DOC: read_clipboard() failed for sphinx in dev-99ae9ce

[michaelaye]

See here:
http://pandas.pydata.org/pandas-docs/dev/io.html#clipboard

and scroll to the clipboard section.
Line [1215] with read_clipboard() failed.

[jreback]

@changhiskhan can you take a look?

[hayd]

probable/possible cause #3848 ?

[jreback]

could possibly be numpy 1.6.2 related

that's what's building docs

and don't think we have any tests for the clipboard stuff

shall we add some?

[cpcloud]

should there be a raise there to make sure that TextParser gets non-empty input?

[jreback]

this should work, non empty is not an error per se
something else going on

[cpcloud]

@jreback should this pass?

class TestClipboard(unittest.TestCase):
    def setUp(self):
        self.data = DataFrame(randn(5, 3))

    def test_round_trip(self):
        self.data.to_clipboard()
        expected = self.data
        result = read_clipboard()
        tm.assert_frame_equal(expected, result)

currently fails b/c columns are read as strings not ints

also should probably not print out the name of the clipboard program

[jreback]

in linux world, works for me

In [1]: df = DataFrame(randn(10,2))

In [2]: df.to_clipboard()
/usr/bin/xclip

In [3]: pd.read_clipboard()
Out[3]: 
          0         1
0  0.086303  0.412240
1 -0.574764 -0.131086
2  0.633972  0.012292
3 -1.034044 -0.115232
4 -0.709054  0.040310
5 -0.307468  0.312964
6  0.673137 -0.700172
7  1.598500 -0.409174
8 -0.511669  0.123703
9  0.010648 -0.205798

In [4]: pd.read_clipboard().dtypes
Out[4]: 
0    float64
1    float64
dtype: object

[cpcloud]

what is the dtype of the column index, i get object bc they are strings, should be int

[jreback]

I think this is expected though; things are transmitted as 'text', and they I wouldn't expect the a numeric looking column to be necessarily converted

just test with ``columns=list('AB')`

In [1]: df = DataFrame(randn(10,2))

In [2]: df.columns.dtype
Out[2]: dtype('int64')

In [3]: df.to_clipboard()
/usr/bin/xclip

In [4]: pd.read_clipboard().dtypes
Out[4]: 
0    float64
1    float64
dtype: object

In [6]: pd.read_clipboard().columns.dtype
Out[6]: dtype('O')

[cpcloud]

ok...the index is converted though....

[jreback]

I don't think we convert columns by default.....even with csv...not sure though

[cpcloud]

ur right read_csv doesn't convert, my bad

[cpcloud]

good news: looks like i managed to repro something similar

https://travis-ci.org/cpcloud/pandas/builds/9213347

bad news: i'm not sure how to work around this

[cpcloud]

is there a way to trick X into thinking there's a display

[cpcloud]

this: http://en.wikipedia.org/wiki/Xvfb might work

[cpcloud]

if docs are being built on a headless machine you'll get this error, i'm not sure how the clipboard works in conjunction with X, i'll investigate later this evening

[cpcloud]

so this is not a numpy version issue, just wanted to point that out...@jreback should i mention this on the ML?

[jreback]

is it because its built headless or some other issue? go ahead and email what u know and solution

[cpcloud]

I think it's the headless issue. The clipboard requires an X server to be running and the DISPLAY variable to be set. Not sure if there are any more details. There might be some specific platform issues, but the travis docs solution works.

[cpcloud]

would i be heretical in suggesting that we move the documentation over to somewhere like http://readthedocs.org?

[cpcloud]

It would be much easier to fix doc issues

[wesm]

looking good now