BUIP098: Bitcoin Unlimited's Strategy for the November 2018 Hard Fork
Proposer: Andrew Stone (@theZerg)
Submitted on: 2018-08-21
Status: draft


“Run Bitcoin Unlimited to vote for compromise”

There are 2 changesets proposed for the November 2018 hard fork that
have a variety of supporters but can be summarized as coming from
Bitcoin ABC and nChain. To review:

1. Increase block size to 128MB
2. Re-activate additional opcodes (OP_MUL, OP_LSHIFT, OP_RSHIFT,
OP_INVERT)
3. Remove restriction on the number of instructions executed per script
(currently 200)
1. Include OP_CHECKDATASIG
2. Limit transaction size to > 100 bytes (to solve a possible but
expensive attack detailed
here)
3. Lexicae transaction ordering
4. Consensus enforcement that scriptsig (spend scripts) contain only
data push instructions

It is ironic that these changesets are mutually compatible, yet both
groups reject the other’s changes. There may be some specific critiques
(see Appendix B) of various proposals, but the core the rationale behind
the rejections seem to be the same used to block Group tokenization –
fewer changes are better because every change introduces risk.
Additionally, there is concern that the blocking of certain features is
happening due to undisclosed patented technologies that compete with the
proposed features. By blocking the feature, the patent remains valuable.

Representatives of Bitcoin Unlimited have explored the idea of
compromise with representatives from both groups with no success so far,
even the smallest changes (like changing a constant to one better
suited) have been rejected. Given the “no changes, no matter how
reasonable, except mine” strategy being pursued by both of these
organizations, I can only sadly conclude that this is again about power

I believe that the proponents of Bitcoin Cash need to stick together and
come to a compromise, rather than fork and face another dispersion of
economic activity. This is the essential conclusion of Metcalfe’s law.
With the 30 day median block size at 36.6Kb, I invite you to examine the
above feature list and identify those whose inclusion will compensate
for splitting the community due to the dramatic and rapid increase in

When I proposed the original plan of hard forks every 6 months the
intention was to onboard as many people as possible by including many
use cases, and accept the risk these changes implied. This strategy has
been a failure. The periodic hard fork has not been used to activate any
feature that resulted in significant new consumer-focused use cases.
Such changes may modify just a few lines of code, but have large
ramifications on the use of the blockchain and the community is
“bundle” individual organizations’ favorite features into a single
“swallow the sweet with the bitter” package.

I would like to propose a strategy for Bitcoin Unlimited for the near
future. In essence, our message will be “run Bitcoin Unlimited to vote
for compromise”. The Bitcoin Unlimited client will incorporate features
from both organizations and allow these features to either be activated
via BIP135 (a generalized form of BIP9 miner voting via version bits),
explicit configuration, or (development time and feasibility permitting)
emergent consensus. By allowing BIP135, we move to a miner voting
process that allows individual features to gain agreement before
activation. By allowing explicit configuration – that is, allowing a
user to force the feature “on” or “off” – people running the BUcash
full node can quickly react to any hash-power surprises.

Appendix A: A few notes on BIP135

BIP135
is a superset of BIP9. BIP9 has hard-coded activation thresholds and
times and these are quite optimistic. For example, it proposed a 95%
activation threshold, yet during the fight for larger blocks it became
clear that well over 5% of the hash power actually had much larger
investments in alt-coin hashing
hardware
.
Although the economic model of Bitcoin assumes that 51% of the hash
power wants what is “best” for the currency, its is a flawed to assume
that for 100% of the hash power.

BIP135 allows activation thresholds and times to be configured. Note
that these can be configured with the BIP9 values to make a particular
activation bit backwards compatible with BIP9-only full nodes.

Note that BIP135 allows for a grace period after a feature is “locked
in” and before it actually activates on mainnet. This period is used to
allow clients that have not implemented the feature to actually
implement it.

BIP135 also defines an end to the voting process, so failed initiatives
can be removed from clients that pre-implemented them. It is our
expectation that part of a reasonable path forward would be a common
understanding that version bits voting should be used when at least one
implementation has the corresponding feature set available and
well-tested.

Finally, note that BIP135 implicitly allows feature obsolescence and
removal some time after activation. The removal of a something can
itself be defined as a “feature” and assigned a bit.

