You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@hive.apache.org by Xuefu Zhang <xz...@cloudera.com> on 2013/11/22 22:42:46 UTC

Review Request 15804: HIVE-5866: Hive divide operator generates wrong results in certain cases

-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/15804/
-----------------------------------------------------------

Review request for hive.


Bugs: HIVE-5866
    https://issues.apache.org/jira/browse/HIVE-5866


Repository: hive-git


Description
-------

Fixed the problem. Added a unit test. Corrected the output of a few q tests.


Diffs
-----

  ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFBaseNumeric.java a1015e9 
  ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFOPDivide.java 0b902e9 
  ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFOPMinus.java 538c07e 
  ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFOPMod.java 472e1dd 
  ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFOPMultiply.java 2e8d364 
  ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFOPPlus.java 35f639e 
  ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFPosMod.java 6b18303 
  ql/src/test/org/apache/hadoop/hive/ql/udf/generic/TestGenericUDFOPDivide.java 581c1a8 
  ql/src/test/results/clientpositive/decimal_precision.q.out 2ee3578 
  ql/src/test/results/clientpositive/decimal_udf.q.out ed5bc65 
  ql/src/test/results/clientpositive/ql_rewrite_gbtoidx.q.out 83787ee 
  ql/src/test/results/clientpositive/vectorization_5.q.out 54aad90 
  ql/src/test/results/clientpositive/vectorization_short_regress.q.out c9296e1 

Diff: https://reviews.apache.org/r/15804/diff/


Testing
-------


Thanks,

Xuefu Zhang


Re: Review Request 15804: HIVE-5866: Hive divide operator generates wrong results in certain cases

Posted by Xuefu Zhang <xz...@cloudera.com>.
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/15804/#review29332
-----------------------------------------------------------



ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFOPDivide.java
<https://reviews.apache.org/r/15804/#comment56518>

    I don't think I get you. Could you give an example about how implementation here is different from sql server's? The decimal part of a decimal number is as important as the integer part in applications where decimal type is required. Otherwise, double might be better. Thus, a decimal number of a certain decimal type needs to comply with the type's precision/scale. I don't think we should store number 456.78 to a type decimal(6,4), about which we already concluded the discussion.


- Xuefu Zhang


On Nov. 22, 2013, 9:42 p.m., Xuefu Zhang wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/15804/
> -----------------------------------------------------------
> 
> (Updated Nov. 22, 2013, 9:42 p.m.)
> 
> 
> Review request for hive.
> 
> 
> Bugs: HIVE-5866
>     https://issues.apache.org/jira/browse/HIVE-5866
> 
> 
> Repository: hive-git
> 
> 
> Description
> -------
> 
> Fixed the problem. Added a unit test. Corrected the output of a few q tests.
> 
> 
> Diffs
> -----
> 
>   ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFBaseNumeric.java a1015e9 
>   ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFOPDivide.java 0b902e9 
>   ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFOPMinus.java 538c07e 
>   ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFOPMod.java 472e1dd 
>   ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFOPMultiply.java 2e8d364 
>   ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFOPPlus.java 35f639e 
>   ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFPosMod.java 6b18303 
>   ql/src/test/org/apache/hadoop/hive/ql/udf/generic/TestGenericUDFOPDivide.java 581c1a8 
>   ql/src/test/results/clientpositive/decimal_precision.q.out 2ee3578 
>   ql/src/test/results/clientpositive/decimal_udf.q.out ed5bc65 
>   ql/src/test/results/clientpositive/ql_rewrite_gbtoidx.q.out 83787ee 
>   ql/src/test/results/clientpositive/vectorization_5.q.out 54aad90 
>   ql/src/test/results/clientpositive/vectorization_short_regress.q.out c9296e1 
> 
> Diff: https://reviews.apache.org/r/15804/diff/
> 
> 
> Testing
> -------
> 
> 
> Thanks,
> 
> Xuefu Zhang
> 
>


Re: Review Request 15804: HIVE-5866: Hive divide operator generates wrong results in certain cases

Posted by Prasad Mujumdar <pr...@cloudera.com>.
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/15804/#review29643
-----------------------------------------------------------

Ship it!


