Following Pull Request #89, it seems that the way gradient_t is defined (using rowVectors) is causing it to fail.
Indeed, in the line
https://github.com/roboptim/roboptim-core/blob/master/include/roboptim/core/numeric-quadratic-function.hxx#L120
the assignment gradient.noalias() = 2 * a_ * x; triggers an unwanted allocation, that causes the test suite to fail if ROBOPTIM_DO_NOT_CHECK_ALLOCATION is set to true.
Following Pull Request #89, it seems that the way gradient_t is defined (using rowVectors) is causing it to fail.
Indeed, in the line
https://github.com/roboptim/roboptim-core/blob/master/include/roboptim/core/numeric-quadratic-function.hxx#L120
the assignment gradient.noalias() = 2 * a_ * x; triggers an unwanted allocation, that causes the test suite to fail if ROBOPTIM_DO_NOT_CHECK_ALLOCATION is set to true.