Numba could not report the error of out of array in njit?

[chen-erqi]

My code is here:

@numba.njit
def f():
    a = np.zeros(10)
    print(a[10])

f()

this program can work and print "1.3351156154062717e-306"

the numba version is 0.54.0
python version is 3.8.10
numpy version is 1.20.3

[gmarkall]

An array with 10 elements has indices 0-9 - if you print the element at index 10 this is a value beyond the end of the array. You can catch errors like this with the boundscheck kwarg to the @njit decorator:

import numba
import numpy as np

@numba.njit(boundscheck=True)
def f():
    a = np.zeros(10)
    print(a[10])

f()

prints:

$ python repro.py 
debug: IndexError: index 10 is out of bounds for axis 0 with size 10
Traceback (most recent call last):
  File "/home/gmarkall/numbadev/issues/8914/repro.py", line 10, in <module>
    f()
IndexError: index is out of bounds

Generally boundscheck=True is slower than running without it, so it would usually only be used for debugging purposes.

[esc]

Generally boundscheck=True is slower than running without it, so it would usually only be used for debugging purposes.

Indeed, the relevant docs are here: https://numba.pydata.org/numba-doc/dev/reference/pysemantics.html?highlight=boundscheck#bounds-checking

[chen-erqi]

Generally boundscheck=True is slower than running without it, so it would usually only be used for debugging purposes.

Indeed, the relevant docs are here: https://numba.pydata.org/numba-doc/dev/reference/pysemantics.html?highlight=boundscheck#bounds-checking

thanks! I learn a lot from the docs.

[chen-erqi]

An array with 10 elements has indices 0-9 - if you print the element at index 10 this is a value beyond the end of the array. You can catch errors like this with the boundscheck kwarg to the @njit decorator:

import numba
import numpy as np

@numba.njit(boundscheck=True)
def f():
    a = np.zeros(10)
    print(a[10])

f()

prints:

$ python repro.py 
debug: IndexError: index 10 is out of bounds for axis 0 with size 10
Traceback (most recent call last):
  File "/home/gmarkall/numbadev/issues/8914/repro.py", line 10, in <module>
    f()
IndexError: index is out of bounds

Generally boundscheck=True is slower than running without it, so it would usually only be used for debugging purposes.

thanks a lot.
Futhermore, could you please explain why numba is not design with a boundary check, which seems like a fundmental work of coding.

[gmarkall]

Futhermore, could you please explain why numba is not design with a boundary check, which seems like a fundmental work of coding.

Because it's faster to omit the boundscheck - it's not designed without it, it's just that you have to opt-in to it.

[chen-erqi]

thanks a lot!
I learn a lot from these setting and will consider how to apply it to accelerate my structure of evolutionary algorithm.
In addition, I sincerely thanks all members of numba, because of numba, i can accelerate my code and finally my papar is accepted by a journal with if=8.653.
repect to you all!