Failed test for testExpln

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
9 messages Options
Reply | Threaded
Open this post in threaded view
|

Failed test for testExpln

SergeStinckwich
Dear all,

I modify a little bit the Jenkis config on the CI server in order to
run the ArbitraryPrecisionFloat tests. Apparently there is a bug in
all the builds for the testExpLn tests.

Look at the CI job here:
https://ci.inria.fr/pharo-contribution/job/SciSmalltalk/

This test is green on my own computer (under OS X) ...
The CI slaves are Linux machines. Something related to Linux VM ?

Regards,
--
Serge Stinckwich
UCBN & UMI UMMISCO 209 (IRD/UPMC)
Every DSL ends up being Smalltalk
http://www.doesnotunderstand.org/

--
You received this message because you are subscribed to the Google Groups "SciSmalltalk" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [hidden email].
For more options, visit https://groups.google.com/d/optout.
Uko
Reply | Threaded
Open this post in threaded view
|

Re: Failed test for testExpln

Uko
Hi, have you tried to load the error into your image and debugging it?

Uko

On 21 May 2015, at 14:15, Serge Stinckwich <[hidden email]> wrote:

Dear all,

I modify a little bit the Jenkis config on the CI server in order to
run the ArbitraryPrecisionFloat tests. Apparently there is a bug in
all the builds for the testExpLn tests.

Look at the CI job here:
https://ci.inria.fr/pharo-contribution/job/SciSmalltalk/

This test is green on my own computer (under OS X) ...
The CI slaves are Linux machines. Something related to Linux VM ?

Regards,
--
Serge Stinckwich
UCBN & UMI UMMISCO 209 (IRD/UPMC)
Every DSL ends up being Smalltalk
http://www.doesnotunderstand.org/

--
You received this message because you are subscribed to the Google Groups "SciSmalltalk" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [hidden email].
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "SciSmalltalk" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [hidden email].
For more options, visit https://groups.google.com/d/optout.
Reply | Threaded
Open this post in threaded view
|

Re: Failed test for testExpln

SergeStinckwich
Yes I open the error but I have only a postmortem interface and the
test is true.



On Thu, May 21, 2015 at 2:23 PM, Yuriy Tymchuk <[hidden email]> wrote:

> Hi, have you tried to load the error into your image and debugging it?
> ArbitraryPrecisionFloatTest-testExpLn.fuel
>
> Uko
>
> On 21 May 2015, at 14:15, Serge Stinckwich <[hidden email]>
> wrote:
>
> Dear all,
>
> I modify a little bit the Jenkis config on the CI server in order to
> run the ArbitraryPrecisionFloat tests. Apparently there is a bug in
> all the builds for the testExpLn tests.
>
> Look at the CI job here:
> https://ci.inria.fr/pharo-contribution/job/SciSmalltalk/
>
> This test is green on my own computer (under OS X) ...
> The CI slaves are Linux machines. Something related to Linux VM ?
>
> Regards,
> --
> Serge Stinckwich
> UCBN & UMI UMMISCO 209 (IRD/UPMC)
> Every DSL ends up being Smalltalk
> http://www.doesnotunderstand.org/
>
> --
> You received this message because you are subscribed to the Google Groups
> "SciSmalltalk" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to [hidden email].
> For more options, visit https://groups.google.com/d/optout.
>
>
> --
> You received this message because you are subscribed to the Google Groups
> "SciSmalltalk" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to [hidden email].
> For more options, visit https://groups.google.com/d/optout.



--
Serge Stinckwich
UCBN & UMI UMMISCO 209 (IRD/UPMC)
Every DSL ends up being Smalltalk
http://www.doesnotunderstand.org/

--
You received this message because you are subscribed to the Google Groups "SciSmalltalk" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [hidden email].
For more options, visit https://groups.google.com/d/optout.
Reply | Threaded
Open this post in threaded view
|

Re: Failed test for testExpln

werner kassens-2
Hi Serge,
this test fails, because it assumes that pharo calcs #exp correctly. but in this case pharo rounds wrongly. i'd guess that Nicolas simply wants - in the best scientific tradition - to document this, since this _is a problem. in my humble opinion it is not really unimportant to document this pharo behaviour, because these things _do influence the decision of scientists to use a certain language or another, or btw of corporations or organisations, who use scientific software. mmmhm, i dont know whether i may say that - well ok i usually say what i think, hence bluntly: since you belong to the inner circle, it could eventually not be the worst (?) idea if you would start a discussion with the pharo-chiefs and the vm-chiefs (its a vm thing), how to solve this pharo problem with the help of Nicolas.
werner

--
You received this message because you are subscribed to the Google Groups "SciSmalltalk" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [hidden email].
For more options, visit https://groups.google.com/d/optout.
Reply | Threaded
Open this post in threaded view
|

Re: Failed test for testExpln

werner kassens-2
and it does not fail on OS X since the vm uses different ways to calc this on different operating systems.
werner

--
You received this message because you are subscribed to the Google Groups "SciSmalltalk" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [hidden email].
For more options, visit https://groups.google.com/d/optout.
Reply | Threaded
Open this post in threaded view
|

Re: Failed test for testExpln

Nicolas Cellier


2015-05-22 14:17 GMT+02:00 werner kassens <[hidden email]>:
and it does not fail on OS X since the vm uses different ways to calc this on different operating systems.
werner


Well, this is a problem that should be fixed: functions should better evaluate identically on different machines.
There used to be a FloatMathPlugin based on fdlibm for that purpose...
But it's not really the problem of ArbitraryPrecisionFloat.

