merge of int and float column results in column of dtype object

[gnebehay]

Code Sample, a copy-pastable example if possible

import pandas as pd

A = pd.DataFrame({'X': [1, 2, 3]})
B = pd.DataFrame({'Y': [1.0, 3.0]})

B = B.merge(A, left_on='Y', right_on='X')

print(B.dtypes)

Problem description

The output is

Y    object
X     int64
dtype: object

meaning that Y was incorrectly upcast to object.

In pandas 0.19, the dtype of B.Y after merging is float64.
In pandas 0.20, the dtype of B.Y after merging is object.

The problem only occurs if some keys in B.Y are missing (try adding 2.0)

Expected Output

Y    float64
X      int64
dtype: object

Output of pd.show_versions()

---

commit: None python: 3.5.3.final.0 python-bits: 64 OS: Linux OS-release: 4.9.0-2-amd64 machine: x86_64 processor: byteorder: little LC_ALL: None LANG: en_US.UTF-8 LOCALE: en_US.UTF-8

pandas: 0.20.1
pytest: None
pip: 9.0.1
setuptools: 35.0.2
Cython: None
numpy: 1.12.1
scipy: 0.19.0
xarray: None
IPython: 6.0.0
sphinx: None
patsy: None
dateutil: 2.6.0
pytz: 2017.2
blosc: None
bottleneck: None
tables: None
numexpr: None
feather: None
matplotlib: 2.0.2
openpyxl: None
xlrd: None
xlwt: None
xlsxwriter: None
lxml: None
bs4: None
html5lib: 0.999999999
sqlalchemy: None
pymysql: None
psycopg2: None
jinja2: 2.9.6
s3fs: None
pandas_gbq: None
pandas_datareader: None

[jreback]

this does looks like a bug. I don't think we have this as a test. This is a fairly narrow case though, the floats have to be exactly equal to the integers. welcome to have a PR. I can point you to where the issue is.

[gnebehay]

I encountered this issue while getting data via pd.read_sql(). There, primary keys are returned as ints (since there are no missing values), but nullable foreign keys are upcast to float (so that nulls can be represented as nan). If you then do a merge on these columns, you get what is described above. If you point me to a code location, I will see what I can do.

[jreback]

Its right about here: https://github.com/pandas-dev/pandas/blob/master/pandas/core/reshape/merge.py#L911

We allow differing int types, e.g. int64 and int8 are ok. but int64 and float will fall thru. you can add a check like

In [8]: lk = np.array([1.0, 2, 3])

In [9]: rk = np.array([2, 3, 4])

In [11]: if is_float_dtype(lk) and is_integer_dtype(rk):
    ...:     if (lk == lk.astype('int64')).all():
    ...:         continue
             # check opposite way as well