Adding element to empty object Series changes dtype

[toobaz]

Code Sample, a copy-pastable example if possible

In [2]: s = pd.Series(dtype='object')

In [3]: s.loc[0] = 1

In [4]: s
Out[4]: 
0    1
dtype: int64

In [5]: s = pd.Series(dtype='object')

In [6]: s.loc[0] = 1.

In [7]: s
Out[7]: 
0    1.0
dtype: float64

Compare with:

In [8]: s = pd.Series({1 : 2}, dtype='object')

In [9]: s.loc[0] = 1

In [10]: s
Out[10]: 
1    2
0    1
dtype: object

In [11]: s = pd.Series({1 : 2.}, dtype='object')

In [12]: s.loc[0] = 1.

In [13]: s
Out[13]: 
1    2
0    1
dtype: object

Problem description

The dtype of a Series should change, on addition of new elements, only when it has to:: as a corollary, it should never change when it is object.

Vaguely related to #17261, maybe to #8902.

Expected Output

Always object dtype.

Output of pd.show_versions()

INSTALLED VERSIONS

commit: None
python: 3.5.3.final.0
python-bits: 64
OS: Linux
OS-release: 4.9.0-5-amd64
machine: x86_64
processor:
byteorder: little
LC_ALL: None
LANG: it_IT.UTF-8
LOCALE: it_IT.UTF-8

pandas: 0.23.0.dev0+229.gd5a7e7c94
pytest: 3.2.3
pip: 9.0.1
setuptools: 36.7.0
Cython: 0.25.2
numpy: 1.12.1
scipy: 0.19.0
pyarrow: None
xarray: None
IPython: 6.2.1
sphinx: 1.5.6
patsy: 0.4.1
dateutil: 2.6.1
pytz: 2017.2
blosc: None
bottleneck: 1.2.0dev
tables: 3.3.0
numexpr: 2.6.1
feather: 0.3.1
matplotlib: 2.0.0
openpyxl: 2.3.0
xlrd: 1.0.0
xlwt: 1.3.0
xlsxwriter: 0.9.6
lxml: 4.1.1
bs4: 4.5.3
html5lib: 0.999999999
sqlalchemy: 1.0.15
pymysql: None
psycopg2: None
jinja2: 2.10
s3fs: None
fastparquet: None
pandas_gbq: None
pandas_datareader: 0.2.1

[jreback]

see #6485
and #6942

this is actually a feature

[toobaz]

I don't think it is related: #6485 is an unwanted upcasting (while I'm seeing an unwanted downcasting), example in #6942 is not complete but seems to be specific to datetimelike stuff.

Most importantly, this one specifically affects empty Series only. Are you referring to this when you say it is a feature? (Might be, but I think it goes against the principle of least surprise, because in the examples above I specifically passed a dtype=object, which is not being respected)

[jreback]

yes the point is a common use case
is to create an empty object and iteratively fill it
not a great pattern but one which is supported
inference on first expansion is ingrained

[toobaz]

is to create an empty object and iteratively fill it

Ideally, we would then have the possibility of dtype-less empty objects. So that if you do pass a dtype, it's not overridden.

[jreback]

i agree in theory we should have an Any type which does exactly this