[Bitcoin-development] Embedded consensus system upgrade procedures
pete at petertodd.org
Sun Feb 9 17:12:14 UTC 2014
We have an embedded consensus system and we want to be able to upgrade
it with new rules. There inevitably will be a transition period where
some users use clients that interpret the new rules, while others only
interpret the old rules. Since we only rely on the host consensus system
for timestamped proof-of-publication the the miner-vote soft-fork
upgrade mechanism;(1) there are no validating miners in the system to
whome trust can be outsourced.
We have a problem: messages encoding actions, such as moving as asset
from one owner to another, can be published on the the blockchain
according to new and old rules simultaneously, double-spending the
asset. Potentially a user with the old v1 software may be tricked into
accepting an asset when the consensus of the v2 software is that the
asset has already been spent, and the v1-visible transaction is invalid.
Split actions into a separate "decrement" and "increment" operations,
and ensure that v1 software can see the "decrement" of a balance, spend
of a transaction output etc. even if it does not see the corresponding
increment operation. This solves the double-spend problem and ensures v1
users can't be ripped off. With obvious analogy to the PoW case, we will
refer to this general principle as a embedded consensus system
Note how with the Colored Coins technology this principle happens
implicitly and with miner validation: colored coins are valid
transaction outputs known to the host consensus system and moving them
from one owner to another is guaranteed to result in the desctruction of
the colored coin from the point of view of any older software version.
Older software that does not support the newer colored coin kernel
specified by the new asset definition will simply see the respective
coins be destroyed in invalid transactions. Note how this implies that
asset definitions created by issuers should be careful to ensure that
kernels chosen should be designed such that the actioned specified by
one kernel can-not be interpreted differently by another; kernels should
be clearly incompatible with each other.
Mastercoin is a balance-based system where transactions increment and
decrement balances. Being balance-based, and lacking pruning, an even
simplier "scorched earth" approach will be used where each address is
associated with a maximum version number seen by transactions signed by
the address. Addresses with a max version number higher than what the
software understands are considered to be null and have no value of any
kind. (counterparty would be wise to do the same)
Implementations should record in their databases the blockhash
associated with transactions that were not recognized yet affected the
state of the consensus. For instance a colored coin implementation
should record the blockhash and transaction ID where a given coin was
destroyed in an invalid transaction; after upgrading these "last
transaction understood" markers can be used to replay blockchain data to
arrive at the new consensus.
Similarly in the case of the Mastercoin system balances associated with
addresses that have been frozen should be still allowed to increment so
that replaying blockchain data from the last recognized transaction
arrives at a upgraded consensus.
As an aside, any embedded consensus system would be wise to have a way
of generating a master digest representing the state of the consensus in
the database. The Bitcoin Core gettxoutsetinfo command is a good model,
which provides hash_serialized, a digest representing the entire UTXO
set. In all systems this is useful for ensuring that different
implementations and instances have in fact arrived at a consensus.
1) BIP-16, Pay to Script Hash,
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 685 bytes
Desc: Digital signature
More information about the bitcoin-dev