On Monday 2010-08-16 19:55 Matthew Toseland wrote: > The Summer of Code firm pencils down deadline is about to expire. The > below is concerned with the question "what must be changed to get > this mergeable"...
> NewPacketFormat.tryDecipherPacket: > - Should it be updated on every packet? Obviously it should never go > backwards... Why shouldn't it be updated on every packet (every time highestReceivedSeqNum changes that is)? It only updates the entries that are outside the window, so assuming packets are received in order we update 1 entry per packet. > 2e33916fee9106b0e985f1bf974ed3d8fda40f42 > - Any particular reason for always starting from 0 on new format? Both sides need to know the number because of the watchlist, so it doesn't have to be 0, but it has to be generated using something other than random.nextInt(). > 501bb6dfe50a6e54bd0c19d867c1136eed37128c > - Not sure this is a good idea, old format is limited to 512 packets > in flight (or was it 256?) and has lots of retransmit stuff etc you > don't need. > - This has the advantage of keeping the would block handling, but on > the other hand it has a nasty packet window limit, lots of code > related to old format retransmission code (which is bad); the > retransmission logic e.g. is completely inappropriate. I forgot that the window was in PacketTracker, so I'll probably revert this one and wrap the sequence numbers instead of using them per key. Other things that need to be fixed: Support for padding The easiest way of doing this might be to use the flag combination that is currently illegal (not fragmented and not first packet) to indicate that the rest of the packet is padding. Longer HMAC Currently the HMAC length is set to 4 bytes which is too short, and having a variable length has also been mentioned. Busy-loop in PacketSender/maybeSendPacket() -- Martin Nyhus -------------- next part -------------- A non-text attachment was scrubbed... Name: signature.asc Type: application/pgp-signature Size: 198 bytes Desc: not available URL: <https://emu.freenetproject.org/pipermail/devl/attachments/20100819/37f9342c/attachment.pgp>
