Le 24/02/2011 18:17, sebb a écrit :
> No, the patch does not work.
>
> However, the following code does seem to work for me on both 1.5 and 1.6
>
> if (ra == 0) { // Fix up the sign so atan works correctly
> ra = copySign(0.0, y);
> }
>
> i.e. always copy the sign of y, not just when negative.
I don't see the difference with my patch.
Luc
>
>> Please, could one of those who see the bug check if the following patch
>> fixes it or not ?
>>
>> Thanks
>> Luc
>>
>>
>> ===================================================================
>> --- src/main/java/org/apache/commons/math/util/FastMath.java (révision
>> 1073250)
>> +++ src/main/java/org/apache/commons/math/util/FastMath.java (copie de
>> travail)
>> @@ -3046,8 +3046,8 @@
>> rb = -(temp - ra - rb);
>> ra = temp;
>>
>> - if (ra == 0 && (y < 0)) { // Fix up the sign so atan works correctly
>> - ra = -0.0;
>> + if (ra == 0) { // Fix up the sign so atan works correctly
>> + ra = copySign(0.0, y);
>> }
>>
>> // Call atan
>>
>>
>>
>>
>>
>> ----- "luc maisonobe" <luc.maison...@free.fr> a écrit :
>>
>>> ----- "Oliver Heger" <oliver.he...@oliver-heger.de> a écrit :
>>>
>>>> Am 23.02.2011 21:08, schrieb Luc Maisonobe:
>>>>> Le 23/02/2011 20:35, Simone Tripodi a écrit :
>>>>>> I wonder if the cause of the fails on my runs could be an
>>> Apple's
>>>> JVM bug...
>>>>>> Any idea?
>>>>>
>>>>> I think Sebastian encountered this bug with Windows XP.
>>>>>
>>>>> Luc
>>>>
>>>> I see the same problem on Windows 7 and JDK 1.5:
>>>>
>>>> Results :
>>>>
>>>> Failed tests:
>>>>
>>>> test1[3](org.apache.commons.math.util.FastMathStrictComparisonTest)
>>>>
>>>> Tests run: 2451, Failures: 1, Errors: 0, Skipped: 1
>>>>
>>>> Here is my setup:
>>>> Apache Maven 2.2.1 (r801777; 2009-08-06 21:16:01+0200)
>>>> Java version: 1.5.0_21
>>>> Java home: C:\Program Files\Java\jdk1.5.0_21\jre
>>>> Default locale: de_DE, platform encoding: Cp1252
>>>> OS name: "windows 7" version: "6.1" arch: "x86" Family: "windows"
>>>>
>>>> With JDK 1.6 I can build successfully. So it seems to be a JDK
>>>> issue(?).
>>>
>>> I think so.
>>> One should also be aware that the error is at a discontinuity.
>>> Basically, this test computes atan2(y, x) with y being almost 0
>>> (slightly positive) and x being a very large negative number. If y was
>>> exactly 0 and x was exactly NEGATIVE_INFINITY, then the correct result
>>> would be either +PI or -PI. Here, due to the signs of x and y, the
>>> result should be +PI and it is -PI which is the same modulo 2PI.
>>>
>>> So it is not a completely absurd result.
>>>
>>> Luc
>>>
>>>>
>>>> Oliver
>>>>
>>>>>
>>>>>> Simo
>>>>>>
>>>>>> http://people.apache.org/~simonetripodi/
>>>>>> http://www.99soft.org/
>>>>>>
>>>>>>
>>>>>>
>>>>>> On Wed, Feb 23, 2011 at 7:03 PM, Jörg
>>>> Schaible<joerg.schai...@gmx.de> wrote:
>>>>>>> Hi Luc,
>>>>>>>
>>>>>>> Luc Maisonobe wrote:
>>>>>>>
>>>>>>>> Tag:
>>>>>>>>
>>>>
>>> http://svn.apache.org/repos/asf/commons/proper/math/tags/MATH_2_2_RC3/
>>>>>>>>
>>>>>>>> Distributions: http://people.apache.org/~luc/math-2.2-RC3/
>>>>>>>>
>>>>>>>> Maven artifacts:
>>>> http://people.apache.org/~luc/math-2.2-RC3/maven/
>>>>>>>>
>>>>>>>> Documentation bundled with the binary distribution:
>>>>>>>> http://people.apache.org/~luc/math-2.2-RC3/docs/
>>>>>>>>
>>>>>>>> Output of maven:site run against the source distribution:
>>>>>>>> http://people.apache.org/~luc/math-2.2-RC3/site/
>>>>>>>>
>>>>>>>> Clirr report:
>>>>>>>>
>>>> http://people.apache.org/~luc/math-2.2-RC3/site/clirr-report.html
>>>>>>>>
>>>>>>>> Release notes:
>>>>>>>> http://people.apache.org/~luc/math-2.2-RC3/RELEASE-NOTES.txt
>>>>>>>>
>>>>>>>> Votes, please. This vote will close in 72 hours,
>>>> 2011-02-26T12:00:00 UTC
>>>>>>>>
>>>>>>>> [ ] +1 Release these artifacts
>>>>>>>> [ ] +0 OK, but...
>>>>>>>> [ ] -0 OK, but really should fix...
>>>>>>>> [ ] -1 I oppose this release because...
>>>>>>>
>>>>>>> +1
>>>>>>>
>>>>>>> I've tested the source distribution and can build and run all
>>>> tests
>>>>>>> flawlessly with M3 using Sun JDK 1.5 + 1.6, IMB JDK 1.5 + 1.6
>>> and
>>>> OpenJDK
>>>>>>> 1.6. However, I am running also on Linux/AMD64 architecture.
>>>>>>>
>>>>>>> Cheers,
>>>>>>> Jörg
>>>>>>>
>>>>>>>
>>>>>>>
>>>>
>>> ---------------------------------------------------------------------
>>>>>>> To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
>>>>>>> For additional commands, e-mail: dev-h...@commons.apache.org
>>>>>>>
>>>>>>>
>>>>>>
>>>>>>
>>>>
>>> ---------------------------------------------------------------------
>>>>>> To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
>>>>>> For additional commands, e-mail: dev-h...@commons.apache.org
>>>>>>
>>>>>
>>>>>
>>>>>
>>>>
>>> ---------------------------------------------------------------------
>>>>> To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
>>>>> For additional commands, e-mail: dev-h...@commons.apache.org
>>>>>
>>>>
>>>>
>>>>
>>> ---------------------------------------------------------------------
>>>> To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
>>>> For additional commands, e-mail: dev-h...@commons.apache.org
>>>
>>> ---------------------------------------------------------------------
>>> To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
>>> For additional commands, e-mail: dev-h...@commons.apache.org
>>
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
>> For additional commands, e-mail: dev-h...@commons.apache.org
>>
>>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
> For additional commands, e-mail: dev-h...@commons.apache.org
>
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
For additional commands, e-mail: dev-h...@commons.apache.org
