We noticed a regression in Python3.10.0a7 (it works properly in Python3.10.0a6) when running Django's test suite. `convert_date()`[1] is called and crashes when fetching an empty value from `date` column on SQLite:

File "python3.10/lib/sqlite3/dbapi2.py", line 64, in convert_date
    return datetime.date(*map(int, val.split(b"-")))
ValueError: invalid literal for int() with base 10: b''

I will bisect a regression and submit PR.

[1] https://github.com/python/cpython/blob/50616223d1043f0f83534ffec38709439b8a49ab/Lib/sqlite3/dbapi2.py#L63-L64

A regression test:

    def test_convert_null_date(self):
        con = sqlite3.connect(":memory:", detect_types=sqlite3.PARSE_DECLTYPES)
        cur = con.cursor()
        cur.execute("CREATE TABLE t (x DATE NULL)")
        cur.execute("INSERT INTO t (x) VALUES (NULL)")
        cur.execute("SELECT * FROM t")
        values = [x[0] for x in cur.fetchall()]
        self.assertEqual(values, [None])

Thanks for the report and the test case! :)

Regression was introduced by e161ec5dd7ba9355eb06757b9304019ac53cdf69, where it was assumed that the handling of zero-sized blobs should be consistent. Unfortunately, the handling of zero-sized blobs is not consistent. Zero-sized blobs _with_ converters must return None; zero-sized blobs _without_ converters must return b"".

BTW, you might be interested in this bpo, Mariusz: bpo-43553

> Zero-sized blobs _with_ converters must return None; zero-sized blobs _without_ converters must return b"".

Maybe the docs should be updated to reflect this inconsistency.

New changeset 6f1e8ccffa5b1272a36a35405d3c4e4bbba0c082 by Erlend Egeberg Aasland in branch 'master':
bpo-43752: Fix sqlite3 regression for zero-sized blobs with converters (GH-25228)
https://github.com/python/cpython/commit/6f1e8ccffa5b1272a36a35405d3c4e4bbba0c082

Thank you for the report and reproducer, Mariusz and thank you for the PR, Erlend.