API: silent overflow in Series(bigints, dtype="int8")

[jbrockmendel]

• closes #xxxx
• tests added / passed
• Ensure all linting tests pass, see here for how to run them
• whatsnew entry

This is made complicated by a) #40110 and b) MaskedArray behavior, which i need to check but may be changeable.

[jorisvandenbossche]

So the current behaviour is to wrap around, which follows numpy:

In [5]: pd.Series([1, 1000], dtype="int8")
Out[5]: 
0     1
1   -24
dtype: int8

In [6]: np.array([1, 1000], dtype="int8")
Out[6]: array([  1, -24], dtype=int8)

Personally, I think it would be good to be more strict (and only allow safe casting by default) in the constructors, as you are doing in this PR.
But I think this could (relatively easily) be done with a deprecation warning that it will raise in the future?

[jbrockmendel]

But I think this could (relatively easily) be done with a deprecation warning that it will raise in the future?

It could, but leaving it until 2.0 will block most remaining simplification of the construction/casting code, which is bound to uncover more bugs.

We do explicitly catch OverflowError when passing negative integers and specifying a uint dtype, so there's a precedent for not matching numpy's behavior.

[jbrockmendel]

mothballing to clear the queue