Buffer types are only allowed as local function variables, but what I do

Question

Cython doesn't like numpy arrays in closure?

%%cython
import numpy as np
cimport numpy as np

def f(np.ndarray[double, ndim=1] a):
    def g (double b):
        return a+b

    return g(1)

Using the stable version 0.24, I get:

Error compiling Cython file:
------------------------------------------------------------
...
import numpy as np
cimport numpy as np

def f(np.ndarray[double, ndim=1] a):
     ^
------------------------------------------------------------

cython_magic.pyx:4:6: Buffer types only allowed as function local variables

If I get rid of the definition g, it compiles / works fine.

Answer 1

There are several rounds:

• Assign a type to a variable only inside an internal function:

  def f(a):
      def g (double b):
          cdef np.ndarray[double, ndim=1] a_typed = a
          return a_typed+b
  
      return g(1)
  

  This is a small cost associated with checking the type of each call g, the value of which depends on what other work you are doing in g.

• Use a mixture of memory and untyped variables.

  def f(a):
     cdef double[:] memview_of_a = a
  
     def g(double b):
         memview_of_a[0] = 0 # example of indexing operation
         return a+b
  
     return g(1)
  

  , memview_of_a a , . . , , , .

, , ( ).