unicode encode error on index joins

[cpcloud]

The following throws a UnicodeDecodeError.

In [8]: df = mkdf(10, 10, data_gen_f=lambda r,c:randn(), r_idx_type='dt', c_idx_type='u')

In [9]: s = Series(randn(5,), df.index[:5])

In [10]: s.index.join(df.columns, how='outer').join(s.index)

[cpcloud]

admittedly this is probably not the most common use case having mixed unicode and datetime indices

[jreback]

problem is you are mixing strings and Unicode
so that error is legit

[cpcloud]

where are the strings coming in? datetimes are their own objecct...should raise a typeerror according to tslib.pyx

[jreback]

when u join different dtypes they are casted to object
which makes them strings from the datetimes (Unicode stays that way though)

[cpcloud]

they are timestamp objects though...

[jreback]

I think the index joining ultimately hashes strings/ Unicode or something

[cpcloud]

hm i will change to error reporting then and see what i can do about it.

[jreback]

isn't this related to #3878 (though resolution is to raise I think)

[cpcloud]

sort of but slightly different problem

[cpcloud]

the other is in python space, this is raising all the way down in the base class of Timestamp

[cpcloud]

@jreback is there a sane way to debug cython? i feel like i'm using php right now

[jreback]

print statements! or gdb I think

[cpcloud]

nevermind found the docs

[cpcloud]

this is insane i'm blowing the stack by calling isinstance(other, unicode) in _Timestamp...what?!

[cpcloud]

there's a cycle in the type graph somewhere...

[cpcloud]

and it's not object

[cpcloud]

it's somewhere around periodindex construction...tracking it down now..