Re: REVIEW: WIP: plpgsql - foreach in

On Mon, Jan 24, 2011 at 20:10, Pavel Stehule <pavel(dot)stehule(at)gmail(dot)com> wrote:
> we have to iterate over array's items because it allow seq. access to
> array's data. I need a global index for function "array_get_isnull". I
> can't to use a buildin functions like array_slize_size or
> array_get_slice, because there is high overhead of array_seek
> function. I redesigned the initial part of main cycle, but code is
> little bit longer :(, but I hope, it is more readable.

The attached patch only includes changes in plpgsql. I tested it
combined with the previous one, that includes other directories.

* src/pl/plpgsql/src/gram.y
| for_variable K_SLICE ICONST
The parameter for SLICE must be an integer literal. Is it a design?
I got a syntax error when I wrote a function like below. However,
FOREACH returns different types on SLICE = 0 or SLICE >= 1.
(element type for 0, or array type for >=1) So, we cannot write
a true generic function that accepts all SLICE values even if
the syntax supports an expr for the parameter.

CREATE FUNCTION foreach_test(int[], int) RETURNS SETOF int[] AS
$$
DECLARE
a integer[];
BEGIN
FOREACH a SLICE $2 IN $1 LOOP -- syntax error
RETURN NEXT a;
END LOOP;
END;
$$ LANGUAGE plpgsql;

* /doc/src/sgml/plpgsql.sgml
+ FOREACH <replaceable>target</replaceable> <optional> SCALE
s/SCALE/SLICE/ ?

* Why do you use "foreach_a" for some identifiers?
We use "foreach" only for arrays, so "_a" suffix doesn't seem needed.

* accumArrayResult() for SLICE has a bit overhead.
If we want to optimize it, we could use memcpy() because slices are
placed in continuous memory. But I'm not sure the worth; I guess
FOREACH will be used with SLICE = 0 in many cases.

--
Itagaki Takahiro