[
https://issues.apache.org/jira/browse/CALCITE-5552?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17695959#comment-17695959
]
Julian Hyde commented on CALCITE-5552:
--------------------------------------
I did some investigation. (Thanks for the excellent repro project, by the way!)
The problem seems to be this. the {{JdbcResultSet.frame}} method is called two
times:
* First, with offset = 0, and a list of 100 rows, from {{JdbcResultSet.create}}
(line 101), with a {{Calendar}} whose offset is 0 (UTC);
* Second, with offset = 100, and a list of 1 row, {{JdbcMeta.fetch}} (line
825), with a {{Calendar}} whose offset is -8h (my local time).
The result of the different calendar offsets is that timestamps are shifted in
the second fetch, which they should not be;
the cause of the different calendars is that {{JdbcResultSet.create}} gets its
calendar by calling {{DateTimeUtils.calendar()}} (line 90) and
{{JdbcMeta.calendar}} is initialized using {{Unsafe.localCalendar()}}.
{{JdbcResultSet}} is in Avatica, therefore this is an Avatica bug.
> Returned timestamp is incorrect after the 100th row
> ---------------------------------------------------
>
> Key: CALCITE-5552
> URL: https://issues.apache.org/jira/browse/CALCITE-5552
> Project: Calcite
> Issue Type: Bug
> Affects Versions: 1.33.0
> Reporter: Magnus Mogren
> Priority: Critical
> Attachments: TSTAMPS.csv
>
>
> When fetching data that contains timestamps the returned timestamp after row
> 100 is incorrect.
> This can be reproduced using the CSV adapter (calcite-csv) using the
> TSTAMPS.csv file attached. It contains 101 timestamps with the value
> 1900-01-01 00:00:00. The first 100 is returned correctly, but number 101 in
> the result has the value 1899-12-31 23:00:00 instead.
> Marking this bug as critical since not beeing able to trust the values
> returned by calcite is as bad as it gets in my opinion.
> I do not know if the bug is in calcite or avatica.
> I have created a project that reproduces the issue. You can find that here:
> [nytro77/calcite-timestamp-bug: Showcase bug in Calcite or Avatica that
> causes faulty timestamps to be returned
> (github.com)|https://github.com/nytro77/calcite-timestamp-bug]
> It starts an AvaticaServer that serves the attached file as a table using
> calcite-csv and a unit tests that connects to the server and fetches the data.
> Run it withÂ
> {code:java}
> ./mvnw test{code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
