create type pt as (a int, b int); postgres=# select json_populate_record('(10,20)'::pt, '{}'); WARNING: TupleDesc reference leak: TupleDesc 0x7f413ca325b0 (16560,-1) still referenced
On Thu, Feb 26, 2015 at 05:31:44PM -0500, Tom Lane wrote: > Andrew Dunstan <andrew@dunslane.net> writes: > > 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? > > I think Pavel's patch is probably OK as-is, because the tupdesc returned > by get_call_result_type isn't reference-counted; but I agree the code > would look cleaner your way. If the main exit isn't bothering to > distinguish this then the early exits should not either. > > What I'm wondering about, though, is this bit at line 2125: > > /* same logic as for json */ > if (!have_record_arg && rec) > PG_RETURN_POINTER(rec); > > If that's supposed to be the same logic as in the other path, then how > is it that !have_record_arg has anything to do with whether the JSON > object is empty? Either the code is broken, or the comment is.
