Connection.text_factory can be used to control what objects are
returned for the TEXT data type. An excerpt from the docs:

    For efficiency reasons, there’s also a way to return str
    objects only for non-ASCII data, and bytes otherwise. To
    activate it, set this attribute to sqlite3.OptimizedUnicode.

However, it always returns Unicode strings now. There's even a
test for this feature which is obviously wrong:

    def CheckOptimizedUnicode(self):
        self.con.text_factory = sqlite.OptimizedUnicode
        austria = "Österreich"
        germany = "Deutchland"
        a_row = self.con.execute("select ?", (austria,)).fetchone()
        d_row = self.con.execute("select ?", (germany,)).fetchone()
        self.assertTrue(type(a_row[0]) == str, "type of non-ASCII row must be str")
        self.assertTrue(type(d_row[0]) == str, "type of ASCII-only row must be str")

It checks for str in both cases even though it should test for
bytes in the latter case.

---

The user can get bytes if he wants to by saying so explicitly.
Having the library mix bytes and unicode by itself makes it
harder for the user. Furthermore, I don't really buy
the "efficiency" reason here, so I'd vote for removing the whole
OptimizeUnicode thing. It has never worked for Py3k so it would
be safe.