Using pointer to C++ vectors in Cython

I was trying define pointers to C++ vectors in Cython, and I noticed a strange behavior that I cannot explain with my Cython knowledge. Below, I have created a pointer to a vector of double based on T range. I either assign a to a_low or a_high.

from libcpp.vector cimport vector

cdef vector[double] a_low = [1.0, 2e-03, -4e-07, 6.3e-10, -2.21e-5, -3.45e+01];
cdef vector[double] a_high = [-2.21e-01, 4.1e-03, -6.12e-02, 1.65e-09, -9.3e-6, -1.4e+02];

def h_calc(double T):
    cdef vector[double] *a
    cdef double h
    if (T<1000.0):
        a = &a_low;
    else:
        a = &a_high;
    print(a[0])
    h = 0.0;
    return h;

executing h_calc(1200); gives out [-0.221, 0.0041, -0.0612, 1.65e-09, -9.3e-06, -140.0] which results from print(a[0]). Here, because T>1000, we see a_high was displayed as output.

What I don't understand is why a[0] does not return the first value of the vector a_high , but rather the whole vector?

In Cython dereferencing (i.e. *a in c++) is spelled as a[0] as the former is not valid python syntax while the latter is - see the "Types" section here

I.e. whats happening is:

• your a is of type vector[double] *,
• a[0] is the dereferenced version (i.e. just a vector[double])
• print(a[0]) converts vector[double] to a python list and prints that (i.e. the whole list)
• (this is also why a[1] crashes - it's effectively doing *(a+1), and getting a memory violation)

Also for what it's worth the assignment from the python list should work - Cython effectively rewrites it to loop over the python list calling push_back(...) on each item for you.

[] is a PyObject. You can't assign it to a c++ type (as evidenced with cdef) of a vector. The conversion isn't implicit. Likely the whole string of values is being treated as the first element. What happens when you access a[1]?

You have to declare the vector, and assign the values to it separately.


cdef vector[double] a_low



a_low.push_back(1.0)
a_low.push_back(2e-03)
...


Perhaps you can also try experimenting with
vector::insert (iterator position, const value_type& val);, which can take an array of elements. Again, you would have to pass it a c array type, so you would have to declare and fill your array before passing it.

cdef double array[6]

array[:] = [1.0, 2e-03, -4e-07, 6.3e-10, -2.21e-5, -3.45e+01] then pass it to vector::insert.

As you can see you can fill a C style array but not a vector in Cython in a one-liner. This raises the question, why not just use arrays in the first place?