BUG?: should ComplexBlock._can_hold_element/should_store check itemsize?

[jbrockmendel]

arr = np.arange(5).astype("c8")
ser = pd.Series(arr)
val = np.finfo(np.float64).max
val = val.astype("c16")

>>> val
(1.7976931348623157e+308+0j)

ser[0] = val

>>> ser
0             inf+0.000000e+00j
1    1.000000e+00+0.000000e+00j
2    2.000000e+00+0.000000e+00j
3    3.000000e+00+0.000000e+00j
4    4.000000e+00+0.000000e+00j
dtype: complex64

For IntBlock we check the itemsize to avoid overflows, but for FloatBlock and ComplexBlock we do not, so we get the inf here. Should we instead be upcasting?

[jbrockmendel]

@jorisvandenbossche when convenient, i'd like to get your thoughts on this, #33469, and #33830

[jorisvandenbossche]

For this issue, can you show a similar example using int?

[jbrockmendel]

Welp I was wrong about IntegerBlock avoiding overflows:

arr = np.arange(5).astype("i4")
ser = pd.Series(arr)
val = np.int64(np.iinfo(np.int64).max)

ser[0] = val

>>> ser
0   -1
1    1
2    2
3    3
4    4
dtype: int32