[
https://issues.apache.org/jira/browse/DERBY-4079?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12678967#action_12678967
]
Rick Hillegas commented on DERBY-4079:
--------------------------------------
Thanks for the experimental patch, Dag. The overall approach looks good to me.
I have a couple small comments:
ResultColumn
The new methods have generic (possibly cloned?) javadoc.
CursorNode
It seems to me that since the new offset and fetchFirst fields are known to be
NumericConstantNodes, then they should be declared as the most refined type
possible, rather than as generic ValueNodes. I think this will make the code
easier to understand.
DMLStatementNode
Thanks for the explanation of why you are wrapping the inner result set at that
particular point in the code. I find that these explanations are easier to
understand if they include a sample problem query which would break if the code
were placed somewhere else.
sqlgrammar.jj
Do the new FETCH and OFFSET keywords need to be reserved? I realize that the
standard may suggest that they should be, but there are plenty of examples of
Derby downgrading reserved keywords to non-reserved keywords in order to avoid
breaking legacy applications. Same question for ROW, which appears to have been
moved from non-reserved to reserved status.
Similar comment to that for CursorNode: if the new clauses are just
NumericConstantNodes, then it makes sense to me to declare them as that refined
type.
I wonder if the offsetClause() and fetchFirstClause() productions could use
numericLiteral() rather than literal(). With a little more work, you might be
able to get the parser to guarantee that the literal is a non-negative integer
and you might be able to skip this check at bind() time.
GenericResultSetFactory
I wonder if all of the duplicated debug code could be pulled up into a
superclass constructor. Something like
if (SanityManager.DEBUG &&
SanityManager.DEBUG_ON("DumpResultSetGeneration")) {
SanityManager.GET_DEBUG_STREAM().print("Generated " +
getClass().getName() + "\n");
}
> Add support for SQL:2008 <result offset clause> and <fetch first clause> to
> limit result set cardinality
> --------------------------------------------------------------------------------------------------------
>
> Key: DERBY-4079
> URL: https://issues.apache.org/jira/browse/DERBY-4079
> Project: Derby
> Issue Type: New Feature
> Components: SQL
> Reporter: Dag H. Wanvik
> Assignee: Dag H. Wanvik
> Fix For: 10.5.0.0
>
> Attachments: derby-4079-1.diff, derby-4079-1.stat,
> derby-4079-docs-1.diff, derby-4079-docs-1.stat, derby-4079-docs-1.zip, ref.zip
>
>
> SQL 2008 has added new syntax to support a direct way to limit the
> returned set of rows in a result set. This allows an application to
> retrieve only some rows of an otherwise larger result set, similar to
> the popular LIMIT clauses use in some databases.
> Up till now, in Derby (and SQL) we have had to use the ROW_NUMBER()
> function in a nested subquery to achieve the effect of the <fetch
> first clause>, cf. DERBY-2998, a method which is rather more indirect
> and still not efficient (DERBY-3505), and primarily intended for OLAP
> functionality, perhaps.
> There has been no direct way to achieve the effect of the <result
> offset clause> via SQL.
> Syntax (cf. SQL 2008, section 7.13):
> <result offset clause> ::= OFFSET <n> {ROW | ROWS}
> <fetch first clause> ::= FETCH {FIRST | NEXT} [<n>] {ROW | ROWS}
> ONLY
> where <n> is an integer. The two clauses syntactically follow the ORDER BY
> clause in the grammar.
> Note that both ORDER BY and the new clauses above are allowed also in
> subqueries in the new version of the SQL standard (section 7.13). I
> only propose to include this at the top level in DERBY for now. (ORDER
> BY is presently also not allowed in subqueries in Derby since SQL
> didn't allow for this until SQL 2008 either).
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
