_round produces some unexpected results on certain numbers

[aecorn]

Describe the bug

Even with ROUND_HALF_UP, some pre-rounded numbers will not work well.
The recommendation is not rounding several times during a production line, but it still happens.

To Reproduce

import numpy as np
from decimal import ROUND_HALF_UP, Decimal, localcontext
import pandas as pd

x = np.float64(103.55)
with localcontext() as ctx:
    ctx.rounding = ROUND_HALF_UP
    result = round(Decimal(x), 1)
result

Expected behaviour

np.float64(103.55) with ROUND_HALF_UP to one decimal place is expected to be 103.6 nor 103.5.
This happens because the actual np.float is 103.54999999...
A possible "hack" around this is using the str() version of the number. But this might be dangerous business dependant on the datatype of the float...

Because of the extra str(), this gives 103.6

import numpy as np
from decimal import ROUND_HALF_UP, Decimal, localcontext
import pandas as pd

x = np.float64(103.55)
with localcontext() as ctx:
    ctx.rounding = ROUND_HALF_UP
    result = round(Decimal(str(x)), 1)
result

Platforms and Environments

DaplaLab with Jupyter

Version

1.4.0

Error messages or logs