[
https://issues.apache.org/jira/browse/TAJO-391?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13842162#comment-13842162
]
Jihoon Son edited comment on TAJO-391 at 12/7/13 10:53 AM:
-----------------------------------------------------------
Basically, I agree with changing the default parsed type of real values from
float4 to float8, because float8 covers a wider range of values. Also, the
attached patch looks good.
But, this issue seems to propose that every real value should be float8 for the
more precise math function results.
Hyunsik said that the precision lose problem is occurred only when the constant
values are used as function parameters, but, in my opinion, it is hard to guess
the reason.
Currently, the results of math functions are tested by comparing with the exact
values of them.
If the precision problem of the real numbers bothers the testing math
functions, it is better to use assertEquals(double expected, double actual,
double delta) instead of assertEquals(double expected, double actual).
was (Author: jihoonson):
Basically, I agree with changing the default type of real values from float4 to
float8, because float8 covers a wider range of values. Also, the attached patch
looks good.
But, this issue seems to propose that every real value should be float8 for the
more precise math function results.
Hyunsik said that the precision lose problem is occurred only when the constant
values are used as function parameters, but, in my opinion, it is hard to guess
the reason.
Currently, the results of math functions are tested by comparing with the exact
values of them.
If the precision problem of the real numbers bothers the testing math
functions, it is better to use assertEquals(double expected, double actual,
double delta) instead of assertEquals(double expected, double actual).
> Change setting default type in query from FLOAT4 to FLOAT8
> ----------------------------------------------------------
>
> Key: TAJO-391
> URL: https://issues.apache.org/jira/browse/TAJO-391
> Project: Tajo
> Issue Type: Improvement
> Components: operator/function/udf
> Reporter: DaeMyung Kang
> Assignee: DaeMyung Kang
> Priority: Minor
> Fix For: 0.8-incubating
>
> Attachments: TAJO-391.patch
>
>
> FLOAT4 is default parsed float type in tajo.
> but it can cause some trobles when we use it as FLOAT8.
> for examples.
> 0.4(f) will changed to 0.40000000000005(d)
> so it cause different result from java.Math functions.
--
This message was sent by Atlassian JIRA
(v6.1#6144)
