Обсуждение: idea: simple variadic functions in SQL and PL/pgSQL

Hello,

I found easy implementation of variadic functions. It's based on
adapation FuncnameGetCandidates. When I found variadic function, then
I should create accurate number of last arguments (diff between
pronargs and nargs). Variadic function can be signed via flag or via
some pseudotype. Flag is better - allows variadic arguments of any
type. In static languages (like SQL or PL/pgSQL) variadic variables
can ba accessed via array (variadic arguments can be only nonarray).
This isn't problem in C language, there are arguments available
directly.

Sample:

CREATE OR REPLACE FUNCTION Least(anyelement)
RETURNS anyelement AS $$ SELECT MIN($1[i])    FROM generate_series(1, array_upper($1,1)) g(i);
$$  LANGUAGE SQL IMMUTABLE VARIADIC.

This sample is really simple. The goal is support sophistic libraries
like JSON support: http://www.mysqludf.org/lib_mysqludf_json/index.php

Main change in FuncnameGetCandidates.

if (!OidIsValid(variadic_oid))
{               memcpy(newResult->args, procform->proargtypes.values,                                 pronargs *
sizeof(Oid));
}
else
{               int     j;               /* copy nonvariadic parameters */               memcpy(newResult->args,
procform->proargtypes.values,                                  pronargs * sizeof(Oid));               /* set variadic
parameters,!!!!!!!!!!! */               for (j = pronargs - 1; j < nargs; j++)
newResult->args[j]= variadic_oid;
 
}

I invite any ideas, notes

Regards
Pavel Stehule

Pavel Stehule wrote:
> Hello,
>
> I found easy implementation of variadic functions. It's based on
> adapation FuncnameGetCandidates. When I found variadic function, then
> I should create accurate number of last arguments (diff between
> pronargs and nargs). Variadic function can be signed via flag or via
> some pseudotype. Flag is better - allows variadic arguments of any
> type. In static languages (like SQL or PL/pgSQL) variadic variables
> can ba accessed via array (variadic arguments can be only nonarray).
> This isn't problem in C language, there are arguments available
> directly.
>   

There are a whole slew of questions around this, ISTM.

For example: What will be the type inferred for the array of variadic 
args in plpgsql?

If we are going to do this I think we need some discussion on design 
before we rush into it.

cheers

andrew

Hello

On 26/02/2008, Andrew Dunstan <andrew@dunslane.net> wrote:
>
>
>  Pavel Stehule wrote:
>  > Hello,
>  >
>  > I found easy implementation of variadic functions. It's based on
>  > adapation FuncnameGetCandidates. When I found variadic function, then
>  > I should create accurate number of last arguments (diff between
>  > pronargs and nargs). Variadic function can be signed via flag or via
>  > some pseudotype. Flag is better - allows variadic arguments of any
>  > type. In static languages (like SQL or PL/pgSQL) variadic variables
>  > can ba accessed via array (variadic arguments can be only nonarray).
>  > This isn't problem in C language, there are arguments available
>  > directly.
>  >
>
>
> There are a whole slew of questions around this, ISTM.

sure. It's time to thing about it.

>
>  For example: What will be the type inferred for the array of variadic
>  args in plpgsql?

SQL and PL/SQL should be limited.  In my prototype I use fragment:
               /* replace ANYNONARRAYS argument with ARRAY OF some */              argtype = get_array_type(argtype);
          if (!OidIsValid(argtype))                                   ereport(ERROR,
   (errcode(ERRCODE_DATATYPE_MISMATCH),                                         errmsg("could not determine
 
actual array type for variadic argument")));

SPI is unlimited (C language). I can use list of arrays in perl or
python - there I don't see complications too. Domains and arrays can
be problem. Using variadic array variables isn't typical. So I will to
solve only domains. But it's more general problem. Maybe 8.4 can
support array of domains.

>
>  If we are going to do this I think we need some discussion on design
>  before we rush into it.
>

yes, I agree. Variadic functions are simple. And I belive it can
simplify life to some programmers that have to manage functions like:

foo(int),
foo(int, int)
foo(int, int, int)...

regards
Pavel



>  cheers
>
>
>  andrew
>
>
>
>