read_csv index_col ignores dtype if specified

[alzmcr]

Hi, I'm not sure if this is intended but when using the index_col parameter in read_csv it ignore the input format specified in dtype. It's reproducible as following using pandas 0.18.0 and numpy 1.11.0.

>>> from StringIO import StringIO
import pandas as pd
df_csv = """request_hour,request_date,size
03,2016-04-26,2580954.0
04,2016-04-26,12003662.0
05,2016-04-26,13042624.0
06,2016-04-26,2899309.0
07,2016-04-26,-1.0"""

>>> pd.read_csv(StringIO(df_csv), dtype={'request_hour': 'string'}).set_index('request_hour')

             request_date        size
request_hour                         
03             2016-04-26   2580954.0
04             2016-04-26  12003662.0
05             2016-04-26  13042624.0
06             2016-04-26   2899309.0
07             2016-04-26        -1.0
# This is what I would expected as output

>>> pd.read_csv(StringIO(df_csv), dtype={'request_hour': 'string'}, index_col=0)

             request_date        size
request_hour                         
3              2016-04-26   2580954.0
4              2016-04-26  12003662.0
5              2016-04-26  13042624.0
6              2016-04-26   2899309.0
7              2016-04-26        -1.0
# I'm surprise that the index has been converted to int

I couldn't find any specs on this anywhere, so I wonder if something with the read_csv or I'm doing something wrong.

Thanks!

[jreback]

dupe of #9435