Re: [COMMITTERS] pgsql: Arrange for quote_identifier() and pg_dump to not quote keywords

Поиск
Список
Период
Сортировка
От Gregory Stark
Тема Re: [COMMITTERS] pgsql: Arrange for quote_identifier() and pg_dump to not quote keywords
Дата
Msg-id 87fy4ej40e.fsf@oxford.xeocode.com
обсуждение исходный текст
Ответы Re: [COMMITTERS] pgsql: Arrange for quote_identifier() and pg_dump to not quote keywords  (Gregory Stark <stark@enterprisedb.com>)
Re: [COMMITTERS] pgsql: Arrange for quote_identifier() and pg_dump to not quote keywords  (Gregory Stark <stark@enterprisedb.com>)
Список pgsql-hackers
"Tom Lane" <tgl@postgresql.org> writes:

> For the moment, lie about WITH's status in the table so it will still get
> quoted --- this is because of the expectation that WITH will become reserved
> when the SQL recursive-queries patch gets done.

Out of curiosity I'm checking what consequences some other future grammer
changes might have for us. Today I checked out full spec compliant GROUP BY
syntax including ROLLUP, CUBE, and GROUPING SETS.

There are two conclusions of note:

1) ROLLUP and CUBE would have to be col_name_keyword keywords. 
  That could be an annoyance for the cube contrib package because it defines  a few constructors like cube(float8[]).
Youcould still have a type named  "cube" but the functions would have to be renamed. Personally I always  found "cube"
astrange name anyways, I think of this data type more as a  n-space vector than a cube anyways.
 
  quote_identifier would start quoting "cube" and "rollup" everywhere. My  first inclination was that it's probably not
necessaryto start  preemptively quoting them this release because people are more likely to  use them as column names
thanfunction names anyways. But perhaps that's  not true given the contrib module.
 

2) Assuming we keep our extension of allowing arbitrary expressions in GROUP  BY lists then there is a conflict between
ourundecorated row constructor  '(' expr_list ')' and the spec which allows parenthesized sublists in the  grouping
list.
  I'm not sure this is a real problem though. As near as I can tell the  semantics of grouping by a ROW(a,b) and
groupingby columns (a,b) as a  grouping set element are basically the same anyways. So I think we can just  accept any
arbitraryexpression including row constructors as what the spec  calls an "ordinary grouping set".
 




For what it's worth here's the grammar I get by basically copying the grammar
straight out of the spec and then cleaning up the conflicts including making
ordinary_grouping_set a straight expr_list as described above:

opt_group_set_clause:           DISTINCT           | ALL           | /*EMPTY*/       ;

group_clause:           GROUP_P BY opt_group_set_clause grouping_element_list           | /*EMPTY*/       ;

grouping_element_list:   grouping_element   | grouping_element_list ',' grouping_element   ;

grouping_element:   a_expr   | rollup_list   | cube_list   | grouping_sets_specification   | empty_grouping_set   ;

rollup_list:   ROLLUP '(' expr_list ')'   ;

cube_list:   CUBE '(' expr_list ')'   ;

grouping_sets_specification:   GROUPING SETS '(' grouping_set_list ')'   ;

grouping_set_list:   grouping_set   | grouping_set_list ',' grouping_set   ;

grouping_set:   a_expr   | rollup_list   | cube_list   | grouping_sets_specification   | empty_grouping_set   ;

empty_grouping_set: '(' ')'   ;

--  Gregory Stark EnterpriseDB          http://www.enterprisedb.com



В списке pgsql-hackers по дате отправления:

Предыдущее
От: Robert Treat
Дата:
Сообщение: Re: How do we create the releases?
Следующее
От: Gregory Stark
Дата:
Сообщение: Re: [COMMITTERS] pgsql: Arrange for quote_identifier() and pg_dump to not quote keywords