Strange error with `ushprod`

[juliasmoothoptimizers]

Strange error with `ushprod`

[dpo]

The Fortran source code says that VECTOR, the vector with which you compute the product, should be of size n. That's probably an error.

[dpo]

This results in no errors at all:

diff --git a/issue126.jl b/issue126.jl
index ed23ffa..24b3de3 100644
--- a/issue126.jl
+++ b/issue126.jl
@@ -22,12 +22,14 @@ function foo()
     err_sum1 = 0.0
     err_sum2 = 0.0
 
+    vect = ones(Cdouble, n)
+
     nnzr, Ir, r = Cint[0], Array{Cint}(n), zeros(n)
-    ushprod(io_err, Cint[n], goth, x0, Cint[1], Cint[1], Cdouble[1.0], nnzr, Ir, r)
+    ushprod(io_err, Cint[n], goth, x0, Cint[1], Cint[1], vect, nnzr, Ir, r)
     CUTEst.@cutest_error
     for i = 1:N
       nnzr2, Ir2, r2 = Cint[0], Array{Cint}(n), zeros(n)
-      ushprod(io_err, Cint[n], goth, x0, Cint[1], Cint[1], Cdouble[1.0], nnzr2, Ir2, r2)
+      ushprod(io_err, Cint[n], goth, x0, Cint[1], Cint[1], vect, nnzr2, Ir2, r2)
       CUTEst.@cutest_error
       if norm(r - r2) > 1e-6
         err_count1 += 1
@@ -41,7 +43,7 @@ function foo()
 
     for i = (N+1):2N
       nnzr2, Ir2, r2 = Cint[0], Array{Cint}(n), zeros(n)
-      ushprod(io_err, Cint[n], goth, x0, Cint[1], Cint[1], Cdouble[1.0], nnzr2, Ir2, r2)
+      ushprod(io_err, Cint[n], goth, x0, Cint[1], Cint[1], vect, nnzr2, Ir2, r2)
       CUTEst.@cutest_error
       if norm(r - r2) > 1e-6
         err_count2 += 1
@@ -64,4 +66,4 @@ function foo()
   end
 end
 
-foo()
\ No newline at end of file
+foo()

The size of VECTOR is consistent in the library but I wonder if it's intentional.

[abelsiqueira]

Ah, thanks. I though it would use the value of nnz_vector. Not a bug, so closing.

[dpo]

We should be sure our doc is clear about that.

[abelsiqueira]

For the core interface I just copied the DESCRIPTION in the man. The only further description is the function call and the types of each parameter. Should I add the disclaimer above as additional information?

[dpo]

It should say somewhere that the input and output vectors are of size n, otherwise, it's tempting to think that they should only contain the nonzero elements.