Network Protocol (almost) Complete

Pettycoin relies on partial knowledge, but until today the code assumed we needed to know transaction in every shard in every block.

Today I implemented the sending and receiving of protocol_pkt_get_txmap and protocol_pkt_txmap packets. When a node discovers a block, it asks for the entire contents of every shard it is interested in using protocol_pkt_get_shard packets. But as transactions can cross shards, it will also be interested in some transactions which are not in its own shard: this is where protocol_pkt_get_txmap comes in. The protocol_pkt_txmap returned is a bitmap of whatever transactions you’re interested in in that shard of transactions: you then ask for them individually using protocol_pkt_get_tx_in_block which provides the transaction and proof that it’s at that position in the block.

I also implemented the receipt and checking of every kind of complaint packet: these prove a problem with a block, and need to be checked carefully to make sure they really do prove it. There are five kinds:

• protocol_pkt_complain_tx_misorder

  This proves that two transactions in a shard are in the wrong order. We insist that they always be ordered, mainly because it should speed transmission of blocks by allowing nodes to guess the contents.

• protocol_pkt_complain_tx_invalid

  This proves that a transaction in a block is invalid. There’s no code which generates such complaints today, since there’s no way to send an invalid transaction, but in future we could use this for a buggy miner who produced invalid transactions.

• protocol_pkt_complain_tx_bad_input

  This proves that a transaction in a block has a bad input (eg. trying to spend someone else’s transaction) by providing the transaction and the input it tried to use.

• protocol_pkt_complain_tx_bad_amount

  This proves that a transaction in a block doesn’t add up by providing the transaction and all the inputs it tried to use.

• protocol_pkt_complain_bad_input_ref

  This proves that a transaction in a block has a bad input reference. Input references are inserted by the miner after each transaction, and show where each input for that transaction should be (blocks ago, shard number and transaction offset in the shard). This proves that the transaction referred to by the reference isn’t the one the transaction wanted.

There are only two parts of the network protocol now missing. First is dealing with skiplist of blocks before the horizon (protocol_pkt_horizon) and the second is broadcast of double spends and complaining about them if they’re inside a block. [Wait, scratch that. I still have fee collection and to-gateway payments to implement as well]

But first, I took some time to write some more unit tests: I’m way behind, still.