[
https://issues.apache.org/jira/browse/DRILL-5549?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16028633#comment-16028633
]
Paul Rogers commented on DRILL-5549:
------------------------------------
Will check in a (disabled) test: {{TestCsvHeaders.testEmptyHeaderStar()}}.
> SELECT * against a CSV file with empty headers produces error
> -------------------------------------------------------------
>
> Key: DRILL-5549
> URL: https://issues.apache.org/jira/browse/DRILL-5549
> Project: Apache Drill
> Issue Type: Bug
> Affects Versions: 1.10.0
> Reporter: Paul Rogers
> Priority: Minor
>
> See DRILL-5548 for background. This test is very similar except that the
> input file has a single blank line. Since the CSV plugin is set up to read
> headers, this represents a non-empty file, with no headers and no data.
> Again use a star query:
> {code}
> SELECT * FROM `dfs.data.emptyHeader.csv`
> {code}
> The result this time is somewhat different:
> {code}
> org.apache.drill.common.exceptions.UserRemoteException:
> SYSTEM ERROR: HeaderError: The file must define at least one header.
> {code}
> If we fix the issue in DRILL-5548 (empty file), we should use the same
> solution for a file with an empty header.
> Suppose the file was not entirely empty, suppose it was:
> {code}
> 1:
> 2: fred
> 3: barney
> {code}
> (Note: line numbers shown to force display of blank line...)
> Here, we have an empty header, but we have data. We can't know that we have
> data while reading the header. While this is a pathological case, and an
> invalid CSV file, this second case does suggest that the right solution to
> the two empty cases is to use the special {{columns}} array when the header
> is empty. This will allow Drill to gracefully handle the case above for a
> file with no header but with data.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
