Going to put it in, thanks. But the check still had a purpose: hiding the dtype=... for longlong and for long both when they have the same itemsize.
So to get that back working, as an enhancement, you would need to add it back as an itemsize check.

But the check still had a purpose: hiding the dtype=... for longlong and for long both when they have the same itemsize.

IIUC, long is an alias for int_ so is already covered. So then the missing case is when longlong has the same number of bytes as int_, is that right? So the check might look something like:

if np.longlong(1).itemsize == np.int_(1).itemsize:
    _typelessdata.append(np.longlong)

or have I misunderstood?

Int is long, so its either longlong or cint. Both paths that existed are necessary. One for windows (mainly) one for others (the longlong).