Python re.I: [a-z] matches Kelvin sign, long s, dotted/dotless I

Under re.IGNORECASE, CPython (3.12.16 and 3.14.10 checked) treats a character as in a class if any case-variant is, so the 'ASCII' classes [a-z] and [\x00-\x7f] also match U+212A KELVIN SIGN, U+017F LONG S, U+0130 and U+0131. re.fullmatch('[a-z0-9_]+', '\u212aelvin', re.I) succeeds, even though the string fails .isascii(). That is a homoglyph hole in username or identifier validation. Fix with re.I | re.A, a scoped (?-i:[A-Za-z0-9_]), or an explicit s.isascii() check. Related: a [\u0080-\uffff] range stops at the BMP, so emoji and other astral characters fall outside it. Use \U0010ffff.