LGTM.

I guess this will cause the some of the decimal results to change over 0.12, but this is the right way to compute it.

- Prasad Mujumdar


On Nov. 24, 2013, 4:13 p.m., Xuefu Zhang wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/15804/
> -----------------------------------------------------------
> 
> (Updated Nov. 24, 2013, 4:13 p.m.)
> 
> 
> Review request for hive.
> 
> 
> Bugs: HIVE-5866
>     https://issues.apache.org/jira/browse/HIVE-5866
> 
> 
> Repository: hive-git
> 
> 
> Description
> -------
> 
> Fixed the problem. Added a unit test. Corrected the output of a few q tests.
> 
> 
> Diffs
> -----
> 
>   ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFBaseNumeric.java a1015e9 
>   ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFOPDivide.java 0b902e9 
>   ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFOPMinus.java 538c07e 
>   ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFOPMod.java 472e1dd 
>   ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFOPMultiply.java 2e8d364 
>   ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFOPPlus.java 35f639e 
>   ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFPosMod.java 6b18303 
>   ql/src/test/org/apache/hadoop/hive/ql/udf/generic/TestGenericUDFOPDivide.java 581c1a8 
>   ql/src/test/queries/clientpositive/decimal_precision.q d46d3ce 
>   ql/src/test/queries/clientpositive/decimal_udf.q d8fd835 
>   ql/src/test/results/clientpositive/decimal_precision.q.out 2ee3578 
>   ql/src/test/results/clientpositive/decimal_udf.q.out ed5bc65 
>   ql/src/test/results/clientpositive/ql_rewrite_gbtoidx.q.out 83787ee 
>   ql/src/test/results/clientpositive/vectorization_5.q.out 54aad90 
>   ql/src/test/results/clientpositive/vectorization_short_regress.q.out c9296e1 
> 
> Diff: https://reviews.apache.org/r/15804/diff/
> 
> 
> Testing
> -------
> 
> 
> Thanks,
> 
> Xuefu Zhang
> 
>


Re: Review Request 15804: HIVE-5866: Hive divide operator generates wrong results in certain cases

Posted by Xuefu Zhang <xz...@cloudera.com>.
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/15804/
-----------------------------------------------------------

(Updated Nov. 24, 2013, 4:13 p.m.)


Review request for hive.


Bugs: HIVE-5866
    https://issues.apache.org/jira/browse/HIVE-5866


Repository: hive-git


Description
-------

Fixed the problem. Added a unit test. Corrected the output of a few q tests.


Diffs (updated)
-----

  ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFBaseNumeric.java a1015e9 
  ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFOPDivide.java 0b902e9 
  ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFOPMinus.java 538c07e 
  ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFOPMod.java 472e1dd 
  ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFOPMultiply.java 2e8d364 
  ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFOPPlus.java 35f639e 
  ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFPosMod.java 6b18303 
  ql/src/test/org/apache/hadoop/hive/ql/udf/generic/TestGenericUDFOPDivide.java 581c1a8 
  ql/src/test/queries/clientpositive/decimal_precision.q d46d3ce 
  ql/src/test/queries/clientpositive/decimal_udf.q d8fd835 
  ql/src/test/results/clientpositive/decimal_precision.q.out 2ee3578 
  ql/src/test/results/clientpositive/decimal_udf.q.out ed5bc65 
  ql/src/test/results/clientpositive/ql_rewrite_gbtoidx.q.out 83787ee 
  ql/src/test/results/clientpositive/vectorization_5.q.out 54aad90 
  ql/src/test/results/clientpositive/vectorization_short_regress.q.out c9296e1 

Diff: https://reviews.apache.org/r/15804/diff/


Testing
-------


Thanks,

Xuefu Zhang


Re: Review Request 15804: HIVE-5866: Hive divide operator generates wrong results in certain cases

Posted by Xuefu Zhang <xz...@cloudera.com>.
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/15804/
-----------------------------------------------------------

(Updated Nov. 24, 2013, 7:27 a.m.)


