BUG: read_json -> 'DataFrame' object has no attribute 'dtype'

[Komnomnomnom]

Came across this when creating #4376. Not sure what's going on as I was under the impression that DataFrame always has a dtype attribute.

The error only seems to occur when the columns are not unique:

In [5]: import pandas as pd

In [6]: df = pd.DataFrame([['a','b'],['c','d']], index=[1,2], columns=['x','y'])

In [7]: df
Out[7]: 
   x  y
1  a  b
2  c  d

In [8]: pd.read_json(df.to_json(orient='split'), orient='split')
Out[8]: 
   x  y
1  a  b
2  c  d

In [9]: df = pd.DataFrame([['a','b'],['c','d']], index=[1,2], columns=['x','x'])

In [10]: df
Out[10]: 
   x  x
1  a  b
2  c  d

In [11]: pd.read_json(df.to_json(orient='split'), orient='split')
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-11-aaa9a429e682> in <module>()
----> 1 pd.read_json(df.to_json(orient='split'), orient='split')

/home/kieran/work/git/pandas_kom/pandas/io/json.pyc in read_json(path_or_buf, orient, typ, dtype, convert_axes, convert_dates, keep_default_dates, numpy, precise_float)
    196     obj = None
    197     if typ == 'frame':
--> 198         obj = FrameParser(json, orient, dtype, convert_axes, convert_dates, keep_default_dates, numpy).parse()
    199 
    200     if typ == 'series' or obj is None:

/home/kieran/work/git/pandas_kom/pandas/io/json.pyc in parse(self)
    241         if self.convert_axes:
    242             self._convert_axes()
--> 243         self._try_convert_types()
    244         return self.obj
    245 

/home/kieran/work/git/pandas_kom/pandas/io/json.pyc in _try_convert_types(self)
    453             self._try_convert_dates()
    454         for col in self.obj.columns:
--> 455             new_data, result = self._try_convert_data(col, self.obj[col], convert_dates=False)
    456             if result:
    457                 self.obj[col] = new_data

/home/kieran/work/git/pandas_kom/pandas/io/json.pyc in _try_convert_data(self, name, data, use_dtypes, convert_dates)
    282         result = False
    283 
--> 284         if data.dtype == 'object':
    285 
    286             # try float

/home/kieran/work/git/pandas_kom/pandas/core/frame.pyc in __getattr__(self, name)
   2086             return self[name]
   2087         raise AttributeError("'%s' object has no attribute '%s'" %
-> 2088                              (type(self).__name__, name))
   2089 
   2090     def __setattr__(self, name, value):

AttributeError: 'DataFrame' object has no attribute 'dtype'

[jreback]

it is expecting the object to be a series at that point....let me look

[jreback]

actually this is a problem with iterating over the duplicate columns that are read back...easy fix

[Komnomnomnom]

Nice, thanks @jreback!

[jreback]

turns out I discovered another bug with this...but all fixed up now...thanks for your help!