Closed
Description
I don't have access to a PC so cannot replicate directly (if anyone else can would be appreciated) but going through #28510 there were a few Windows with dtype preservation, which would look something as follows:
>>> df = pd.DataFrame(range(3))
>>> df.dtypes
0 int64
dtype: object
>>> pd.read_json(df.to_json()).dtypes
0 int32
dtype: object
Note that this may also be irrelevant if #28512 happens