[Bitcoin-development] Bitcoin TX fill-or-kill deterministic behavior

Mike Hearn mike at plan99.net
Fri Apr 13 10:04:07 UTC 2012

It sounds OK as long as you exclude nLockTimed transactions.

That said, if you broadcast a transaction that does not meet the fee
rules, you should be able to notice that it wasn't accepted by your
peers immediately. Today it's painful because the protocol isn't very
chatty - in bitcoinj I plan to do this by announcing to half the
connected peers and waiting to see if the transaction comes back on
the other half. Getting a response from a peer that the TX was dropped
for reasons {x,y,z} is a better design but needs another protocol

So having transactions expire would address the case where somebody
broadcasts a transaction that successfully propagates across the
network, but then isn't actually accepted by miners for some reason.
For instance due to a change in the default fee schedules. That risk
can be mitigated somewhat by being careful about such changes (timed
phase ins set multiple months out so people have time to upgrade,
alerts announcing it, etc).

I'm not sure we should be encouraging users to attach fees to
transactions though. Even if you can replace a transaction after a
couple of days, the user experience of trying to get the fee "right"
is atrocious. I don't think any sensible merchant will actually be
willing to put their customers through this nonsense. If somebody
broadcasts a transaction that successfully propagates across a big
chunk of the network but then gets stuck due to lacking sufficient
fees, the best fix is for the merchant to broadcast another
transaction that spends the first and increases the fees on it that
way. After this happens a few times, if I was a merchant I'd be
tempted to just ask buyers to submit the TX to me directly and I'll
handle keeping up with what miners currently charge and attaching
fees. I don't want my customers to have to think about this and have
trades spuriously fail when they forget.

That design requires a minor change to how fees are calculated inside
the memory pool, to include fees on un-included dependencies. But that
seems fairly uncontroversial to me. It's best for users, merchants and
miners to not leave chains of transactions in limbo when together
their fees add up to the minimum required amount.

More information about the bitcoin-dev mailing list