pandas.read_csv(..., skiprows=2, engine='c') with unix-style line breaks crashes python on windows

[e-pet]

The following makes python crash ("Kernel died, restarting" in IPython) on my windows 7 machine:

import pandas
myfile = open("test.csv", "w", newline="\n")
myfile.write("blah\n\ncol_1,col_2,col_3\n\n")
myfile.close()
dat = pandas.read_csv("test.csv", skiprows=2, encoding="utf-8", engine="c")

Note the unix-style line breaks.

The test case seems to be pretty precise, since changing about anything leads to working code. I tried, e.g.,

• with Windows-style line breaks ('\r\n' instead of '\n')
• without the two initial lines and the skiprows parameter
• with two empty initial lines
• with just one initial line, containing text
• with the 'python' engine,

and everything worked.

Here is the output of pandas.show_versions():

INSTALLED VERSIONS
------------------
commit: None
python: 3.4.3.final.0
python-bits: 64
OS: Windows
OS-release: 7
machine: AMD64
processor: Intel64 Family 6 Model 61 Stepping 4, GenuineIntel
byteorder: little
LC_ALL: None
LANG: None

pandas: 0.15.2
nose: 1.3.4
Cython: 0.22
numpy: 1.9.2
scipy: 0.15.1
statsmodels: 0.6.1
IPython: 3.0.0
sphinx: 1.2.3
patsy: 0.3.0
dateutil: 2.4.2
pytz: 2015.4
bottleneck: None
tables: 3.1.1
numexpr: 2.3.1
matplotlib: 1.4.3
openpyxl: 1.8.5
xlrd: 0.9.3
xlwt: None
xlsxwriter: 0.6.7
lxml: 3.4.2
bs4: 4.3.2
html5lib: None
httplib2: None
apiclient: None
rpy2: None
sqlalchemy: 0.9.9
pymysql: None
psycopg2: None

[jreback]

works on master, might have been #5664 fixed in 0.16.1

In [1]: import pandas
In [2]: myfile = open("test.csv", "w", newline="\n")
In [3]: myfile.write("blah\n\ncol_1,col_2,col_3\n\n")
Out[3]: 25
In [4]: myfile.close()
In [5]: dat = pandas.read_csv("test.csv", skiprows=2, encoding="utf-8", engine="c")
In [6]: dat
Out[6]:
Empty DataFrame
Columns: [col_1, col_2, col_3]
Index: []

In [7]: pd.__version__
Out[7]: '0.16.2+601.g76a4d99

[jreback]

if you'd like to add this as a test case, would take a PR in any event.

[e-pet]

Sorry, I'd like to contribute a test case (and in fact just wrote one), but I don't have the time to set up the build process right now, which seems to be necessary for running tests. So I can't check whether my test actually fails on the version I'm using.

[jreback]

ok, contributing docs are here: http://pandas.pydata.org/pandas-docs/stable/contributing.html

its actually quite easy with conda

[sudheesh001]

Hey, Could I take a shot at this? Does the test for this have to go into io.parser.c_parser_only?
I can add a file with unix endings into the data directory and assign it to a variable in the setup_method of test_parsers along with the other csv files.
The contents of the file are as follows:

Test

col_1,col_2,col_3

Reading the file returns the columns with skiprows=2 as col_1, col_2, col_3. Should the assertion for the test here be the length being 3 and the contents?