ffc team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Wed, Feb 03, 2010 at 01:46:32PM +0100, Anders Logg wrote:
> On Wed, Feb 03, 2010 at 01:27:51PM +0100, Kristian Oelgaard wrote:
> >
> >
> > On 3 February 2010 13:21, Anders Logg <logg@xxxxxxxxx> wrote:
> > >The regression test is failing of HyperElasticity. I guess this is
> > >after the recent work on stripping variables?
> > >
> > >Is this demo ok so we can update the reference?
> >
> > I pushed a fix to ufl, and the tests are green apart from MixedMixedElement for which I get:
> >
> > 346_evaluate_dof_6: values differ
> >  old = 0.645561
> >  new = 0.645562
> > 355_evaluate_dofs: values differ
> >  old = -0.844483 -0.984214 -1.12569 0.04968 0.063168 0.077184 0.645561 0.71467 0.752005 0.055611 0.0594 0.068229 -0.824896 -0.8811 -1.01206
> >  new = -0.844483 -0.984214 -1.12569 0.04968 0.063168 0.077184 0.645562 0.71467 0.752005 0.055611 0.0594 0.068229 -0.824896 -0.8811 -1.01206
> >
> > a tolerance issue.
>
> ok, nice. I'll increase it to 1e-5. Soon we'll be at 0.1...

Fixed, finally, with a tolerance of 1e-9.

The problem was comparison of strings rather than float values with a
tolerance. I thought initially it would be enough to just compare the
strings and make sure those string values are printed out with only a
certain number of decimals, but we kept being screwed over by values
like

  0.0625000000000000001
  0.0624999999999999999

Which are rounded to different numbers...

--
Anders

Attachment: signature.asc
Description: Digital signature