On Tue, Sep 15, 2020, at 06:36, Dmitri Tikhonov wrote: > Assuming that mandating the 4-byte encoding is what's intended. Or is it > rather a "do not truncate" rule (which is also Christian's reading)?
It's "don't truncate". That is, don't send a packet number using an encoding that would drop a bit that is set. If you do, the other side (probably [1]) won't be able to read it. You can achieve that by sending longer encodings, but you probably [1] don't have to. Needing 4 bytes requires conditions I can't really conceive of. [1] It is in theory possible to have the other side receive a significant number of packets and for those packets to be unacknowledged. But that would require at least 256 packets (plus gaps) to be sent and not acknowledged. Thus, this is probably mostly academic, even for those implementations that don't set the packet number to zero for each new packet number space. I've never seen an interop runner test, even at its ridiculous loss rates, reach nearly that many packets.
