> On May 3, 2017, 5:34 a.m., Prasanth_J wrote:
> > llap-client/src/java/org/apache/hadoop/hive/llap/LlapRowRecordReader.java
> > Lines 154 (patched)
> > <https://reviews.apache.org/r/58934/diff/1/?file=1706163#file1706163line174>
> >
> >     IIRC, there are utilities already to do this in ObjectInspectorUtils.. 
> > copyToXXX() methods.. can that be reused?
> 
> Jason Dere wrote:
>     I had not seen the methods in ObjectInspectorUtils, yeah we might be able 
> to use this.

copyToXXX() ends up saving decimal/char/varchar values to their Hive 
equivalents (HiveDecimal/HiveChar/HiveVarchar). I'd prefer the row interface 
expose these values as non-Hive classes (BigDecimal/String/String) for ease of 
use. So I think it is better off not using the ObjectInspecterUtils.copyToXXX 
methods.


> On May 3, 2017, 5:34 a.m., Prasanth_J wrote:
> > llap-common/src/java/org/apache/hadoop/hive/llap/TypeDesc.java
> > Lines 154 (patched)
> > <https://reviews.apache.org/r/58934/diff/1/?file=1706165#file1706165line154>
> >
> >     This also looks repetitive. TypeInfoUtils already has something like 
> > this I guess. We need to make sure TypeInfo parser can parse the string 
> > generated by this method. It's easier to reuse TypeInfoUtils or have a 
> > TypeDesc converted to TypeInfo.
> >     
> >     My point there is duplicacy in
> >     TypeInfo
> >     TypeDesc
> >     TypeDescriptor (ORC has this)
> >     
> >     Wondering if TypeInfo or TypeDescriptor from ORC can be reused here. 
> > Thoughts?
> 
> Jason Dere wrote:
>     I had orginally created TypeDesc to try to avoid extra dependencies on 
> what might be considered internal Hive types, but it is true that we are 
> still the serde lib and TypeInfo is available from there. At this point 
> though the TypeDesc has already been created, I'm not totally sure if I want 
> to now go and remove this. If you feel strongly about this let me know.
> 
> Prasanth_J wrote:
>     May be this can be done in a follow up if it involves a lot of surgery.
>     If removing hive dependency is desired or required, then there are 2 
> possibilities (Make TypeDescriptor in ORC a separate module or make TypeInfo 
> in hive as separate module). But at the current state it looks like there is 
> dependecy with hive serde and typeinfo, so it might be easier to reuse code.
>     This way there won't be problem with maintaining compatibility with hive 
> types when new one gets added.

I've tried to update the patch to remove TypeDesc and use TypeInfo instead.


- Jason


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/58934/#review173697
-----------------------------------------------------------


On May 2, 2017, 9:52 p.m., Jason Dere wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/58934/
> -----------------------------------------------------------
> 
> (Updated May 2, 2017, 9:52 p.m.)
> 
> 
> Review request for hive, Gunther Hagleitner, Prasanth_J, and Siddharth Seth.
> 
> 
> Bugs: HIVE-16568
>     https://issues.apache.org/jira/browse/HIVE-16568
> 
> 
> Repository: hive-git
> 
> 
> Description
> -------
> 
> - Support list/map/struct types in the LLAPRowInputFormat Schema/TypeDesc
> - Support list/map/struct types in the LLAPRowInputFormat Row. Changes in the 
> Row getters/setters needed (no longer using Writable).
> 
> 
> Diffs
> -----
> 
>   
> itests/hive-unit/src/test/java/org/apache/hive/jdbc/TestJdbcWithMiniLlap.java 
> de47412 
>   llap-client/src/java/org/apache/hadoop/hive/llap/LlapRowRecordReader.java 
> ee92f3e 
>   llap-common/src/java/org/apache/hadoop/hive/llap/Row.java a84fadc 
>   llap-common/src/java/org/apache/hadoop/hive/llap/TypeDesc.java dda5928 
>   ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDTFGetSplits.java 
> 9ddbd7e 
> 
> 
> Diff: https://reviews.apache.org/r/58934/diff/1/
> 
> 
> Testing
> -------
> 
> Added test to TestJdbcWithMiniLlap
> 
> 
> Thanks,
> 
> Jason Dere
> 
>

Reply via email to