        return 0;
}

I touched on some of the underlying issues recently.  When we get CVS HEAD
linkable for you again, we'll be in a better position to find a real fix. 


HTH.

--jkl