It's possible that I wanted to use Float as a quick and easy reference when I started implementation...
In this case, this is historical, and we should probably rewrite these tests...
But as Werner said, it certainly was also my purpose to document behavior of Float math functions, so I don't know if it still has some value...

Nicolas

--
You received this message because you are subscribed to the Google Groups "SciSmalltalk" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [hidden email].
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "SciSmalltalk" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [hidden email].
For more options, visit https://groups.google.com/d/optout.
Reply | Threaded
Open this post in threaded view
|

Re: Failed test for testExpln

werner kassens-2
Hi,
in the end ArbitraryPrecisionFloatTest is a test of ArbitraryPrecisionFloat and perhaps it would make sense to hardcode the correct result in the test, this way the test would be as strict as before and would not fail?

and regarding the precision of calculations generally, perhaps two other examples are instructive:
the correct result of  "1.5707963267948968 tan" would be, i dont know, something like "-5e15" but
1.5707963267948968 tan. "--> 8.165889364191922e15"
iow it differs by more than 200% from the correct result. it seems to me it would be nice if at least the sign would be correct.
or the correct result of "-1.5707963267948966 tan" would perhaps be something around -6e16 but pharo returns an error.
of course this happens because pharo calcs these results by itself (#sin is the only trigonometric function that uses a primitive), but perhaps there exist floating math libraries that can do that instead?
and that pharo returns different results on different computers is imo also not ideal for a project like scismalltalk.
werner


--
You received this message because you are subscribed to the Google Groups "SciSmalltalk" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [hidden email].
For more options, visit https://groups.google.com/d/optout.
Reply | Threaded
Open this post in threaded view
|

Re: Failed test for testExpln

SergeStinckwich
On Wed, Jun 3, 2015 at 10:14 AM, werner kassens <[hidden email]> wrote:

> Hi,
> in the end ArbitraryPrecisionFloatTest is a test of ArbitraryPrecisionFloat
> and perhaps it would make sense to hardcode the correct result in the test,
> this way the test would be as strict as before and would not fail?
>
> and regarding the precision of calculations generally, perhaps two other
> examples are instructive:
> the correct result of  "1.5707963267948968 tan" would be, i dont know,
> something like "-5e15" but
> 1.5707963267948968 tan. "--> 8.165889364191922e15"
> iow it differs by more than 200% from the correct result. it seems to me it
> would be nice if at least the sign would be correct.
> or the correct result of "-1.5707963267948966 tan" would perhaps be
> something around -6e16 but pharo returns an error.
> of course this happens because pharo calcs these results by itself (#sin is
> the only trigonometric function that uses a primitive), but perhaps there
> exist floating math libraries that can do that instead?
> and that pharo returns different results on different computers is imo also
> not ideal for a project like scismalltalk.

Yes, I agree.

Can we report this on pharo-dev mailing-list ?
What simple example reply a different result regarding the platform ?

About the FloatMathPlugin: apparently there is a FloatMathPlugin
included with Pharo VM.
This plugin is not based on FDLIBM ?

I will modify the test in order to say that this test will fail.
Thank you.
--
Serge Stinckwich
UCBN & UMI UMMISCO 209 (IRD/UPMC)
Every DSL ends up being Smalltalk
http://www.doesnotunderstand.org/

--
You received this message because you are subscribed to the Google Groups "SciSmalltalk" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [hidden email].
For more options, visit https://groups.google.com/d/optout.
Reply | Threaded
Open this post in threaded view
|

Re: Failed test for testExpln

werner kassens-2

Hi Serge,

On Sunday, July 19, 2015 at 8:09:48 PM UTC+2, Serge Stinckwich wrote:
Can we report this on pharo-dev mailing-list ?

well, i cant do that as i am only member of the pharo-users mailing list and i dont want to change that, i'm happy being a simple user. i know that this can look like a rather stupid behaviour, please excuse that, but thats just the way it is.
 
What simple example reply a different result regarding the platform ?

#ln calculations in macs contra other os's. Nicolas brought that problem up eg in fogbugs case 15453.

werner

--
You received this message because you are subscribed to the Google Groups "SciSmalltalk" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [hidden email].
For more options, visit https://groups.google.com/d/optout.