API/BUG: Handling Dtype Coercions in Series/Index

[gfyoung]

Off master (bd169d):

# Case I: Overflow on int64
>>> Index([np.iinfo(np.uint64).max-1],dtype='int64')
...
OverflowError: Python int too large to convert to C long

# Case II: Coercion to uint64
>>> Index([-1], dtype='uint64')
UInt64Index([18446744073709551615], dtype='uint64')

# Case III: Ignoring coercion to int
>>> Index([1, 2, 3.5], dtype=int)
Float64Index([1.0, 2.0, 3.5], dtype='float64')

So we got some coercions that fail but others that work. Although all of these issues involve Index, the first two are also applicable to Series (in the last issue, it does coerce all elements to integer).

How should we handle failed coercions? Is the second case even a failure? Should iron out this.

xref #12758
xref #15187

[jreback]

I think our rules are pretty well established (FYI, we should actually document these).

If dtype=None (on Series/Index). then we find a best-fit dtype.
If dtype=..... Then we coerce to that dtype if its safe, otherwise should raise (a useful message)

Case I: Overflow on int64
Index([np.iinfo(np.uint64).max-1],dtype='int64')
...
OverflowError: Python int too large to convert to C long

I think this is ok (let's test it). Maybe a more user friendly message though.

Case II: Coercion to uint64
Index([-1], dtype='uint64')
UInt64Index([18446744073709551615], dtype='uint64')

This should raise

Case III: Ignoring coercion to int
Index([1, 2, 3.5], dtype=int)
Float64Index([1.0, 2.0, 3.5], dtype='float64')

This should raise

[ucals]

Now that we were able to close 1 novice and 2 intermediate bugs, I'd like to try an advanced bug - this one :)

[jreback]

sure!