Yan Zhou commented on PIG-1026:

Another problem is that during STORE, when  handleMapSplit method is called to 
set up the CG schema mapping with the Table schema, an incremented index was 
used as the column indices to the CG schemas. This works ok if the columns in 
the CG schemas are in the same order as in the table schema, but will be wrong 
if the orders are different, causing no values to be stored in some MAP columns 
in any CG and hence, during LOAD, getValue() returns null.

> [zebra] map split returns null
> ------------------------------
>                 Key: PIG-1026
>                 URL: https://issues.apache.org/jira/browse/PIG-1026
>             Project: Pig
>          Issue Type: Bug
>    Affects Versions: 0.6.0
>            Reporter: Jing Huang
>            Assignee: Yan Zhou
>             Fix For: 0.6.0
>         Attachments: PIG_1026.patch
> Here is the test scenario:
>  final static String STR_SCHEMA = "m1:map(string),m2:map(map(int))";
>   //final static String STR_STORAGE = "[m1#{a}];[m2#{x|y}]; [m1#{b}, 
> m2#{z}];[m1]";
>  final static String STR_STORAGE = "[m1#{a}, m2#{x}];[m2#{x|y}]; [m1#{b}, 
> m2#{z}];[m1,m2]";
> projection: String projection2 = new String("m1#{b}, m2#{x|z}");
> User got null pointer exception on reading m1#{b}.
> Yan, please refer to the test class:
> TestNonDefaultWholeMapSplit.java 

This message is automatically generated by JIRA.
You can reply to this email to add a comment to the issue online.

Reply via email to