[
https://issues.apache.org/jira/browse/CALCITE-806?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14632770#comment-14632770
]
Vladimir Sitnikov commented on CALCITE-806:
-------------------------------------------
I don't like discussing the spec, however it looks like we don't have a choice.
{quote}Does that mean that it is illegal to use ROW_NUMBER with any other kind
of bound?{quote}
This is exactly how I read the following in the spec:
{quote}
If <rank function type> or ROW_NUMBER is specified, then:
b) The window framing of WDX shall not be present.{quote}
In other words, all rank functions+row_number cannot have framing.
I believe we have a validation exactly for this reason:
{{RankWithFrame=ROW/RANGE not allowed with RANK or DENSE_RANK functions}}
Well, it makes sense adding ROW_NUMBER there as well (or parametrize the error
message with actual name of the function).
Ok, we continue:
{quote}{{ROW_NUMBER() OVER WNS}} is equivalent to the <window function>:
{{COUNT \(*) OVER (WNS1 ROWS UNBOUNDED PRECEDING)}}{quote}
I read this as follows: whenever you see {{ROW_NUMBER() OVER WNS}}, you take
window specification (remember, it does not contain framing), overlay it with
{{ROWS UNBOUNDED PRECEDING}}, and replace {{ROW_NUMBER}} with {{COUNT\(*)}}.
PostgreSQL just silently ignores {{ROWS BETWEEN 3 PRECEDING AND 2 FOLLOWING}}
in row_number/rank/dense_rank.
Oracle does not allow framing for rank/row_number.
> ROW_NUMBER should emit distinct values
> --------------------------------------
>
> Key: CALCITE-806
> URL: https://issues.apache.org/jira/browse/CALCITE-806
> Project: Calcite
> Issue Type: Bug
> Reporter: Julian Hyde
> Assignee: Julian Hyde
> Fix For: next
>
>
> ROW_NUMBER should emit distinct values even for rows that have identical sort
> keys, or if there is no sort key. (This is distinct from RANK and DENSE_RANK,
> which emit the same value for rows with the same sort key.)
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
