na_filter=False ignored when index_col set

[cancan101]

Given the following CSV file:

u1,u2,u3,d1,d2,d3,d4
Good Things,C,,1,1,1,1
Good Things,R,,1,1,1,1
Bad Things,C,,1,1,1,1
Bad Things,T,,1,1,1,1
Okay Things,N,B,1,1,1,1
Okay Things,N,D,1,1,1,1
Okay Things,B,,1,1,1,1
Okay Things,D,,1,1,1,1

First I parse with na_filter=True:

In [13]: pd.read_csv("/home/alex/nan_issue.csv", na_filter=True)
Out[13]: 
            u1 u2   u3  d1  d2  d3  d4
0  Good Things  C  NaN   1   1   1   1
1  Good Things  R  NaN   1   1   1   1
2   Bad Things  C  NaN   1   1   1   1
3   Bad Things  T  NaN   1   1   1   1
4  Okay Things  N    B   1   1   1   1
5  Okay Things  N    D   1   1   1   1
6  Okay Things  B  NaN   1   1   1   1
7  Okay Things  D  NaN   1   1   1   1

then I parse with na_filter=False:

In [12]: pd.read_csv("/home/alex/nan_issue.csv", na_filter=False)
Out[12]: 
            u1 u2 u3  d1  d2  d3  d4
0  Good Things  C      1   1   1   1
1  Good Things  R      1   1   1   1
2   Bad Things  C      1   1   1   1
3   Bad Things  T      1   1   1   1
4  Okay Things  N  B   1   1   1   1
5  Okay Things  N  D   1   1   1   1
6  Okay Things  B      1   1   1   1
7  Okay Things  D      1   1   1   1

then index_cols set:

In [11]: pd.read_csv("/home/alex/nan_issue.csv", na_filter=False,index_col=[0,1,2],)
Out[11]: 
                    d1  d2  d3  d4
u1          u2 u3                 
Good Things C  NaN   1   1   1   1
            R  NaN   1   1   1   1
Bad Things  C  NaN   1   1   1   1
            T  NaN   1   1   1   1
Okay Things N  B     1   1   1   1
               D     1   1   1   1
            B  NaN   1   1   1   1
            D  NaN   1   1   1   1

Finally setting na_values=[], keep_default_na=False seems to fix the issue:

In [14]: pd.read_csv("/home/alex/nan_issue.csv", na_filter=False,index_col=[0,1,2],na_values=[], keep_default_na=False)
Out[14]: 
                   d1  d2  d3  d4
u1          u2 u3                
Good Things C       1   1   1   1
            R       1   1   1   1
Bad Things  C       1   1   1   1
            T       1   1   1   1
Okay Things N  B    1   1   1   1
               D    1   1   1   1
            B       1   1   1   1
            D       1   1   1   1

[jreback]

There are very limited tests with na_filter=False and its a pretty silly parameter, so moving to low-priority. You are welcome to do a PR if you'd like.

[cancan101]

TBI I can work around for now. My observation is that there probably too many parameters on the method having to do with handling of nans.It would be great to clean this up.
keep_default_na, na_filter, na_values

There is also interaction between these parameters.

[jreback]

sure.....though to be honest, easiest just to drop na_filter...but if you come up with a better API, gr8