polyfit test fails

[Ben Abbott]

On Friday, May 22, 2009, at 02:19AM, "Benjamin Lindner" <lindnerben at gmx.net> wrote:
>Marco Atzeri wrote:
>> Dear all,
>> on latest hg source, on cygwin-1.7, I have
>> 
>> 
>>>>>>> processing /pub/hg/octave_clone/scripts/polynomial/polyfit.m
>>   ***** test
>>  x = [ -1196.4, -1195.2, -1194, -1192.8, -1191.6, -1190.4, -1189.2, -1188, \
>>        -1186.8, -1185.6, -1184.4, -1183.2, -1182];
>>  y = [ 315571.7086, 315575.9618, 315579.4195, 315582.6206, 315585.4966,    \
>>        315588.3172, 315590.9326, 315593.5934, 315596.0455, 315598.4201,    \
>>        315600.7143, 315602.9508, 315605.1765 ];
>>  [p1, s1] = polyfit (x, y, 10);
>>  [p2, s2, mu] = polyfit (x, y, 10);
>>  assert (2*s2.normr < s1.normr)
>> !!!!! test failed
>> 
>> 
>> 
>> octave:5> 2*s2.normr
>> ans =  0.12583
>> octave:6>  s1.normr
>> ans =  0.11466
>> 
>> 
>> is this fault cygwin specific or do you see it 
>> on other platform ?
>> 
>
>I see this test also failing with a mingw32 build.
>A win-specific feature ?
>
>benjamin

If I understand what is happening correctly, it *appears* that when the test fails for Windows, it does so because the s1 results are better than expected.

Does anyone have an idea why that is so? ... also are the s1 results really better (which looks surprising) or is the s1.normr incorrect?

Ben