Encoding non-ascii characters in to_csv with encodings

[jseabold]

Right now if you want to use an encoding, you have to make sure that all fields that contain non-ascii characters are in that encoding. Maybe this is okay, but I'm working with a lot of data and it is a bit cumbersome for me to be doing these checks constantly. For example

from StringIO import StringIO
import pandas

df = pandas.read_table(StringIO('Ki\xc3\x9fwetter, Wolfgang;Ki\xc3\x9fwetter, Wolfgang'), sep=";", header=None)
df["X.1"] = df["X.1"].apply(lambda x : x.decode('utf-8'))
df.to_csv("blah.csv", encoding="utf-8")

The question is, is this something that the user should be worrying about or is there a "safe_encode" that could be used instead, similar idea to #1804?

[jseabold]

Just ran into this again. I'm starting to have the feeling that the I/O operations should do the conversion. For example, I want to read in a utf-8 file and then convert it to latin-1. It would be great if I could do

read_csv(..., 'utf-8')
to_csv(..., 'latin-1')

but currently you have to go through the intermediate apply step for each string-like column. Might work on a PR for this if I can get some time this afternoon.

[wesm]

Note you can use x.str.decode('utf-8') now. Not that that helps too much

After a week of digging through the unicode nether-regions of the tree, I support jseabold's
comment on strictly enforcing unicode internally and encoding/decoding at I/O
points only.

Otherwise, you end up having brittle assumptions about encodings and if clauses handling corner-cases
all over the codebase.

I apologize for all these useless messages, github is doing unexpected things, and I can't
figure out how to purge these. hopefully they'll go away when I tear down the PR branches.

I'll be more cautious in the future.

[wesm]

Passing encoding='utf-8' yields unicode strings now. Closing this issue

In [4]: df = pandas.read_table(StringIO('Ki\xc3\x9fwetter, Wolfgang;Ki\xc3\x9fwetter, Wolfgang'), sep=";", header=None, encoding='utf-8')

In [5]: df
Out[5]: 
                    X0                   X1
0  Kißwetter, Wolfgang  Kißwetter, Wolfgang

In [6]: df['X0']
Out[6]: 
0    Kißwetter, Wolfgang
Name: X0

In [7]: df['X0'][0]
Out[7]: u'Ki\xdfwetter, Wolfgang'