Numba: Cannot determine the type of nested functions

I have this numba script (in a file called 'MySource.py') i'm trying to compile:

from numpy import empty_like, inf, minimum, max
from numba import pycc

cc = pycc.CC('base_model')
cc.verbose = True


@cc.export('cumMin','float64[:](float64[:])')  
def cumMin(A):
    r = empty_like(A)
    t = inf
    for i in range(len(A)):
        t = minimum(t, A[i])
        r[i] = t
    return r

# nm -C base_model.so > bingo.txt
@cc.export('da_comp','List(float64)(float64[:, :])')  
def da_comp(rolled_ndarray):

    this_series_p = rolled_ndarray[:, 0]
    this_series_v = rolled_ndarray[:, 1]    

    return [max(cumMin(this_series_p)), max(cumMin(this_series_v))]

    
if __name__ == "__main__":
    cc.compile()

In bash, I do:

 python3 MySource.py

but I get a compilation error:

    raise TypingError(msg, loc=inst.loc)
numba.core.errors.TypingError: Failed in nopython mode pipeline (step: nopython frontend)
Untyped global name 'cumMin': Cannot determine Numba type of <class 'function'>

File "MySource.py", line 24:
def da_comp(rolled_ndarray):
    <source elided>

    return [max(cumMin(this_series_p)), max(cumMin(this_series_v))]

I use numba 0.57.1 How to fix this?

#Edit:
I tried splitting the functions into two files as suggested in the answer by Eric below but this doesn't solve the issue (I get the same error).

It looks like the problem is that cumMin is not recognized by numba as a proper type. Numba, when asked to compile with the @cc.export, doesn't look for dependencies in global scope, so cumMin is not found when compiling da_comp. A solution is to compile both methods seperately and then load the module before compiling the second method.

Here is a possible solution:

1. compile cumMin separately.

from numba.pycc import CC
from numba import njit

@njit('float64[:](float64[:])')
def cumMin(A):
    r = empty_like(A)
    t = inf
    for i in range(len(A)):
        t = minimum(t, A[i])
        r[i] = t
    return r

# numba.pycc.CC() returns an instance of a subclass of numba.pycc.CompilerBase
cc = CC('cumMin_module')

# Export cumMin function
cc.export('cumMin', 'float64[:](float64[:])')(cumMin)
cc.compile()

2. load resulting module and use it:

from cumMin_module import cumMin

cc = CC('base_model')
cc.verbose = True

@cc.export('da_comp','float64[:](float64[:, :])')  
def da_comp(rolled_ndarray):
    this_series_p = rolled_ndarray[:, 0]
    this_series_v = rolled_ndarray[:, 1]    
   
    return [max(cumMin(this_series_p)), max(cumMin(this_series_v))]

cc.compile()

3. run code with bash:

python3 MySource.py

This way cumMin is compiled first and then available in da_comp.

so I solved the problem by replacing

@cc.export('cumMin','float64[:](float64[:])') 

by:

@numba.jit(nopython = True)

import numpy
import numba
import numba.pycc

cc = numba.pycc.CC('base_model')
cc.verbose = True


@numba.jit(nopython = True)
def cumMin(A):
    r = numpy.empty(len(A))
    t = numpy.inf
    for i in range(len(A)):
        t = numpy.minimum(t, A[i])
        r[i] = t
    return r

@cc.export('da_comp','List(float64)(float64[:, :])')  
def da_comp(rolled_ndarray):

    this_series_p = rolled_ndarray[:, 0]
    this_series_v = rolled_ndarray[:, 1]    

    t1 = cumMin(this_series_p)
    t2 = cumMin(this_series_v)


    return [numpy.max(t1), numpy.max(t2)]

    
if __name__ == "__main__":
    cc.compile()

#python3 MySource.py