ALPN values are not limited to ASCII

[msoxzw]

According to ALPN Identification Sequence, Python should accordingly not confine ALPN values to ASCII strings in

cpython/Lib/ssl.py

Lines 565 to 575 in 1c8b3b5

	def set_alpn_protocols(self, alpn_protocols):
	protos = bytearray()
	for protocol in alpn_protocols:
	b = bytes(protocol, 'ascii')
	if len(b) == 0 or len(b) > 255:
	raise SSLError('ALPN protocols must be 1 to 255 in length')
	protos.append(len(b))
	protos.extend(b)
	self._set_alpn_protocols(protos)

and

cpython/Modules/_ssl.c

Lines 2064 to 2081 in 1c8b3b5

	/*[clinic input]
	_ssl._SSLSocket.selected_alpn_protocol
	[clinic start generated code]*/
	static PyObject *
	_ssl__SSLSocket_selected_alpn_protocol_impl(PySSLSocket *self)
	/*[clinic end generated code: output=ec33688b303d250f input=442de30e35bc2913]*/
	{
	const unsigned char *out;
	unsigned int outlen;
	SSL_get0_alpn_selected(self->ssl, &out, &outlen);
	if (out == NULL)
	Py_RETURN_NONE;
	return PyUnicode_FromStringAndSize((char *)out, outlen);
	}

The OpenJDK proposed to use "latin-1" encoding instead of ASCII. However, @jlaine views this workaround as a hack and suggests a more agreeable and maintainable solution aiortc/aioquic#275 (comment) .

Therefore, which is the more ideal solution, or another better fix?

[ronaldoussoren]

According to RFC7301 the alpn identifier is a sequence of octets without a specified encoding. Most registrations are currently ASCII strings, except for some GREASE values.