Using asserts for type hinting

In PyCharm 2016.2, If there was some code like `df = pandas.concat(df_list)`, you could use `assert isinstance(df, pandas.DataFrame)` to do both runtime type checking and type hinting for PyCharm, but in 2016.3 you have to do `# type: pandas.DataFrame` for type hinting for that line of code.

So in order to do both runtime checking and type hinting you would have to do:

df = pandas.concat(df_list)  # type: pandas.DataFrame
assert isinstance(df, pandas.DataFrame)

vs just

df = pandas.concat(df_list)
assert isinstance(df, pandas.DataFrame)

which was possible in the previous version.

Is there a reason that this was removed? Or is there a way that this can be turned on as a default option?, since this would force me to add a lot of unnecessary code, or remove type hinting. This is especially problematic for pandas related code, since a lot of the functions work on multiple types so the return types are often ambiguous.

type hinting also doesn't work anymore for cases like:

if not isinstance(obj, str):
    raise TypeError

as well.

Edit:

Apparently this problem only occurs when a value has a type hint from a function, but later is trying to be overridden by an assert.

For example in:

df = pandas.concat(df_list)
assert isinstance(df, pandas.DataFrame)

pandas.concat has a return type hint of `type`, so df has the inferred type `type` from the function, but after the assert, instead of changing to inferred type `DataFrame` df stays as inferred type `type` which was originally assigned from the function.

In cases like:

version = payload['version']
assert isinstance(version, int)

then version has the correct inferred type after the assert, since payload which is a dict, returns an inferred type `Any` due to not having any annotations.

Could there be some clarification about how the new type hinting system works? or whether there are plans to leave it as is or to change it in the future, since it has changed since the previous version?