Thread: Allowed to return possibly TOASTed datums?

Hi,

I would like to implement a function that simply returns its argument
(for the unary plus operator of a custum data type).

Is it allowed to use the following function, assuming the data type is
TOASTable?

PG_FUNCTION_INFO_V1(mytype_pos);
Datum mytype_pos(PG_FUNCTION_ARGS) {                             
  PG_RETURN_DATUM(GETARG_DATUM(0));
}

CREATE FUNCTION "mytype_pos"("mytype") RETURNS "mytype"
  STRICT IMMUTABLE LANGUAGE c AS 'mylib.so', 'mytype_pos';

CREATE OPERATOR + (
  rightarg = "mytype",
  function = "mytype_pos"
);

Or do I need to always return a DETOASTed value? In addition to the
answer, I would like to know where I can find such information. It
seems to be difficult (for me) to find it in the source. Is there any
other documentation I can use, or any hints on WHERE in the source I
find more information about when to DETOAST and when not?

Also: Is my approach an idiomatic way to implement unary plus? Are there
pre-existing functions for that?

Many thanks in advance for your help and kind regards

Jan Behrens

Jan Behrens <jbe-mlist@magnetkern.de> writes:
> Is it allowed to use the following function, assuming the data type is
> TOASTable?

> PG_FUNCTION_INFO_V1(mytype_pos);
> Datum mytype_pos(PG_FUNCTION_ARGS) {                             
>   PG_RETURN_DATUM(GETARG_DATUM(0));
> }

Absolutely.  You can find things equivalent to that in the
core code, I think.

            regards, tom lane

Ron Johnson <ronljohnsonjr@gmail.com> writes:
> Since the purpose of TOAST is just (I think) to store "large field values"
> on disk by dividing into multiple chunks, what would be the purpose of
> returning the TOASTed value to Jan's function?

Probably not much.  There are scenarios where not detoasting
saves work overall, because the object never has to get detoasted;
but I think that's rare.  It'd probably look like

    UPDATE t SET c = no_op_function(c) WHERE ...

which is something few people would write.

            regards, tom lane