string truncation when using Series.astype(str)

[mariusvniekerk]

When converting a pandas Series object to type string using astype(str), long strings are truncated to 64 characters silently.

Pandas version: 0.12
Numpy version: 1.7.1

In [1]: s = '0123456789' * 10

In [2]: tmp = np.array([s]).astype(str)
         tmp
Out[2]: array([ '0123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789'], 
      dtype='|S100')

In [3]: tmp = pd.Series([s]).astype(str)
       tmp[0]
Out[3]: '0123456789012345678901234567890123456789012345678901234567890123'

In [4]: len(tmp[0])
Out[4]: 64

[cpcloud]

actually this looks like a numpy bug

In [1]: s = '0123456789' * 10

In [2]: tmp = np.array([s]).astype(str)

In [3]: s = Series([s])

In [4]: s.values
Out[4]: array([ '0123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789'], dtype=object)

In [5]: s.values.astype(str)
Out[5]:
array(['0123456789012345678901234567890123456789012345678901234567890123'],
      dtype='|S64')

[jreback]

and astype(str) should just be converted to astype(object) internally (as str type is not a valid type)

[cpcloud]

if you need to do this then you can do

In [29]: s.values.astype('|S%d' % len(s[0]))

[jreback]

str must implcity do S64 somewhere (in numpy)

[cpcloud]

maybe fixed in np master

numpy/numpy#3270

[cpcloud]

i don't have the patience right now to recompile everything...anyone want to try this out on numpy master? i guess we leave it open until that numpy release