ffc team mailing list archive

[Kristian Oelgaard <k.b.oelgaard@xxxxxxxxxx>]

I have decreased the precision. Try to run the tests again and see if it still
complains.

Kristian

Quoting Ilmar Wilbers <ilmarw@xxxxxxxxx>:

> The error is only numerical:
> 
> 932c932
> <     {1.9272482233e+00, 1.8283482007e+00, 2.9908590898e+00,
> -1.2440333788e+00, 9.5014618758e-01, 5.7406445631e+00, -1.6996731712e-01,
> -7.2111025509e-01, -1.1796107330e+00, 4.9065338146e-01, 2.9439202888e-01,
> 4.9761335153e+00, 5.7817446700e+00, -3.2268221135e+00, -2.1547290184e-01,
> -1.3018558182e+00, 3.8544964466e-02, 1.6353243453e-01, 2.6751056944e-01,
> -1.1126972805e-01, 0.0000000000e+00, 2.6463875384e+00, 1.0763664872e+01,
> -3.5732138395e+00, 1.3753306481e+00, -2.9587512153e-02, -5.0012023566e-01,
> -5.8108720315e-01, 3.2430782472e-01, 2.1655841452e-02, 1.3084143273e-01,
> -3.8739146877e-03, -1.6435628072e-02, -2.6885823827e-02, 1.1183028439e-02,
> 0.0000000000e+00, 0.0000000000e+00, 0.0000000000e+00, 0.0000000000e+00,
> 0.0000000000e+00, 0.0000000000e+00, 0.0000000000e+00, 0.0000000000e+00,
> 0.0000000000e+00, 0.0000000000e+00, 0.0000000000e+00, 0.0000000000e+00,
> 0.0000000000e+00, 0.0000000000e+00, 0.0000000000e+00, 0.0000000000e+00,
> 0.0000000000e+00, 0.0000000000e+00, 0.0000000000e+00, 0.0000000000e+00,
> 0.0000000000e+00},
> ---
> >     {1.9272482233e+00, 1.8283482007e+00, 2.9908590898e+00,
> -1.2440333788e+00, 9.5014618758e-01, 5.7406445631e+00,
> -1.6996731712e-01, -7.2111025509e-01, -1.1796107330e+00,
> 4.9065338146e-01, 2.9439202888e-01, 4.9761335153e+00,
> 5.7817446700e+00, -3.2268221135e+00, -2.1547290184e-01,
> -1.3018558182e+00, 3.8544964466e-02, 1.6353243453e-01,
> 2.6751056944e-01, -1.1126972805e-01, 0.0000000000e+00,
> 2.6463875384e+00, 1.0763664872e+01, -3.5732138395e+00,
> 1.3753306481e+00, -2.9587512153e-02, -5.0012023566e-01,
> -5.8108720315e-01, 3.2430782472e-01, 2.1655841452e-02,
> 1.3084143273e-01, -3.8739146876e-03, -1.6435628072e-02,
> -2.6885823827e-02, 1.1183028439e-02, 0.0000000000e+00,
> 0.0000000000e+00, 0.0000000000e+00, 0.0000000000e+00,
> 0.0000000000e+00, 0.0000000000e+00, 0.0000000000e+00,
> 0.0000000000e+00, 0.0000000000e+00, 0.0000000000e+00,
> 0.0000000000e+00, 0.0000000000e+00, 0.0000000000e+00,
> 0.0000000000e+00, 0.0000000000e+00, 0.0000000000e+00,
> 0.0000000000e+00, 0.0000000000e+00, 0.0000000000e+00,
> 0.0000000000e+00, 0.0000000000e+00},
> 
> but it still breaks the code (the buildbot is not happy). Any way to avoid
> numerical precision errors? Like including a function in regression.py for
> setting the output with a limited number of decimals, or taking the
> absolute value of the numbers from the verification file and the output,
> and ignoring errors smaller then a certain epsilon?
> 
> Also, would be nice to have an option for viewdiff that does not require
> X, that is, diff or the like instead of meld. Maybe a command line
> argument?
> 
> ilmar
> 
> > The buildbot address is http://fenics.org:8010/ffc/, by the way. You can
> look at the output of the different steps by clicking on the 'stdio'
> link.
> >
> > ilmar
> >
> > Ilmar Wilbers skrev:
> >> I am getting regression errors on the Buildbot for both 32-bit Linux
> and
> >> Mac, 64-bit Linus seems fine, for some reason:
> >> *** Regression test failed for 1 of 23 forms:
> >>   P5tet.form
> >> Will try to include the output from meld, but it will have to wait
> 'till
> >> I get to work, just thought I'd let you know in advance.
> >> ilmar
> >> _______________________________________________
> >> FFC-dev mailing list
> >> FFC-dev@xxxxxxxxxx
> >> http://www.fenics.org/mailman/listinfo/ffc-dev
> > _______________________________________________
> > FFC-dev mailing list
> > FFC-dev@xxxxxxxxxx
> > http://www.fenics.org/mailman/listinfo/ffc-dev
> >
> 
> 
> 
> 
> _______________________________________________
> FFC-dev mailing list
> FFC-dev@xxxxxxxxxx
> http://www.fenics.org/mailman/listinfo/ffc-dev
>