MIME-Version: 1.0
Sender: gmaxwell@gmail.com
In-Reply-To: <0213B102-7595-4177-A76C-FE4E8D7F0EDF@xbt.hk>
References: <0213B102-7595-4177-A76C-FE4E8D7F0EDF@xbt.hk>
From: Gregory Maxwell <greg@xiph.org>
Date: Sat, 17 Feb 2018 02:33:32 +0000
Message-ID: <CAAS2fgT_AjhaZ8AtG8H-8=0tiXTULK9ciLy-a-vW6u==MyBbcQ@mail.gmail.com>
To: Johnson Lau <jl2012@xbt.hk>, 
	Bitcoin Protocol Discussion <bitcoin-dev@lists.linuxfoundation.org>
Content-Type: text/plain; charset="UTF-8"
Subject: Re: [bitcoin-dev] Alternative way to count sigops
Precedence: list

On Fri, Feb 16, 2018 at 10:49 PM, Johnson Lau via bitcoin-dev
<bitcoin-dev@lists.linuxfoundation.org> wrote:
> Since we have a block weight limit of 4,000,000 and sigop limit of 80,000,
> each sigop could not use more than 50 weight unit on average. For new script
> proposals we could count the actual number of sigop at execution (i.e. skip
> unexecuted sigop, skip 0-size signature, count the actual checksig
> operations in multi-sig), and make sure the number of executed sigop * 50 is
> not greater than the size of the input.

We have a related policy rule in Bitcoin Core for some time now, the
weight of the transaction for the purpose of mining is
max(weight,lambda*sigops), though we set lambda a bit lower than makes
sense due to how checkmultisig.  This policy rule replaced an earlier
one which was almost equivalent to your proposal: it rejected
transactions with too many sigops per the byte count, but we found it
block actual more or less sensible transactions.

Going forward I don't think this is a great framework.  It works if
the only expensive operations all involve large input data, but I
think many proposals people have made for new operations would have
computational cost which requires relatively small amounts of
additional input-- aggregation is just one fairly minor example.