read_table/csv unexpected type dependence on delimiter

[dsm054]

While answering a question on SO I came across something which puzzled me:

>>> import pandas as pd
>>> pd.__version__
'0.10.0b1'
>>> repr(open('cusip.txt').read())
"'65248E10 11\\n55555E55 22\\n'"
>>> !cat cusip.txt
65248E10 11
55555E55 22
>>> df = pd.read_table("cusip.txt", header=None, sep=" ")
>>> df
              0   1
0  6.524800e+14  11
1  5.555500e+59  22
>>> type(df[0][0])
<type 'numpy.float64'>
>>> df = pd.read_table("cusip.txt", header=None, sep=r"\s+")
>>> df
                     0   1
0      652480000000000  11
1 -9223372036854775808  22
>>> type(df[0][0])
<type 'numpy.int64'>

Changing the delimiter from " " to r"\s+" somehow triggered the interpretation of the first column as integers instead of floats.

[wesm]

Yeah the other code path does not detect overflows. I haven't gotten around to adding regular expression capability to the new parser engine (it won't be toooo hard)

[wesm]

Actually no that's weird. I'll take a close look now