Re: idea: allow AS label inside ROW constructor - Mailing list pgsql-hackers

2014-10-22 18:35 GMT+02:00 Merlin Moncure <mmoncure@gmail.com>:

On Wed, Oct 22, 2014 at 11:21 AM, Pavel Stehule <pavel.stehule@gmail.com> wrote:
> Hi
>
> with new functions row_to_json(b), there is more often usage of ROW
> constructor. Using names in fields is relative difficult. Because ROW has
> special clause in parser, I am thinking so we can enable labeling inside ROW
> constructor
>
> so instead currently supported:
>
> select row_to_json(r) from (select 10 as a, 20 as b) r;
>
> users can to write:
>
> select row_to_json(row(10 as a,20 as b));
>
> labeling will be enabled "only" inside ROW constructor. I don't propose
> enable it everywhere.
>
> What do you think about it?

It's a neat idea -- maybe a better alternative to what I was thinking
here: http://postgresql.1045698.n5.nabble.com/Support-UPDATE-table-SET-tp5823073p5823944.html

Some questions:
*) What would the parser transformation resolve to

row:            ROW '(' expr_list ')'                                 { $$ = $3; }
                        | ROW '(' ')'                                                   { $$ = NIL; }
                        | '(' expr_list ',' a_expr ')'                  { $$ = lappend($2, $4); }
                ;

we can replace a expr_list by target_list. I know only so it doesn't enforce a problems with gramatic  - bison doesn't raise any warning.

*) Are we ok with SQL standard

SQL standard doesn't think named attributes in row - so it is out of range ANSI. But it is not in conflict with standard. "AS name" is used more in SQL/MM, SQL/XML -- and function named parameters has different syntax "parameter_name <= value" - I checked it against SQL99.

 

*) Do you think this (or some similar variant) would work?

select row_to_json(row(foo.*)) from foo;

It looks like independent feature and can work too - it is more natural for user.

 

merlin