Change from of 'c' to 'python' engine in read_csv gives unexpected side-effect

[olajir]

Code Sample

>>> import pandas as pd
>>>
>>> csv_file_contents=(
... """File: small.csv,,
... 10010010233,0123,654
... foo,,bar
... 01001000155,4530,898""")
>>> with open('small.csv', 'w') as fp:
...     fp.write(csv_file_contents)
>>>
>>> pd.read_csv('small.csv', header=None, engine='c',
...             names=['col1', 'col2', 'col3'],
...             dtype={'col1': str, 'col2': str, 'col3': str})

              col1  col2 col3
0  File: small.csv   NaN  NaN
1      10010010233  0123  654
2              foo   NaN  bar
3      01001000155  4530  898

>>> pd.read_csv('small.csv', header=None, engine='c',
...             names=['col1', 'col2', 'col3'],
...             dtype={'col1': str, 'col2': str, 'col3': str}).dropna()

          col1  col2 col3
1  10010010233  0123  654
3  01001000155  4530  898

>>> pd.read_csv('small.csv', header=None, engine='python',
...             names=['col1', 'col2', 'col3'],
...             dtype={'col1': str, 'col2': str, 'col3': str}).dropna()

              col1  col2 col3
0  File: small.csv   nan  nan
1      10010010233  0123  654
2              foo   nan  bar
3      01001000155  4530  898

Problem description

Reading an empty field from a csv-file returns NaN (dtype float) if read with the 'c' engine and 'nan' (string representation) if read with the 'python' engine.
In the example above the digits that shall be treated as strings and lines with empty data-fields shall be dropped. This works fine with the above code snippet if the 'c' engine is used. However if the 'python' engine is used the 'nan' fields are not dropped. This may break working code if one needs to switch from the 'c' engine to the 'python' engine.

Expected Output

It is expected that the two engines gives consistent output.

Output of pd.show_versions()

[paste the output of pd.show_versions() here below this line]
INSTALLED VERSIONS

commit: None
python: 3.6.5.final.0
python-bits: 64
OS: Darwin
OS-release: 17.5.0
machine: x86_64
processor: i386
byteorder: little
LC_ALL: en_US.UTF-8
LANG: en_US.UTF-8
LOCALE: en_US.UTF-8

pandas: 0.22.0
pytest: 3.5.1
pip: 9.0.3
setuptools: 39.0.1
Cython: None
numpy: 1.14.2
scipy: 1.0.1
pyarrow: None
xarray: None
IPython: 6.3.1
sphinx: None
patsy: 0.5.0
dateutil: 2.7.2
pytz: 2018.4
blosc: None
bottleneck: None
tables: None
numexpr: None
feather: None
matplotlib: 2.2.2
openpyxl: 2.5.3
xlrd: 1.1.0
xlwt: None
xlsxwriter: None
lxml: None
bs4: None
html5lib: 1.0.1
sqlalchemy: None
pymysql: None
psycopg2: None
jinja2: 2.10
s3fs: None
fastparquet: None
pandas_gbq: None
pandas_datareader: None

[WillAyd]

Is the issue just that the Python engine returns a string representation of NaN whereas the C parser returns the actually NaN value? If so, can you update your example to reflect that?

[olajir]

Yes, that's the issue. The example has been updated.

[mroeschke]

This looks to work on master now. Could use a test.

In [72]: >>> pd.read_csv('small.csv', header=None, engine='c',
    ...: ...             names=['col1', 'col2', 'col3'],
    ...: ...             dtype={'col1': str, 'col2': str, 'col3': str}).dropna()
Out[72]:
          col1  col2 col3
1  10010010233  0123  654
3  01001000155  4530  898

In [73]: >>> pd.read_csv('small.csv', header=None, engine='python',
    ...: ...             names=['col1', 'col2', 'col3'],
    ...: ...             dtype={'col1': str, 'col2': str, 'col3': str}).dropna()

Out[73]:
          col1  col2 col3
1  10010010233  0123  654
3  01001000155  4530  898

In [74]: pd.__version__
Out[74]: '0.26.0.dev0+593.g9d45934af'