Aggressive typecasting with specialization leads to silent wrong value output

[ZzEeKkAa]

Reporting a bug

• I have tried using the latest released version of Numba (most recent is
  visible in the release notes
  (https://numba.readthedocs.io/en/stable/release-notes-overview.html).
• I have included a self contained code sample to reproduce the problem.
  i.e. it's possible to run as 'python bug.py'.

Following code implicitly type cast float64 to int32, so the data after dot is getting lost.

import numba
from numba import int32


@numba.njit(int32(int32))
def inc(a):
    return a+1

if __name__ == "__main__":
    b = 1.5
    b1 = inc(b) # <- does not raise exception
    assert b1 == 2.5 # actual value is 2, because float(1.5) was cast to int32. 

Instead I would expect to see exception based on the https://en.cppreference.com/w/cpp/language/implicit_conversion . It may be either exception, either warning, but never silent loss of data.

[kc611]

Hi @ZzEeKkAa , Thanks for reporting this bug. I can reproduce the issue locally.

This does seem like a case of implicit unsafe casting behaviour that would need to be addressed.