BUG: Series dtype not as specified when values are missing

[emmanuelito]

Pandas version checks

• I have checked that this issue has not already been reported.

• I have confirmed this bug exists on the latest version of pandas.

• I have confirmed this bug exists on the main branch of pandas.

Reproducible Example

# In 1.4.2
import pandas as pd
s = pd.Series( index=[1, 2, 3 ], dtype=int)
s.dtype
# -> dtype('float64') *ouch*

# ---- vs in 1.3.5
s = pd.Series( index=[1, 2, 3 ], dtype=int)
s.dtype
# -> dtype('int64') *correct*

Issue Description

pd.Series() does not respect the specified dtype and breaks compatibility with previous (1.3.5) version.

Expected Behavior

Should be backward compatible with 1.3.5: the result should have the dtype specified by the caller.

This issue seems (I'm not sure) related to #46672

The problem is that missing values (unspecified in the constructor) are initialized to NaNs, and thus floats, whereas in Pandas 1.3.5 they were initialized with zeroes.

Installed Versions

[jreback]

this was incorrect before and is now fixed

[jreback]

this incorrect before and is now fixed

int cannot hold missing values

[emmanuelito]

But this modification badly breaks code. Isn't Pandas supposed to be backward compatible ?
IMHO raising an exception would be better than changing the dtype.

[jreback]

the usage has always been wrong