Simplifying Generation

Pettycoin has a simple “mining” program called generate. The main program feeds it transaction positions and hashes on stdin, and it outputs a block (if it finds one) on stdout.

There’s a race here: the generate program might find a block before it processes the new transaction(s) we just sent. It handled this through an ugly system where pettycoin would feed it a cookie (actually a pointer to a pending transaction) which gets returned with the solution. However, if a pending transaction was invalidated, it would presumably get freed and thus generate would hand us a stale pointer.

This doesn’t happen at the moment since we insist on complete knowledge of all transactions on a chain to start generating. But in general it could, and fixing it would be ugly.

However, the new network protocol already has a packet to say “here’s a block header” and “here are the hashes of all transactions and input refrences in a shard”. All I had to do is change generate to send those packets, and wire them in. Much code removed.

While moving some code around, I noticed a stupid bug in my “divide difficulty by 4” routine. This is used to solve the problem that we see a new block, but don’t recognize the previous block hash. If the block took a significant amount of work to generate, we should try to find it. Otherwise it may be random spam, so ignore it.

So I wrote a difficulty test case, which not only allowed me to fix that bug, but turned up an unrelated in block_ancestor, which went back to then count’th block, not back count blocks as it advertised.

Testing is good. Next week, I will write a test a day, while trying to push forward and complete the handling of every packet in the protocol.