Re: json_populate_record issue - TupleDesc reference leak

On 02/23/2015 12:56 PM, Pavel Stehule wrote:
> by the way - this feature is undocumented - I though so only value 
> used as type holder is not used.
>
> Should be documented better, - if I understand - it is base stone for 
> implementation #= hstore operator
>
> some nice example
>
> postgres=# select json_populate_record('(10,20)'::pt, '{"a":30}');
>  json_populate_record
> ----------------------
>  (30,20)
> (1 row)
>
> a mentioned bug is corner case - bugfix is simple
>
> diff --git a/src/backend/utils/adt/jsonfuncs.c 
> b/src/backend/utils/adt/jsonfuncs.c
> new file mode 100644
> index a8cdeaa..6e83f78
> *** a/src/backend/utils/adt/jsonfuncs.c
> --- b/src/backend/utils/adt/jsonfuncs.c
> *************** populate_record_worker(FunctionCallInfo
> *** 2114,2119 ****
> --- 2114,2122 ----
>                  */
>                 if (hash_get_num_entries(json_hash) == 0 && rec)
>                 {
> +                       if (have_record_arg)
> +                               ReleaseTupleDesc(tupdesc);
> +
>                         hash_destroy(json_hash);
>                         PG_RETURN_POINTER(rec);
>                 }
>
>
>
>


This doesn't look quite right. Shouldn't we unconditionally release the 
Tupledesc before the returns at lines 2118 and 2127, just as we do at 
the bottom of the function at line 2285?

cheers

andrew