Allow parsing dates of the form "1899-12-30" in a jit context

[bmironov]

• I am 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).

Hello,

First, thanks for your hard work and this great library.
Recently I came across small issue that is stopping me from using numba in my project. We need to parse dates that are adjusted as floats from Dec 30, 1899 (this format is used in Excel).
Below is details of my environment and content of Jupyter cells:

Thanks again and best regards

Numba 0.44.1
Running through Jupyter

import numba
import numpy as np

print(np.datetime64("1899-12-30"))

1899-12-30

@numba.jit
def test():
    t = np.datetime64("1899-12-30")
    return 0

test()

<ipython-input-50-f403da816d8c>:1: NumbaWarning: 
Compilation is falling back to object mode WITH looplifting enabled because Function test failed at nopython mode lowering due to: Cannot cast Literal[str](1899-12-30) to datetime64[]: %".16" = load i8*, i8** %"$const0.3"

File "<ipython-input-50-f403da816d8c>", line 3:
def test():
    t = np.datetime64("1899-12-30")
    ^

[1] During: lowering "$0.4 = call $0.2($const0.3, func=$0.2, args=[Var($const0.3, <ipython-input-50-f403da816d8c> (3))], kws=(), vararg=None)" at <ipython-input-50-f403da816d8c> (3)
  @numba.jit
/Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages/numba/compiler.py:725: NumbaWarning: Function "test" was compiled in object mode without forceobj=True.

File "<ipython-input-50-f403da816d8c>", line 2:
@numba.jit
def test():
^

  self.func_ir.loc))
/Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages/numba/compiler.py:734: NumbaDeprecationWarning: 
Fall-back from the nopython compilation path to the object mode compilation path has been detected, this is deprecated behaviour.

For more information visit http://numba.pydata.org/numba-doc/latest/reference/deprecation.html#deprecation-of-object-mode-fall-back-behaviour-when-using-jit

File "<ipython-input-50-f403da816d8c>", line 2:
@numba.jit
def test():
^

  warnings.warn(errors.NumbaDeprecationWarning(msg, self.func_ir.loc))

[esc]

Hi @bmironov

Thanks for the praise, we really appreciate it! Thank you for considering Numba and taking the time to report this error.

I can confirm that parsing such strings you mentioned in a @jit context currently does not seem to work. What you could do in the meantime as a workaround, is to create your np.datetime objects outside of the context and then pass them in as arguments like so:

In [8]: mydate = np.datetime64("1899-12-30")

In [9]: @numba.njit
   ...: def test(da):
   ...:     print(da)
   ...:

In [10]: test(mydate)
1899-12-30

Would that be feasible for your use-case?

[bmironov]

Hi @esc
Thanks, this idea works for me

Look forward for permanent solution :)

[esc]

@bmironov thanks for reporting back. I will convert your issue into a feature request!