On Wed, Oct 12, 2011 at 15:33, Alex Hunsaker <badalex@gmail.com> wrote:
> On Wed, Oct 12, 2011 at 15:00, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> The core of the problem seems to be that if SvROK(sv) then
>> the code assumes that it must be intended to convert that to an array or
>> composite, no matter whether the declared result type of the function is
>> compatible with such a thing.
>
> Hrm, well 9.0 and below did not get this "right" either:
> create or replace function test_hash() returns text as $$ return
> {'a'=>1}; $$ language plperl;
> select test_array();
> test_array
> -----------------------
> ARRAY(0x7fd92384dcb8)
> (1 row)
>
> create or replace function test_hash() returns text as $$ return
> {'a'=>1}; $$ language plperl;
> select test_hash();
> test_hash
> ----------------------
> HASH(0x7fd92387f848)
> (1 row)
>
> Given the output above (both pre 9.1 and post) it seems unless the
> type is a set or composite we should throw an error. Maybe "PL/Perl
> function returning type %s must not return a reference" ?
>
>> It would be more appropriate to drive the
>> cases off the nature of the function result type, perhaps.
>
> Ill see if I can cook up something that's not too invasive.
PFA my attempt at a fix.
This gets rid of of most of the if/else chain and the has_retval crap
in plperl_handl_func(). Instead we let plperl_sv_to_datum() do most of
the lifting. It also now handles VOIDOID and checks that the request
result oid can be converted from the perl structure. For example if
you passed in a hashref with a result oid that was not an rowtype it
will error out with "PL/Perl cannot convert hash to non rowtype %s".
Arrays behave similarly.
One side effect is you can now return a composite literal where you
could not before. ( We already let you return array literals )
The comments might still be a bit sparse-- Im hoping the added errors
make things a bit more self explanatory.
A large portion of the diff is added regression tests, testing what
happens when you return various references.
Comments?