Incorrect sparse assignment from Map to SparseMatrix

Summary

The last element of the value array is set to zero when mapping from raw buffers (using column compressed storage) to a SparseMatrix (Map -> SparseMatrix).

Environment

  • Operating System : macOS Big Sur 11.7
  • Architecture : x64
  • Eigen Version : 3.3.9
  • Compiler Version : clang 12.0.5
  • Compile Flags : irrelevant
  • Vector Extension : none

Minimal Example

#include "Eigen/Sparse"
#include <iostream>

int main()
{
    /** create a 3x3 diagonal matrix (with {1,2,3} on the diagonal) using column compressed storage format. */
    int* outter = new int [3];
    int* inner  = new int [3];
    double* values = new double [3];

    outter[0] = 0; outter[1] = 1; outter[2] = 2;
    inner[0] = 0; inner[1] = 1; inner[2] = 2;
    values[0] = 1; values[1] = 2; values[2] = 3;

    Eigen::SparseMatrix<double> mat = Eigen::Map<Eigen::SparseMatrix<double>>(3,3,3,outter,inner,values);
    std::cout << mat << "\n";

    return 0;
}

Steps to reproduce

  1. Compile and run the minimal example.
  2. Observe the output:
1 0 0 
0 2 0 
0 0 0
  1. The correct output should be:
1 0 0 
0 2 0 
0 0 3

What is the current bug behavior?

The last element of the value array is replaced by zero (0), perhaps, during the sparse assignment. In this example, a temporary object is also created during the assignment, whereas it should not? I could not track down where exactly this happens but the Map object is created correctly.

What is the expected correct behavior?

We should see:

1 0 0 
0 2 0 
0 0 3

The values array of mat should contain {1,2,3}, instead it is {1,2,0}.