Re: MULTISET patch

On Mon, Dec 27, 2010 at 20:15, Pavel Stehule <pavel(dot)stehule(at)gmail(dot)com> wrote:
> * trim_array - you use a deconstruct_array. It unpack  all fields and
> it could not be effective. Can we limit a unpacked array?

Sure, I'll optimize it.

> I searched on net. This function has a little bit unconptual name -
> DB2 use a synonym for this function array_trim. Can we use this
> synonym too?

IBM DB2 does use TRIM_ARRAY for the name, no? I believe it's the standard.
http://publib.boulder.ibm.com/infocenter/db2luw/v9r7/index.jsp?topic=/com.ibm.db2.luw.apdv.sqlpl.doc/doc/t0053491.html

> Probably there could be a low level optimization - can we limit a
> detoast processing? (It must not be a part of this patch).

I think we could avoid deconstruct_array() in some spaces,
but cannot avoid detoasting.

> Questions:
> should be a MULTISET, SUBMULTISET, MEMBER a reserved keywords?
> I am for marking these words as reserved keywords, but it needs a wide
> agreeement. Without agreement, I don't think so not optimal keyword
> "OF" in MEMBER operator is significant issue.

They are full-reserved keywords in the spec, but I'd like not to
reserve them until we can do nothing but do so.

To be honest, I cannot fix shift/reduce errors in an optional OF
in the syntax even if I marked those variables as reserved keywords.
Can I ask for your help about the usage of bison/flex for such case?

+ /* FIXME: OF is an option in the SQL standard, but I cannot solve
+ shift/reduce errors without OF. To solve the errors, we might need
+ to make OF, MEMBER, and/or SUBMULTISET to reserved keywords. They
+ are reserved keywords in the SQL standard.
+ */

--
Itagaki Takahiro