Re: release 3.0.2

[Jaroslav Hajek]

On Tue, Aug 19, 2008 at 12:21 PM, Michael Goffioul
<address@hidden> wrote:
> On Tue, Aug 19, 2008 at 10:12 AM, Jaroslav Hajek <address@hidden> wrote:
>> It's going to happen, but I have no access. John is just waiting
>> whether all the regular builders can build this release. I
>> pre-announced the release two weeks ago, but perhaps I should have
>> made a second "last call".
>
> I just tried to compile and run 3.0.2 with MSVC. It compiles fine and
> reports 2 test failures:
>
> 1) It seems there's a typo in the test code
>
>>>>>> processing 
>>>>>> d:\Sources\c\octave\octave-forge-svn\admin\Windows\msvc\downloaded_packages\octave-3.0.2\src/DLD-FUNCTIONS\rand.cc
>  ***** test
>  % Test fixed state
>  randp("seed",1);
>  assert(randp(1e9,1,6),[1000006208 1000012224 999981120 999963520
> 999963072 999981440]],1e-6))
> !!!!! test failed
> parse error near line 231 of file ./fntests.m
>
>  syntax error
>
>>>> function [  __random_statistical_tests__] = __test__( 
>>>> __random_statistical_tests__)
>
>  % Test fixed state
>  randp("seed",1);
>  assert(randp(1e9,1,6),[1000006208 1000012224 999981120 999963520
> 999963072 999981440]],1e-6))
> endfunction
>
>                   ^
>
> shared variables {
>  __random_statistical_tests__ = 0
> }
>

Damn, I would swear I've removed that extra bracket. Still, I don't
think this is a serious problem.


> 2) This one, I don't know what to do. When debugging, it seems
> xdnrm2 returns Inf.
>
>>>>>> processing 
>>>>>> d:\Sources\c\octave\octave-forge-svn\admin\Windows\msvc\downloaded_packages\octave-3.0.2\src\data.cc
>  ***** assert(norm([1e200, 1]), 1e200);
> !!!!! test failed
> error: assert (norm ([1e200, 1]),1e200) expected
>  1.0000e+200
> but got
> Inf
> Infs don't match
> shared variables {
>  x =
>
>     1  -3   4   5  -7
>
> }
>
> Michael.
>

Depending on what BLAS you use, I think this can happen. I think the
earlier reference versions from netlib didn't do the "online" scaling.
I've submitted a patch recently to completely rework `norm', so this
issue should go away.


-- 
RNDr. Jaroslav Hajek
computing expert
Aeronautical Research and Test Institute (VZLU)
Prague, Czech Republic
url: www.highegg.matfyz.cz