Thread: Implicit casting for || (was Re: grant execute on many functions)

Implicit casting for || (was Re: grant execute on many functions)

From
Tom Lane
Date:
[ moving thread to a more appropriate list ]

Joe Conway <mail@joeconway.com> writes:
> Tom Lane wrote:
>> select 'GRANT EXECUTE ON ' || 1142::regprocedure;
>> ERROR:  array value must start with "{" or dimension information
>> 
>> BTW, it seems like there's something pretty broken here.  How did
>> arrays get into it?

> regression=# \df array_append
>                            List of functions
>   Result data type |   Schema   |     Name     | Argument data types
> -------------------+------------+--------------+----------------------
>   anyarray         | pg_catalog | array_append | anyarray, anyelement
> (1 row)

> So the "||" operator sees (unknown, regprocedure), and make_op tries to 
> coerce the unknown literal to an array of regprocedure, which of course 
> fails. If instead the literal is explicitly cast:

> select 'GRANT EXECUTE ON '::text || 1142::regprocedure;
> ERROR:  operator does not exist: text || regprocedure
> HINT:  No operator matches the given name and argument type(s). You may 
> need to add explicit type casts.

> I'm not clear on how we can do better :(. Any suggestions?

I'm not sure either, but I can't say that I like this interpretation.
I did find that if there is an implicit coercion to text then the system
will prefer a saner interpretation:

regression=# select 'GRANT EXECUTE ON ' || 1142::regprocedure;
ERROR:  array value must start with "{" or dimension information

regression=# create function text(regprocedure) returns text as '
regression'# begin
regression'#   return $1;
regression'# end' language plpgsql stable strict;
CREATE FUNCTION
regression=# create cast (regprocedure as text) with function text(regprocedure) as implicit;
CREATE CAST

regression=# select 'GRANT EXECUTE ON ' || 1142::regprocedure;               ?column?
-----------------------------------------GRANT EXECUTE ON date_mii(date,integer)
(1 row)

I'm worried though about how stable this choice is.  I'm almost tempted
to add a wart in the coercion routines to discourage matching "unknown"
to "anyarray" ...
        regards, tom lane