Return correct value for pinv of zero vectors and scalars (bug #51246). CMatrix.cc, dMatrix.cc (pseudo_inverse): Change any computed tolerance of 0 to std::numeric_limits<double>::min() so that zero values are later dropped from calculation. fCMatrix.cc, fMatrix.cc (pseudo_inverse): Change calculation of default tolerance to us the <float> version of std::numeric_limits::epsilon () rather than the double version. Change any computed tolerance of 0 to std::numeric_limits<float>::min() so that zero values are later dropped from calculation. * pinv.cc (Fpinv): Add BIST tests.

--- a/liboctave/array/dMatrix.cc	Fri Nov 03 15:49:18 2017 +0100
+++ b/liboctave/array/dMatrix.cc	Fri Nov 03 10:03:36 2017 -0700
@@ -672,6 +672,9 @@
         tol = nr * sigma.elem (0) * std::numeric_limits<double>::epsilon ();
       else
         tol = nc * sigma.elem (0) * std::numeric_limits<double>::epsilon ();
+
+      if (tol == 0)
+        tol = std::numeric_limits<double>::min ();
     }
 
   while (r >= 0 && sigma.elem (r) < tol)