Just to expand a tiny bit here, while the testnet setup of only a few nodes acting as "bridges", mainnet already has many systems which act as effective bridges today - there are several relay networks in use which effectively bypass the P2P network, including my legacy relay network (which many miners historically have used, and I'd expect those who aren't paying attention and don't upgrade will not turn off, fixing the issue for them), ViaBTC's super aggressive bandwidth-wasting block announcement network which pushes blocks from several pools to many nodes globally, and Bitcoin.com's private relay network. (Of course many other miners and pools have private relay networks, but the several other such networks I'm aware of are already segwit-compatible, even for pools not signaling segwit).

Matt

On March 27, 2017 12:22:43 PM PDT, Suhas Daftuar via bitcoin-dev <bitcoin-dev@lists.linuxfoundation.org> wrote:
Hi,

There have been two threads recently that have made references to peer-to-peer implementation details in Bitcoin Core's Segregated Witness code that I would like to clarify.

In the thread "Issolated Bitcoin Nodes" (https://lists.linuxfoundation.org/pipermail/bitcoin-dev/2017-March/013765.html), there was some discussion about how Bitcoin Core's block download logic behaves after segwit activation.  After segwit activation, Bitcoin Core nodes will not (currently) attempt to download any blocks from non-segwit peers (nodes that do not set the NODE WITNESS service bit).  This is a bandwidth optimization to prevent a node from downloading a block that may be invalid only because the sender omitted the witness, requiring re-download until the block is received with the required witness data.

But to be clear, non-segwit blocks -- that is, blocks without a witness commitment in the coinbase, and whose transactions are serialized without witnesses, and whose transactions are not spending segwit outputs which require a witness -- are evaluated under the same rules as prior, pre-segwit versions of the software.  So such non-segwit blocks that are valid to older, pre-segwit nodes are also valid to segwit-nodes.

In https://lists.linuxfoundation.org/pipermail/bitcoin-dev/2017-March/013796.html, Eric Voskuil wrote:

Given the protocol requirements of the segwit proposal this is not the
case. A miner running pre-segwit code will produce blocks that no
segwit node will ever receive.

The phrase "protocol requirements of segwit" is confusing here, because there are two different layers that need consideration: the consensus protocol layer and the peer-to-peer protocol layer.  But in neither layer is the behavior of not downloading blocks from non-NODE WITNESS peers a "requirement".  This is an implementation detail in the Bitcoin Core code that alternate implementations compliant with BIP 144 could implement differently.

At the consensus layer, non-segwit blocks (described above) that are valid to older nodes are also valid to segwit nodes.  That means that if a miner was using an older, pre-segwit version of Bitcoin Core to produce blocks after segwit activates, that blocks they find will be valid to all nodes.

At the p2p layer, though, segwit-enabled Bitcoin Core nodes will only try to download those blocks if announced by a segwit-enabled peer.  But this is not a protocol requirement; other implementations can remain compatible even they take different approaches here.  (As an example, I could imagine an implementation that downloaded a new block from any peer, but if the block has a witness commitment in the coinbase and was received from a non-segwit peer, then the node would attempt re-download from a segwit peer.  I'm sure many other reasonable block download strategies could be devised.)

Still, if a miner wants to continue mining post-segwit activation, but using pre-segwit software, they would need a way to relay their blocks to a segwit-enabled peer.

There are a few ways to do this that I can think of:

- Use the RPC call "submitblock" on a segwit-enabled node.  Calling "submitblock" on a Bitcoin Core 0.13.1 (0.13.0 in the case of testnet) or later node works fine as long as the block is valid (whether or not it has a witness commitment or witness transactions), and once a segwit-enabled peer has the block it will relay to other segwit peers.

- Explicitly deliver the block to a segwit node over the p2p network, even if unrequested.  Currently Bitcoin Core at least will process unrequested blocks, and valid blocks that update the tip will then be relayed to other peers.

- Run a bridge node, which advertises NODE_WITNESS and can serialize blocks with witness data, which downloads blocks even from non-NODE WITNESS peers.  Anyone can do this to bridge the networks for the benefit of the whole network (I have personally been running a few nodes that do this, for several months now), but miners concerned about this issue for their own blocks could explicitly do this themselves to ensure that their own blocks propagate to the segwit-enabled network.

- Peer directly with other miners, bypassing the p2p network.  Many miners do this already, using protocols which may already serve to bridge the network.

So saying that "A miner running pre-segwit code will produce blocks that no segwit node will ever receive" is not really correct, in my view.  If the whole network were just running Bitcoin Core software releases, and the miner was not able/willing to deliver their block to a segwit-enabled node (eg by using the RPC call "submitblock", or one of the other suggestions I had above), then I would agree with the statement.  But given that there are bridge nodes on the network, and that miners have other options to relay their block, I think this is not an accurate portrayal of what would actually happen on the network -- I would expect that non-segwit miners' blocks would still get relayed post-segwit activation, even if only by the handful of bridge nodes that I expect are currently running.

All that said, I do think this is an important detail to highlight and that this behavior should be better documented (I believe it deserves specific mention in a BIP), as this is an important issue for miners to be aware of.