Re: [bug fix] Memory leak in dblink

Joe Conway <mail@joeconway.com> writes:
> On 06/18/2014 08:19 PM, Tom Lane wrote:
>> Actually, I was wondering whether we couldn't remove that 
>> CreateTupleDescCopy call entirely.

> Apparently not, at least without some additional surgery.
> ExecMakeTableFunctionResult() tries to free the tupledesc and segfaults.

Hmm ... oh, I missed this bit:
           /* We must get the tupledesc from call context */           if (rsinfo && IsA(rsinfo, ReturnSetInfo) &&
        rsinfo->expectedDesc != NULL)           {               result = TYPEFUNC_COMPOSITE;               if
(resultTupleDesc)                  *resultTupleDesc = rsinfo->expectedDesc;               /* Assume no polymorphic
columnshere, either */           }
 

So it's passing back the same tupdesc passed in by the caller of
ExecMakeTableFunctionResult.  We can free that all right, but the caller
won't be happy :-(
        regards, tom lane