Cannot convert unicode or char dtype to longer fixed size

[elfjes]

Reporting a bug

• I have tried using the latest released version of Numba (most recent is
  visible in the change log (https://github.com/numba/numba/blob/master/CHANGE_LOG).
• I have included below a minimal working reproducer (if you are unsure how
  to write one see http://matthewrocklin.com/blog/work/2018/02/28/minimal-bug-reports).

I'm trying to increase the (fixed) length string/unicode dtype of an array. I've tried a number of ways, but none of them work. I think the following two methods should work however, but fail with the error given in the comments:

import numba
import numpy as np

@numba.njit
def assign_in_array(source, destination):
    destination[:] = source

@numba.njit
def using_astype(source, destination):
    converted =source.astype(destination.dtype)
    destination[:] = converted


if __name__ == '__main__':
    assert numba.__version__ == '0.51.2'
    src = np.array(['val'], dtype='S4') # also U4/U8 doesn't work
    dst = np.empty_like(src, dtype='S8')

    assign_in_array(src, dst)
    # numba.core.errors.TypingError: Failed in nopython mode pipeline (step: nopython frontend)
    # No implementation of function Function(<built-in function setitem>) found for signature:
    #
    #  >>> setitem(array([unichr x 8], 1d, C), slice<a:b>, array([unichr x 4], 1d, C))
    #
    # There are 16 candidate implementations:
    #   - Of which 16 did not match due to:
    #   Overload of function 'setitem': File: <numerous>: Line N/A.
    #     With argument(s): '(array([unichr x 8], 1d, C), slice<a:b>, array([unichr x 4], 1d, C))':
    #    No match.

    using_astype(src, dst)
    # numba.core.errors.TypingError: Failed in nopython mode pipeline (step: nopython frontend)
    # - Resolution failure for literal arguments:
    # astype([char x 8]) not supported on array([char x 4], 1d, C): cannot convert from [char x 4] to [char x 8]
    # - Resolution failure for non-literal arguments:
    # None

Is there any other way that might work?

[esc]

@elfjes thank you for reporting this, I can confirm the behaviour you describe. I can also confirm that the code will work fine without the Numba decorators.

[elfjes]

@esc thanks for your reply. Is there any other way I might be able to do this within numba? Would it perhaps be possible to do something with an @overload decorator?

[stuartarchibald]

I think this is due to a number of valid casts having not been implemented. One path that should work is this:

import numba
import numpy as np

@numba.njit(boundscheck=True)
def assign_in_array(source, destination):
    for i in range(len(destination)):
        destination[i] = str(source[i])

if __name__ == '__main__':
    src = np.array(['val'], dtype='U4')
    dst = np.empty_like(src, dtype='U8')

    assign_in_array(src, dst)

[esc]

@stuartarchibald I tried that, but only with 'S4' and 'S8' and that failed too with:

Traceback (most recent call last):
  File "issue6484.py", line 25, in <module>
    convert(src, dst)
  File "/Users/esc/git/numba/numba/core/dispatcher.py", line 414, in _compile_for_args
    error_rewrite(e, 'typing')
  File "/Users/esc/git/numba/numba/core/dispatcher.py", line 357, in error_rewrite
    raise e.with_traceback(None)
numba.core.errors.TypingError: Failed in nopython mode pipeline (step: nopython frontend)
No implementation of function Function(<class 'str'>) found for signature:

 >>> str([char x 4])

There are 6 candidate implementations:
  - Of which 6 did not match due to:
  Overload of function 'str': File: <numerous>: Line N/A.
    With argument(s): '([char x 4])':
   No match.

During: resolving callee type: Function(<class 'str'>)
During: typing of call at issue6484.py (19)


File "issue6484.py", line 19:
def convert(source, destination):
    <source elided>
    for i in range(len(source)):
        destination[i] = str(source[i])

[stuartarchibald]

@esc I anticipate it would fail given that implemented casts do not cover that case.

[esc]

I have re-labelled this as a feature request to add the missing casts.