Re: [Octave Forge] Octave 4.0 call for packages

[Oliver Heimlich]

On 11.04.2015 14:42, Oliver Heimlich wrote:
> On 11.04.2015 14:38, Julien Bect wrote:
> > Le 11/04/2015 08:59, Julien Bect a écrit :
> > > Le 11/04/2015 01:49, Oliver Heimlich a écrit :
> > > > > stk 2.2.0    - OK
> > > > PASS 576, FAIL 348 (many failed because of undefined variables)
> > > > 82 (of 222) .m files have no tests.
> > >
> > > Ouch, that is a very high number of failing tests :((
> > >
> > > I have never seen any "undefined variables" failure before and the
> > > tests results sent by JohnD for 4.0.0rc2 32 bits show no failure for
> > > STK.
> > >
> > > So, this must be a Win7/64bits-specific problem, then...
> > >
> > > I don't have access to a Win7/64bits box at home, but I will be able
> > > to test next week (hopefully on 4.0.0rc3, if the installer has been
> > > released by then).
> >
> > Thanks Oliver for uploading your logs.
> >
> > Here is a typical example of the test failures you get with Octave
> > 4.0rc2 under Win7/64bits:
> >
> > ***** shared x t u
> >   t = rand(3, 2);
> >   x = stk_dataframe(t);
> > !!!!! shared variable initialization failed
> >
> > ***** test u = apply(x, 1, @sum);
> > !!!!! test failed
> > 'x' undefined near line 2 column 12
> >
> > ***** assert (isequal(u, sum(t, 1)))
> > !!!!! test failed
> > 'u' undefined near line 2 column 17
> >
> > ...
> >
> > Could you please try to see if there's anything wrong with the code in
> > my %!shared block ?
> >
> > pkg load stk
> > t = rand (3, 2);
> > x = stk_dataframe (t);
>
> I can debug into this later. Currently I rerun all tests with rc3 and
> the Reference BLAS library.

The %!shared blocks don't work. However, they do if I put a comma 
between variables:

%!shared x t u   ## <- broken
%!shared x, t, u ## <- works

I guess that most (if not all) failures come from this kind of error.


I uploaded the rc3 test results. There have been no surprises.
http://wiki.octave.org/wiki/index.php?title=Octave-Forge&diff=6155&oldid=6152

The control package could be changed from “broken” to “ok” by chosing 
ReferenceBLAS instead of OpenBLAS during installation. Maybe the tests 
should better not depend on the BLAS library.

> > Otherwise, does anybody have any idea about why this fails ? Is there
> > anything wrong with the way I use %!shared and %!test blocks ?
> >
> > Finally, I can see on the wiki, in the last column : "Tests don't
> > cleanup temporary files". What does it mean ? How do you see that in the
> > log ?
>
> The files showed up in my working directory after running the tests. I
> don't think they are mentioned in the log. Maybe the files would be
> cleaned up, if the tests didn't fail? I have to investigate further...

The comment “Tests don't cleanup temporary files” accidentally has been 
put in the stk package row, but affects statistics instead. The stk 
package leaves no temporary test files behind in the working directory.

Oliver