cbrt test failure on Win32/MSVC

[John W. Eaton]

On 21-Oct-2011, Michael Goffioul wrote:

| While analyzing test failures under Win32/MSVC I found that cbtr(0) ->
| NaN. while the expected value is 0. MSVC does not provide cbrt
| implementation, so the octave implementation is used, but that
| implementation definitely returns NaN when the argument is 0 (because
| it computes 0/0). Shouldn't the test be changed, as it's weird to have
| a test that is known to fail when using octave's own code?

I guess something should be fixed, but I'm not sure it is the test.  I
didn't even know we had a cbrt function and I had to think for a
moment about what it is.  Ah, Jaroslav added.  Maybe you can get him
to fix the replacement code?  :-/

It looks like gnulib has some tests for cbrt but does not provide a
replacement function even though it seems to acknowledge that it is
not present on all systems.  It looks like glibc provides a version of
cbrt that we could probably use.  It would be nice to have it as a
gnulib module, but if not that, then maybe we could at least use the
code directly in Octave.

jwe