API/BUG: unnecessary upcasting for df.update

[h-vetinari]

I looked through all the open issues with update and dtype in them, but didn't find something - apologies if an issue exists already.

Consider:

>>> import pandas as pd
>>> s = pd.Series([10, 11, 12])
>>> t = pd.Series([61, 63], index=[1, 3])
>>> s.update(t)
>>> s
0    10
1    61
2    12
dtype: int64

So far, so good. Now, for frames:

>>> df = s.to_frame('A')
>>> df.update(t.to_frame('A'))
>>> df
      A
0  10.0
1  61.0
2  12.0

Obviously, the dtype promotion here was neither desired nor necessary (strictly speaking; s and t are both int64), and is simply a by-product of the alignment happening behind the scenes.

[gfyoung]

Your issue makes sense to me!

[lithomas1]

Looks fixed on main at least.