Index out of bounds error raised for negative indices.

[da415]

Accessing out of bounds negative timeseries indices silently return nonsense values.

[jreback]

@da415 can you put an example of this change?

also just check that performance is unchanged
e.g. run test_perf.sh

[ghost]

Looks like the cython (no) bounds-checking monster has roared again.
#3028 , #3029

I get a segfault rather then junk data.

In [6]: s=mkdf(10,5,r_idx_type='dt').C_l0_g0
   ...: s
Out[6]: 
R0
2000-01-03    R0C0
2000-01-04    R1C0
2000-01-05    R2C0
2000-01-06    R3C0
2000-01-07    R4C0
2000-01-10    R5C0
2000-01-11    R6C0
2000-01-12    R7C0
2000-01-13    R8C0
2000-01-14    R9C0
Freq: B, Name: C_l0_g0, dtype: object

In [7]: s[0]
Out[7]: 'R0C0'

In [8]: s[-1]
Out[8]: 'R9C0'

In [9]: s[-11]
** segfault **

#3029 doesn't seem to catch this either, @stephenwlin?

[jreback]

I think this is very easy to fix by putting some minimal error checking in Series.getitem , and using indexing._maybe_convert_indicies (like how we are handling take which is user facing), I think the perf impact will be quite minimal then, rather than trying to catch this in cython (which is very general)

[jreback]

In fact, this is maybe an excuse to do some more unification on series/frame getitem/setitem, @stephenwlin made them nice and user friendly......prob not too hard to 'fix' series

[ghost]

sounds good.

[jreback]

not sure will have time to get out before 0.11...shall we push to 0.12? @wesm

[jreback]

@da415 I believe that your solution fails the tseries/test_resample test..can you confirm?

[da415]

@jreback Apologies, the earliest I can confirm this failure is Tuesday 02/04.

[jreback]

@da415 sorry...didn't mean that...I tried your solution and it fails test_resample (on your commit); this was for sure a problem all along! thanks for catching it. I was just trying to see if maybe I am seeing something different.

[wesm]

I fixed it directly in the segfaulting function and added test case. thanks