Numba: "RuntimeError: missing Environment" using print in second call to cached numba function

[brunojacobs]

I noticed a bug in the latest version of Numba (0.41.0) which involves repeatedly calling a cached Numba JIT compiled function that contains a print statement. To be more precise, everything works as expected for the first call, after which the function is cached. For subsequent calls to the same (now cached) function, a RuntimeError: missing Environment is thrown. This issue is related to issue #2411 which was fixed in a recent release.

Example: numba_print_test.py

from numba import njit

@njit(cache=True)
def test_inner(x):
    print(x)


@njit(cache=True)
def test_outer(x):
    test_inner(x)


test_outer(5)

The issue I observe is as follows: If cache=True for both the inner and outer function, everything works fine for the first call to numba_print_test.py. After this initial call the function is cached and no longer works for subsequent calls:

$ python numba_print_test.py 
5
$ python numba_print_test.py 
Traceback (most recent call last):
  File "numba_print_test.py", line 14, in <module>
    test_outer(5)
RuntimeError: missing Environment

It works fine if cache=False for both the inner and outer function:

$ python numba_print_test.py 
5
$ python numba_print_test.py 
5

It works fine if we set cache=True for the inner function, but keep cache=False for the outer function:

$ python numba_print_test.py 
5
$ python numba_print_test.py 
5

It no longer works if we set cache=False for the inner function and cache=True for the outer function:

$ python numba_print_test.py 
5
$ python numba_print_test.py 
Traceback (most recent call last):
  File "numba_print_test.py", line 14, in <module>
    test_outer(5)
RuntimeError: missing Environment

This seems to suggest that the issue is related to calling the cached outer function. In addition, if I delete the __pycache__ folder I can successfully call numba_print_test.py function once more, after which subsequent calls will again throw the RuntimeError.

[stuartarchibald]

Thanks for the report, I can reproduce, this indeed likely to be related to https://github.com/numba/numba/issues/2411

[songololo]

Likewise, encountering the same issue. In my case it seemed to work fine for a few cases but then started throwing the error predictably after the first run.

Temporarily resolved by disabling caching per @brunojacobs description.

PS - I am using v0.43.1

[songololo]

Curious if this is still on the radar? Thanks.

[stuartarchibald]

@songololo Yes it's still known, there's now a caching label, and sometime next year we're going to have a big caching fixing session as there's a number of interconnected things that need addressing.