to_html vertically expands multiindex cells if there are empty strings

[al-yisun]

Notice in the html below the second index column's 'a' field is given as <th rowspan="2" valign="top">a</th>, expanding over what should be an empty cell.

I actually found this using a pivot table, margins=True, as this creates a row keyed like ('All', '', '').

print df = pd.DataFrame({'c1': ['a', 'b'], 'c2': ['a', ''], 'data': [1, 2]}).set_index(['c1', 'c2'])
       data
c1 c2      
a  a      1
b         2

print df.to_html()
<table border="1" class="dataframe">
  <thead>
    <tr style="text-align: right;">
      <th></th>
      <th></th>
      <th>data</th>
    </tr>
    <tr>
      <th>c1</th>
      <th>c2</th>
      <th></th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <th>a</th>
      <th rowspan="2" valign="top">a</th>
      <td> 1</td>
    </tr>
    <tr>
      <th>b</th>
      <td> 2</td>
    </tr>
  </tbody>
</table>

[cpcloud]

@y-p @jreback r u sure this a bug? _get_level_lengths on line 742 of pandas/core/format.py explicitly tests for the empty string. Getting rid of that check causes tests to fail, and looking at the tests it looks like this kind of output is expected unless the tests are copy and pasted output of what is expected in which case it probably is a bug.

[al-yisun]

> pd.DataFrame({'c1': ['a', 'b'], 'c2': ['a', ''], 'data': [1, 2]}).set_index(['c1', 'c2'])

This table is definitely incorrect.

[cpcloud]

@Adeodatus I asked because the code seems to indicate that an empty string in a MultiIndex means "repeat the previous value", so I'm not sure I would say it's definitely incorrect. Seems more like a design choice issue. Of course, I could be completely wrong.

[ghost]

Empty string labels are a mild abuse IMO. But this does work:

import pandas as pd
df = pd.DataFrame({'c1': ['', 'b'], 'c2': ['a', 'c'], 'data': [1, 2]}).set_index(['c1', 'c2'])
df.ix['']

so It's a bug. Using '' as a sentinel value was a way to reuse the existing formatting code.

fixed by bb4b640

[point6013]

> pd.DataFrame({'c1': ['a', 'b'], 'c2': ['a', ''], 'data': [1, 2]}).set_index(['c1', 'c2'])

This table is definitely incorrect.

in this case; how to make data down to aglin with the C1 and C2