MIME-Version: 1.0
Sender: rgrant@rgrant.org
In-Reply-To: <CAMnpzfr5hLFqRm0_KWb6=N8YsD4dfQBtRBPFwfjz-EHA31d34A@mail.gmail.com>
References: <CAMnpzfr5hLFqRm0_KWb6=N8YsD4dfQBtRBPFwfjz-EHA31d34A@mail.gmail.com>
From: Ryan Grant <bitcoin-dev@rgrant.org>
Date: Sun, 11 Jun 2017 15:31:00 -0400
Message-ID: <CAMnpzfrx5QA0qqOjfRn=q0_HUtcDpbhAKxRLPitU02ycxN1+kQ@mail.gmail.com>
To: bitcoin-dev@lists.linuxfoundation.org
Content-Type: text/plain; charset="UTF-8"
Subject: Re: [bitcoin-dev] extended BIP9 activation of segwit,
	for legacy nodes
Precedence: list

This[1] idea from April would assist in a BIP149-like segwit
activation on November 16th.

Its goal is to be incredibly easy to test and deploy, right now, even
before a decision on revisions to BIP149 is made, and well before such
"BIP149ish" testing is itself complete.

UASFs don't need time for most legacy nodes to upgrade - that's the
point of a soft fork.  UASFs simply need to have inevitability,
which is provided by some nodes more than others.  But for the node
less instrumental in that inevitability, and more relaxed about
scheduling upgrade work, being moved to a miner-protected consensus
ruleset is not as desirable a position as the opportunity to
participate fully.  As a courtesy, the plan for soft forks has
always been to allow legacy nodes time to upgrade to full
participation.  How much time should rollouts allow for this
courtesy?

Extended BIP9 activation of segwit (for legacy nodes) separates
concerns between intending to activate segwit and its method of
deployment, allowing "semi-legacy" nodes that have upgraded to
include this proposal to participate immediately in a successful
segwit activation, without needing any courtesy time to upgrade to
the particular deployment logic.

Code for deployment is included in the original email[1].  There's
nothing missing from the logic shown.  The whole intent of the
proposal is that other deployment specifics are left to be defined
by future proposals.  In the same block that THRESHOLD_ACTIVE is
reached for segwit, require Consensus::DEPLOYMENT_SEGWIT_ALT1 to
also reach THRESHOLD_ACTIVE, and the burden of the future proposal
is fulfilled.

If the idea proves broken or of no benefit, when actually
implementing and testing future deployments, then we can avoid using
DEPLOYMENT_SEGWIT_ALT1 until it expires, and zero nodes get hurt.

Let's look at how this affects options for how to deploy segwit...

/ BIP149ish UASF, without this proposal /

roadmap:
  - BIP149ish debated (handles BIP141 success, unlike BIP149)
  - BIP149ish tested
  - BIP149ish courtesy timeout debated
  - BIP148    fails
  - BIP149ish released
  - BIP141    fails
  - BIP149ish courtesy timeout expires
  - BIP149ish activates
  - segwit activates
  - legacy nodes must upgrade before recognizing BIP149 activation

If we try to restrict ourselves to only the original service bit, we
see a tension between activating soon and leaving legacy nodes on a
miner-protected consensus.  We also see a tension between *planning*
how long it will take to debate and test UASF logic, and setting
expectations for when a reasonable activation date is.

These seemingly small tensions complicate the solution, and push it
out of developer's visions of what is feasible.  It's not clear how
soon it can happen, so it doesn't get started in an urgent manner.
It doesn't get started in an urgent manner, so it's not clear how
soon it can happen.

/ BIP149ish UASF, with this proposal /

roadmap:
  - this proposal debated
  - this proposal tested
  - this proposal released (courtesy begins)
  - BIP149ish     debated  (handles BIP141 success, unlike BIP149)
  - BIP149ish     tested
  - BIP148        fails
  - BIP149ish     released
  - BIP141        fails
  - BIP149ish     activates
  - segwit activates
  - semi-legacy nodes *immediately* use segwit, via this proposal

We can remove the courtesy timeout problem, because the courtesy
begins as soon as this proposal goes live.  This simplifies debate on
how BIP149ish should deploy, and helps make reasonable a much quicker
segwit activation should BIP141 fail.

With a clear route to quick activation for semi-legacy nodes, the
UASF can be planned for activation in as short a window as the key
nodes can upgrade.  Not even all the key nodes need to upgrade: it's
still a soft fork, and UASFs simply need to have inevitability.

These differences can help us aim for activating segwit on November
16th, if BIP141 and BIP148 do not succeed earlier.  Since BIP149 as
originally conceived is slow as molasses, BIP149ish still needs
debate, BIP141 has steadfast enemies, and the community is slow to
adapt to BIP148's complicated commitment requirements, it is prudent
to take this intermediate step allowing quicker BIP149ish activation.

[1] https://lists.linuxfoundation.org/pipermail/bitcoin-dev/2017-April/014160.html