• Python 3.8 and 3.9 are no longer supported, and Python 3.14 and 3.15 are added to testing.
• The original field of the returned validated email object incorrectly un-escaped the local part of the email address. Now it returns the original address, without any display name, as passed to validate_email.
• With allow_quoted_local=True, a quoted local part containing U+037E (Greek question mark), which Unicode NFC normalization changes to a semicolon, was returned without its quotes in the normalized address (e.g. a;b@example.com), getting around the re-validation added in 2.2.0. The normalized local part is now quoted according to its normalized form, giving "a;b"@example.com.
• When parsing display names, multiple closing angle brackets at the end of the email address is now rejected.
• The IPv6 tag of a domain literal is now case-insensitive, as in the RFCs, so me@[ipv6:::1] is accepted (and normalized to me@[IPv6:::1]) when allow_domain_literal=True.
• IPv6 domain literals with a zone index are now rejected.
• Internationalized domains with a right-to-left label are now rejected if any other label breaks the IDNA Bidi Rule (RFC 5893), such as a label starting with a digit. Such domains, like 1.xn--4dbc.com and xn--4dbc.1.com (1.אב.com and אב.1.com in Unicode), could be displayed identically.
• Faster validation of domains without Punycode (xn--) labels.
• Internationalized domains longer than 253 characters are now rejected for their length before IDNA encoding, which could take time quadratic in the length of the input.
• The command line tool now properly reads boolean environment variables and the DEFAULT_TIMEOUT environment variable.
• Improvements in type hinting, code comments, and tests.