+++ b/Lib/test/test_winconsoleio.py
+to real people with real keyborads.
Should be keyboards
There are still assert_() calls in this file (1602_6.patch). Did you miss them?

+++ b/Lib/io.py
+from _io import WindowsConsoleIO
+__all__.append('WindowsConsoleIO')
I think you should either document this class, or remove it from __all__ to clarify it is just an implementation detail.

+++ b/Modules/_io/winconsoleio.c
+_io_WindowsConsoleIO___init___impl
+    PyObject *decodedname = Py_None;
+    Py_INCREF(decodedname);
+    int d = PyUnicode_FSDecoder(nameobj, (void*)&decodedname);
Won’t this leak a reference to Py_None?
(Also, I think needless casting like in the last line can mask mistakes that the compiler would otherwise pick up. Imagine if you got the parameters around the wrong way.)

+read_console_w(HANDLE handle, DWORD maxlen, DWORD *readlen) {
+    /* If we didn't read a full buffer that time, don't try
+       again or we will block a second time. */
I’m not familiar with the Windows APIs involved, but this doesn’t seem robust. What if there were exactly one full buffer waiting, would the next call block without returning anything?