[
https://issues.apache.org/jira/browse/PHOENIX-66?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13908360#comment-13908360
]
Gabriel Reid commented on PHOENIX-66:
-------------------------------------
Hive handles something very similar by using a separate collection item
terminator, defined with the "COLLECTION ITEMS TERMINATED BY" clause in DDL.
This would mean that a CSV file with an array would be encoded as something
like this if semicolon was used as the array value separator:
a,b,c,foo;1;bar,d
I'm thinking that using a separate collection value separator would probably be
easiest in terms of parsing logic, as well as making it easier to deal with the
same file in Hive or Pig or any other tool that wouldn't be aware of a
variable-width record encoding.
> Support array creation from CSV file
> ------------------------------------
>
> Key: PHOENIX-66
> URL: https://issues.apache.org/jira/browse/PHOENIX-66
> Project: Phoenix
> Issue Type: Bug
> Reporter: James Taylor
> Fix For: 3.0.0
>
>
> We should support being able to parse an array defined in our CVS file.
> Perhaps something like this:
> a, b, c, [foo, 1, bar], d
> We'd know (from the data type of the column), that we have an array for the
> fourth field here.
> One option to support this would be to implement the
> PDataType.toObject(String) for the ARRAY PDataType enums. That's not ideal,
> though, as we'd introduce a dependency from PDataType to our CSVLoader, since
> we'd need to in turn parse each element. Also, we don't have a way to pass
> through the custom delimiters that might be in use.
> Another pretty trivial, though a bit more constrained approach would be to
> look at the column ARRAY_SIZE to control how many of the next CSV columns
> should be used as array elements. In this approach, you wouldn't use the
> square brackets at all. You can get the ARRAY_SIZE from the column metadata
> through connection.getMetaData().getColumns() call, through
> resultSet.getInt("ARRAY_SIZE"); However, the ARRAY_SIZE is optional in a DDL
> statement, so we'd need to do something for the case where it's not specified.
> A third option would be to handle most of the parsing in the CSVLoader. We
> could use the above bracket syntax, and then collect up the next set of CSV
> field elements until we hit the unescaped ']'. Then we'd use our standard
> JDBC APIs to build the array and continue on our merry way.
> What do you think, [~jviolettedsiq]? Or [~bruno], maybe you can take a crack
> at it?
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
