API: inference of uint64s are wrong

[jreback]

In [1]: pd.api.types.infer_dtype([2*63, -1])
Out[1]: 'integer'

In [3]: pd.api.types.infer_dtype([2*63])
Out[3]: 'integer'

[1] should be 'mixed' (inference which means object)
[2] should be 'uinteger'

note this will probably break some things in other parts of the suite as we sometimes infer to figure out the dtype and act on it, so 'uinteger' would then be unhandled.

e.g. in Index creation and https://github.com/pandas-dev/pandas/pull/16108/files
xref https://github.com/pandas-dev/pandas/pull/16295/files

[jreback]

cc @gfyoung

[gfyoung]

@jreback : This is a deliberate choice to consider integer to include ALL integer data types (more like a Python int in this sense). Afterwards, we handled uint by checking if all were >= 0. So wouldn't necessarily call this a BUG (more like an API change?)

[jreback]

yes this is more of a change. I think we should differentiate these.