Re: sqrtm test failure

[Marco atzeri]

On 7/29/2011 11:06 PM, John W. Eaton wrote:
> On 29-Jul-2011, Marco atzeri wrote:
>
> | on src/DLD-FUNCTIONS/sqrtm.cc
> | the following test is failing
> |
> | %!assert (isnan (sqrtm ([0 1; 0 0])))
> |
> | but it is not clear to me, what is the expected results
> | as isnan applied to a matrix produces a matrix
>
> The assert will succeed only if all elements of the matrix returned by
> isnan are true.
>
> |>  sqrtm ([0 1; 0 0])
> | warning: sqrtm: matrix is singular, may not have a square root
> | ans =
> |
> |       0   NaN
> |       0   NaN
> |
> |
> |>  isnan(sqrtm ([0 1; 0 0]))
> | warning: sqrtm: matrix is singular, may not have a square root
> | ans =
> |
> |     0   1
> |     0   1
>
> On my system, I see
>
>    octave>  sqrtm ([0,1; 0, 0])
>    warning: sqrtm: matrix is singular, may not have a square root
>    ans =
>
>       NaN - NaNi   NaN - NaNi
>       NaN - NaNi   NaN - NaNi
>
> I'm not sure why this result should be complex.
>
> jwe

probably to avoid that the real or img of NaN is wrongly reported as
a number.

I will add this to my list of tests that fails on
cygwin for different NaN implementation.

Regards
Marco