Review request for hive.


Bugs: HIVE-5866
    https://issues.apache.org/jira/browse/HIVE-5866


Repository: hive-git


Description
-------

Fixed the problem. Added a unit test. Corrected the output of a few q tests.


Diffs (updated)
-----

  ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFBaseNumeric.java a1015e9 
  ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFOPDivide.java 0b902e9 
  ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFOPMinus.java 538c07e 
  ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFOPMod.java 472e1dd 
  ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFOPMultiply.java 2e8d364 
  ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFOPPlus.java 35f639e 
  ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFPosMod.java 6b18303 
  ql/src/test/org/apache/hadoop/hive/ql/udf/generic/TestGenericUDFOPDivide.java 581c1a8 
  ql/src/test/queries/clientpositive/decimal_precision.q d46d3ce 
  ql/src/test/queries/clientpositive/decimal_udf.q d8fd835 
  ql/src/test/results/clientpositive/decimal_precision.q.out 2ee3578 
  ql/src/test/results/clientpositive/decimal_udf.q.out ed5bc65 
  ql/src/test/results/clientpositive/ql_rewrite_gbtoidx.q.out 83787ee 
  ql/src/test/results/clientpositive/vectorization_5.q.out 54aad90 
  ql/src/test/results/clientpositive/vectorization_short_regress.q.out c9296e1 

Diff: https://reviews.apache.org/r/15804/diff/


Testing
-------


Thanks,

Xuefu Zhang


Re: Review Request 15804: HIVE-5866: Hive divide operator generates wrong results in certain cases

Posted by Jason Dere <jd...@hortonworks.com>.
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/15804/#review29330
-----------------------------------------------------------



ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFOPDivide.java
<https://reviews.apache.org/r/15804/#comment56517>

    So I think the issue here is that when  (integer_prec + scale) > max_precision, we prioritize keeping the scale at the expense of the integer portion of the result type. Looks like the SQL Server precision/scale rules mention that it does not let the scale eat into the integer portion of the result type - it goes the other way and will reduce the scale to allow the total precision to fit within max_precision. This might be a better rule to follow than prioritizing the scale value, at least for the purposes of determining the return type.


- Jason Dere


On Nov. 22, 2013, 9:42 p.m., Xuefu Zhang wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/15804/
> -----------------------------------------------------------
> 
> (Updated Nov. 22, 2013, 9:42 p.m.)
> 
> 
> Review request for hive.
> 
> 
> Bugs: HIVE-5866
>     https://issues.apache.org/jira/browse/HIVE-5866
> 
> 
> Repository: hive-git
> 
> 
> Description
> -------
> 
> Fixed the problem. Added a unit test. Corrected the output of a few q tests.
> 
> 
> Diffs
> -----
> 
>   ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFBaseNumeric.java a1015e9 
>   ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFOPDivide.java 0b902e9 
>   ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFOPMinus.java 538c07e 
>   ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFOPMod.java 472e1dd 
>   ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFOPMultiply.java 2e8d364 
>   ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFOPPlus.java 35f639e 
>   ql/src/java/org/apache/hadoop/hive/ql/udf/generic/GenericUDFPosMod.java 6b18303 
>   ql/src/test/org/apache/hadoop/hive/ql/udf/generic/TestGenericUDFOPDivide.java 581c1a8 
>   ql/src/test/results/clientpositive/decimal_precision.q.out 2ee3578 
>   ql/src/test/results/clientpositive/decimal_udf.q.out ed5bc65 
>   ql/src/test/results/clientpositive/ql_rewrite_gbtoidx.q.out 83787ee 
>   ql/src/test/results/clientpositive/vectorization_5.q.out 54aad90 
>   ql/src/test/results/clientpositive/vectorization_short_regress.q.out c9296e1 
> 
> Diff: https://reviews.apache.org/r/15804/diff/
> 
> 
> Testing
> -------
> 
> 
> Thanks,
> 
> Xuefu Zhang
> 
>