This does indeed sound reasonable. The chances of having a cut in the
network consisting of non-upgraded nodes partitioning the network and not
forwarding the segregated witnesses should be minimal, given a long rollout
phase before the activation.

If everybody agrees that this is a better way to approach the normalization
issue we should probably start writing it up and see if we can get critical
mass behind it :-)

On Wed, Nov 4, 2015 at 5:00 AM Peter Todd <pete@petertodd.org> wrote:

> On Tue, Nov 03, 2015 at 09:44:02PM +0000, Christian Decker via bitcoin-dev
> wrote:
> > Ok, so assuming we can get a connected component of upgraded nodes that
> > relay both the transaction and the associated external scripts then we
> > could just piggyback the external scripts on top of the normal messages.
> > Non-upgraded nodes will read the entire two-part message but only parse
> the
> > classical transaction, dropping the external script. Validation rules for
> > upgraded nodes are the same as before: if the attached signatures are
> > invalid the entire TX is dropped. We have to commit to the external
> scripts
> > used during the creation of a block. I think the easiest way to add this
> > commitment is the coinbase input I guess, and following the transaction
> > list a new list of signature lists is shipped with the rest of the block.
> > Non-upgraded will ignore it as before.
> >
> > Would that work? It all hinges on having upgraded miners in a connected
> > component otherwise non-upgraded nodes will drop the external scripts on
> > the way (since they parse and then reconstruct the messages along the
> > path). But if it works this could be a much nicer solution.
>
> FWIW my replace-by-fee fork does preferential peering with other RBF
> nodes to ensure that you'll always be connected to at least some
> full-RBF peers. In practice this works very well, and I'm sure a similar
> scheme could be used in this situation as well.
>
> Basically, conceptually unless you're connected to peers that advertise
> that they relay the new data, you treat the situation as though you're
> not connected to any peers at all. No different than if for some reason
> none of your peers were advertising NODE_NETWORK.
>
> --
> 'peter'[:-1]@petertodd.org
> 00000000000000000247b0e7436a5169ac6f9087be0295d10b07bf0bcbd4c0cc
>