Received-SPF: pass (sog-mx-4.v43.ch3.sourceforge.com: domain of gmail.com
	designates 209.85.216.47 as permitted sender)
	client-ip=209.85.216.47; envelope-from=gmaxwell@gmail.com;
	helo=mail-qw0-f47.google.com; 
MIME-Version: 1.0
Date: Mon, 3 Oct 2011 00:53:51 -0400
Message-ID: <CAAS2fgRSw8ry7wjL5Fao1U+Ps0-6hfDcme_V_OCSkKaRTm2r9w@mail.gmail.com>
From: Gregory Maxwell <gmaxwell@gmail.com>
To: Bitcoin Development <bitcoin-development@lists.sourceforge.net>
Content-Type: text/plain; charset=UTF-8
Subject: [Bitcoin-development] Supermajority mining votes for valid->invalid
	changes.
Precedence: list

It is possible to made changes to the distributed algorithm which make
previously valid txn invalid without necessarily creating any lasting
chain splits.  This has been proposed for the addition of the eval
opcode by using one of the existing NOPs.

One challenge is that if transactions are emitted which are invalid
under the new scheme but valid under the old after the block height
that the rule is coded to take effect and a super-majority of miners
are not yet upgraded the upgrade may cause a long reorganization and
serious disruption.

Here I explain one possible way of avoiding this.

Upgraded nodes get the following rules:
(0) Never forward or mine a txn which would be invalid under the new rule.
(1) Apply old behavior before height X unconditionally.
    (X set far enough in the future to get reasonable deployment by
large miners)
(2) Begin applying the new rule only after the first point in the chain
    after X when none of the last Y blocks have contained an invalid transaction
    under the new rules.

After the software has been released members of the bitcoin community then
begin _intentionally_ transmitting transactions which are invalid under
the new rules. (What would have been an attack under simplest deployment plan)

By setting Y high enough that all major miners have a chance to mine
in the window,
this actually becomes an effective vote for the change by miners with
a stochastic
super-majority threshold.  All nodes are able to exactly determine at what block
the election has completed because it is an objective fact of the chain.

With this scheme the new encoding will only become active when enough mining
capacity supports it (or at least helpfully refuses to mine the who class
of transactions) so that a large reorganization will not happen due to
incompatible blocks during deployment.

This could be further enhanced with conflicting block discouragement (e.g.
refusing to extend or forward a rules violating block until it is burred)
but I think this scheme is sufficient without that, and that this is generally
superior to discouragement for this purpose.

Cheers.