Backport BigDecimal performance patches

[Joe Darcy]

Dr Andrew John Hughes wrote:
> On 12:34 Thu 04 Nov     , Joe Darcy wrote:
>   
>> Andrew Haley wrote:
>>     
>>> This is a backport of the BigDecimal performance rewrite.
>>> http://cr.openjdk.java.net/~aph/6622432-openjdk6-webrev/
>>>
>>> rev 397 : 6622432: RFE: Performance improvements to java.math.BigDecimal
>>> Reviewed-by: darcy
>>> rev 398 : 6850606: Regression from JDK 1.6.0_12
>>> Summary: The returned result from multiply should be constructed by using valueOf to take care of the INFLATED case.
>>> Reviewed-by: darcy
>>> rev 399 : 6876282: BigDecimal's divide(BigDecimal bd, RoundingFormat r) produces incorrect result
>>> Reviewed-by: darcy
>>>
>>> The patches didn't quite apply cleanly, but once I fixed them up by hand
>>> there were no regressions of either the JCK or JTreg.
>>>
>>> Given that 2000 lines of patch is rather hard to read, I have attached
>>> a diff between the patched jdk6 and the current jdk7 tree.
>>>
>>> OK for jdk6?
>>>
>>>   
>>>       
>> Hi Andrew.
>>
>> I approve this backport in principle, but need to look over the diffs 
>> more closely before giving the final okay.  There were some spec changes 
>> in BigInteger/BigDecimal in JDK 7; you seem to have correctly adjusted 
>> the patches for OpenJDK 6, but I just want to verify that.
>>
>> Thanks for including the diff against JDK 7; that is a big aid in 
>> reviewing :-)
>>
>> -Joe
>>     
>
> FWIW, the changes Andrew has backported:
>
> changeset:   1794:98558a60c555
> user:        darcy
> date:        Wed Oct 21 09:53:23 2009 -0700
> summary:     6560935: BigInteger.modPow() throws ArithmeticException for negative exponent
>
> changeset:   1785:0dd3d16e8183
> user:        darcy
> date:        Tue Oct 20 09:51:28 2009 -0700
> summary:     6371401: java.math.BigInteger.shift(Integer.MIN_VALUE) throws StackOverflowError
>
> changeset:   1246:8d2efec31d78
> user:        xlu
> date:        Sun May 24 16:29:57 2009 -0700
> summary:     6622432: RFE: Performance improvements to java.math.BigDecimal
>
> don't include the spec. changes, which do appear in the jdk7 diff he supplied.  They seem to
> have occurred prior to OpenJDK entering Mercurial with b24.  See:
>
> http://hg.openjdk.java.net/jdk7/jdk7/jdk/file/0/src/share/classes/java/math/BigInteger.java
>
> and were then reverted in OpenJDK6:
>
> changeset:   2:39e8fe7a0af1
> tag:         jdk6-b00
> user:        ohair
> date:        Fri Jan 30 16:05:57 2009 -0800
> summary:     6755277: All initial changes to jdk7 to create openjdk 6 build 0
>
> http://hg.openjdk.java.net/jdk6/jdk6/jdk/diff/39e8fe7a0af1/src/share/classes/java/math/BigInteger.java
>
>   
Thanks for the assistance with the bug archeology.  The pre-Hg JDK 7 bug 
for the change in the string constructor is

    5017980 Allow signed positive integer to be parsed instead of a 
NumerFormatException

fixed way back in JDK 7 build 3.  This bug stands out for me in the 
early history of OpenJDK 6.  In JDK 7, because of the delegations among 
the String -> number methods the specifications of 26 methods were 
modified by changing the code for about 5 methods.  The JCK team had 
dutifully written tests included in JCK 6 to verify "+" was not accepted 
as a leading character of those 26 methods so this change was the cause 
of many, many JCK failures against JCK 6 in the initial backwards branch 
of OpenJDK 6 from JDK 7 build 20 or so.

In any case, the patch is good and I approve the changes being backported.

Cheers,

-Joe