db-derby-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Knut Anders Hatlen (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (DERBY-6566) Simplify handling of untyped nulls in CASE and NULLIF expressions
Date Wed, 28 May 2014 09:59:02 GMT

    [ https://issues.apache.org/jira/browse/DERBY-6566?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14010985#comment-14010985
] 

Knut Anders Hatlen commented on DERBY-6566:
-------------------------------------------

That's correct. This change only streamlines how ConditionalNode.bindExpression() infers the
type of an untyped NULL. It's just a compiler change. No change to the stored format.

The possible compatibility issue is that the new logic does not attempt to infer the type
of an already typed NULL. So if you for example say WHEN a=b THEN CAST(NULL AS BLOB), and
the context requires the result to be an INTEGER, we now raise a type-mismatch error instead
of silently rewriting CAST(NULL AS BLOB) to CAST(NULL AS INTEGER). If you soft downgrade to
an earlier version, you'll get the old behaviour again.

I agree that this change should not be backported.

> Simplify handling of untyped nulls in CASE and NULLIF expressions
> -----------------------------------------------------------------
>
>                 Key: DERBY-6566
>                 URL: https://issues.apache.org/jira/browse/DERBY-6566
>             Project: Derby
>          Issue Type: Improvement
>          Components: SQL
>    Affects Versions: 11.0.0.0
>            Reporter: Knut Anders Hatlen
>            Assignee: Knut Anders Hatlen
>            Priority: Minor
>              Labels: derby_backport_reject_10_10
>             Fix For: 10.11.0.0
>
>         Attachments: d6566-1a.diff, releaseNote.html
>
>
> The parser translates both CASE and NULLIF expressions into ConditionalNodes, but it
represents untyped NULLs differently in the two cases.
> In a CASE expression, any branch that is an untyped NULL, is translated into an UntypedNullConstantNode
that's wrapped in a CastNode that casts the value to CHAR(1). The CastNode is replaced with
a cast to the correct type during the bind phase.
> A NULLIF expression is turned into a CASE expression that has a THEN NULL clause. The
parser simply creates an UntypedNullConstantNode for that clause, without wrapping it in a
CastNode. A CastNode is instead added during the bind phase.
> This slight difference in how NULLs are represented by the parser in the two cases, means
that ConditionalNode needs to handle the two cases differently during the bind phase. It would
be better if the parser generated NULLs in the same way for the two cases, so that ConditionalNode
didn't need to know if it was generated for a CASE expression or a NULLIF expression.



--
This message was sent by Atlassian JIRA
(v6.2#6252)

Mime
View raw message