ffc team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Thu, Sep 24, 2009 at 08:08:30AM +0200, Johannes Ring wrote:
> On Wed, Sep 23, 2009 at 8:09 PM, Anders Logg <logg@xxxxxxxxx> wrote:
> > On Wed, Sep 23, 2009 at 03:42:11PM +0200, Johannes Ring wrote:
> >> On Wed, Sep 23, 2009 at 11:17 AM, Anders Logg <logg@xxxxxxxxx> wrote:
> >> > On Wed, Sep 23, 2009 at 10:12:43AM +0100, Garth N. Wells wrote:
> >> >>
> >> >>
> >> >> Anders Logg wrote:
> >> >> > On Wed, Sep 23, 2009 at 10:06:00AM +0100, Garth N. Wells wrote:
> >> >> >>
> >> >> >> Anders Logg wrote:
> >> >> >>> On Wed, Sep 23, 2009 at 09:40:05AM +0200, Johannes Ring wrote:
> >> >> >>>> On Tue, Sep 22, 2009 at 2:46 PM, Anders Logg <logg@xxxxxxxxx> wrote:
> >> >> >>>>> Why is the buildbot failing for FFC on winxp-mingw32? Is this
> >> >> >>>>> something that could be fixed?
> >> >> >>>> The problem is that floats are represented slightly differently
> >> >> >>>> because of the underlying C library. On Windows, float(1e-8) is 1e-008
> >> >> >>>> while it is 1e-08 on Ubuntu.
> >> >> >>> I don't understand why this should matter. Where does this problem show up?
> >> >> >>>
> >> >> >> When performing a diff on the generated and reference files.
> >> >> >
> >> >> > You mean the regression test fail just because floats are formatted
> >> >> > differently on the two platforms?
> >> >> >
> >> >>
> >> >> Yes.
> >> >
> >> > Johannes, could you send an example of a test that fails so I can look
> >> > at it in detail? I won't be able to test myself on Windows, but
> >> > perhaps you can find some solution by modifying the following lines in
> >> >
> >> >  ffc/compiler/ufcformat.py:
> >> >
> >> >  # Set number of digits for floating point and machine precision
> >> >  precision = int(options["precision"])
> >> >  f1 = "%%.%dg" % precision
> >> >  f2 = "%%.%de" % precision
> >> >  def floating_point(v):
> >> >      if abs(v) < 100.0:
> >> >          return f1 % v
> >> >      else:
> >> >          return f2 % v
> >> >  self.format["floating point"] = floating_point
> >>
> >> I guess it should work if "%%.%df" is used instead but that is
> >> probably not a good solution. I can't think of anything else that will
> >> make this work. At least FFC is working on Windows as the errors can
> >> be ignored.
> >>
> >> Anders: If you like, I can give you access to the Windows buildslave
> >> so that you can try it out yourself.
> >
> > Yes, that would be nice. How do I log in?
> 
> I will tell you off list.

Thanks, I'll take a look later today.

-- 
Anders

Attachment: signature.asc
Description: Digital signature