Appendix B: General Arguments Against Various Features & BU Specific Notes

Note: I don’t necessarily agree or disagree with any of the arguments
presented here. This is a general summary of arguments [and rebuttals
in brackets] that I have heard. Actually, I think that quite a few of
them are invalid, but make up your own mind. Please comment if you would
like another argument or rebuttal added.

1. 128MB block size increase:
1. There is no need at this time. Blocks are not even close to 1MB!
It’s just for marketing. [when adoption comes it will be a
tsunami]
2. Dangerous. We have not even exercised 32MB blocks on mainnet. No
pool has mined a block > 8MB.
3. BU note: We already support 128MB blocks
2. Re-activation of opcodes
1. Proposal came too late – it missed the generally-agreed upon
and code does seem to be available within the date]
2. BU note: Code here:
1. D1592 - Add SCRIPT_ENABLE_MAGNETIC_OPCODES flag -
https://reviews.bitcoinabc.org/D1592
2. D1593 - Expand IsOpcodeDisabled() -
https://reviews.bitcoinabc.org/D1593
3. D1594 - rename monolith_opcodes.cpp to … -
https://reviews.bitcoinabc.org/D1594
4. D1631 - enable magnetic opcodes in … -
https://reviews.bitcoinabc.org/D1631
5. D1598 - OP_MUL implementation -
https://reviews.bitcoinabc.org/D1598
6. D1606 - OP_INVERT implementation -
https://reviews.bitcoinabc.org/D1606
7. D1638 - OP_LSHIFT & OP_RSHIFT -
https://reviews.bitcoinabc.org/D1638
8. D1631 - enable magnetic upgrade in tests -
https://reviews.bitcoinabc.org/D1631
3. Removal of instruction execution count restrictions
1. Significant time is needed for people to think about attack
scripts, consuming too much CPU or memory, for example.
2. no use cases
3. provided use cases can be done off chain with zero knowledge
proofs
4. The idea of freezing the instruction set and then using hundreds
or thousands of instructions to implement general purpose
primitives like EC multiplication is incredibly wasteful of UTXO
and blockchain history space
5. BU note: small LOC change (nchain commit:
b47906926fe5b71549d1b422f2219ccdd10a5a0d)
4. OP_CHECKDATASIG
1. Is not part of Bitcoin’s original instruction set [we have
already diverged from the original instruction set. Limiting
changes to restoring the original set is the authority
(nostalgia?) fallacy]
2. Can be used to enable wagering which may be illegal in some
jurisdictions [bitcoin itself is illegal in some jurisdictions,
information is not generally illegal, and the blockchain should
not and realistically cannot enforce legality on its
participants just like cars cannot and do not enforce speed
limits]
3. BU note: We have significant experience here and can implement
quickly
5. Limit transaction size to > 100 bytes (fixes
this)
1. Transaction != 64 bytes fixes the problem, why choose 100?
2. Attack is extremely expensive and only tricks SPV wallets [but
SPV wallet access is very important for bitcoin cash, and the
attack will become cheaper]
3. There may be a change to the MERKLEBLOCK protocol message that
also fixes the problem – that is, a non-consensus fix may exist
(https://lists.linuxfoundation.org/pipermail/bitcoin-dev/2018-August/016298.html)
4. BU note: small LOC change
6. Make scriptsig contain only data push instructions
1. This is already enforced at the network level, so probably not
contentious, yet not in the nChain list
7. Lexical Transaction Ordering
1. All the value is available through voluntary,
non-consensus-enforced, transaction ordering. [But by enforcing
it, we ensure that the most efficient is always used]
2. The stated original purpose (to enable parallel validation) is
actually a feature of removing dependency ordering, not adding
lexical ordering, and it even turns out that that’s not true –
the proposed parallel validation algorithm can be implemented to
efficiently verify dependency ordering.
3. It is a major change, for little to no impact. [it prepares us
for huge 1GB+ future blocks]
4. There is no spec, just marketing documents
5. There is no rush, so let’s defer and investigate the
possibilities
(https://gist.github.com/gavinandresen/e20c3b5a1d4b97f79ac2#canonical-ordering-of-transactions)