Re: Early WIP/PoC for inlining CTEs

On Mon, Jan 21, 2019 at 10:28 AM Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Andreas Karlsson <andreas@proxel.se> writes:
> > I have a minor biksheddish question about the syntax.
> > You proposed:
> > WITH cte_name [[NOT] MATERIALIZED] AS (query) main_query
> > While Andrew proposed:
> > WITH cte_name AS [[NOT] MATERIALIZED] (query) main_query
> > Do people have any preference between these two?
>
> FWIW, I'd independently thought that the latter is more readable,
> and probably less likely to have syntax problems with future
> extensions (since AS is already fully reserved).  Didn't get
> around to mentioning it yet, but +1 for putting AS first.

It seems to me that as long as the query has to be surrounded by
non-optional parentheses and the options list never starts with a
parenthesis, there isn't much room for a grammar conflict either way.
If the query didn't have to be surrounded by parentheses, you would
want to put the options before the word AS so that the word AS would
serve as an unambiguous terminator for the options specification, but
if the query must be preceded by an opening parenthesis then as long
as the options list can't include an option that begins with such a
parenthesis we are in good shape.

However, generally we have not had great luck with just sticking
keywords in there (cf. VACUUM, ANALYZE, EXPLAIN, COPY) which is why I
suggested using a flexible syntax with parenthesized options.  And
then you're going to have trouble getting bison to figure out what to
do after...

WITH something AS ( some_keyword

...because some_keyword might be the an option name or the first word
of a query.  And putting that syntax before AS isn't any better,
because now it can be confused with a column name list.

I am not deeply worked up about this, just proposing some things to think about.

-- 
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company