Gregory Maxwell [ARCHIVE] on Nostr: 📅 Original date posted:2015-09-30 📝 Original message:On Wed, Sep 30, 2015 at ...
📅 Original date posted:2015-09-30
📝 Original message:On Wed, Sep 30, 2015 at 5:11 PM, Mike Hearn <hearn at vinumeris.com> wrote:
> Hi Gregory,
>
>>
>> I'm surprised to see this response
>
>
> Why? I have objected to the idea of soft forks many times. I wrote an entire
> article about it in August.
Yes, your article contained numerous factual and logical inaccuracies
which I corrected (many of which you had been previously corrected on
as well by others). (For example:
https://www.reddit.com/r/Bitcoin/comments/3griiv/on_consensus_and_forks_by_mike_hearn/cu0yv0r
)
I would have hoped that after so many corrections you would have
updated your beliefs.
> There's no such thing as a "real" hard fork - don't try and move the goal
> posts. SPV clients do not need any changes to do the right thing with BIP
> 101, they will follow the new chain automatically, so it needs no changes.
Yes, because what 101 does is not a hard-fork from the perspective of
BitcoinJ clients. Please do not conflate BitcoinJ with all of SPV; a
SPV client could validate the information received more extensively or
respond to alerts in reject rule violating blocks-- BitcoinJ does
not, but this is BitcoinJ's design decision to lack security in this
respect and not something inherent to SPV).
Directly fixing the time-warp attack, for example, would be a hard
fork from the perspective of BitcoinJ clients. Recovering the fixed 0
bits in the header for use as extra-nonce would be a hard fork from
the perspective of BitcoinJ clients. Changing the transaction format
to include an explicit nonce for ECDH (e.g. stealth addresses) would
be a hard fork from the perspective of BitcoinJ clients. Increasing
the precision of Bitcoin by 1000 would be a hard fork from the
perspective of BitcoinJ. As would adjusting the hashtree to commit to
fees, including fees under OP_CHECKSIGs hash, or switching to the
segregated witness commitment structure from elements alpha that
allows syncing the chain without fetching signatures... all that would
be hardforks from the perspective of BitcoinJ.
Because of an cheaply avoidable the lack of validation in BitcoinJ no
increase of the blocksize is a hard-fork from its perspective. Nor
would increasing the subsidy to miners, or allowing third parties to
confiscate coins. But other SPV clients could, if they wanted to,
reject blocks the violated most of these criteria.
The argument you are presenting against BIP65 is that it is bad
because it is silently accepted. But this applies no less to 101 for
SPV clients, and in 101's case it's a failure to enforce pre-existing
rules which the users might care a great deal about. Worse, counting
on this kind of behavior can build a dependence on weak security forms
of SPV and inhibits the use of full security SPV.
In truth, both of BIP101 and BIP65 are detectable even by the most
simplistic and pre-change clients due to the voluntarily use of block
version signaling. Any participant in the network is free to take
whatever action they choose to take in response to such an event.
Bitcoin Core's behavior is to issue alerts to the user when unexpected
block versions show up on the network. Users and implementer are free
to turn changes like BIP65 into hardforks from the perspective of
their own system, necessitating manual intervention, by simply forcing
the block version to be a particular value (or shutting down when
there are many blocks of a new version; until manually authorized to
continue).
> Several people have asked several times now: given the very real and widely
> acknowledged downsides that come with a soft fork, what is the specific
> benefit to end users of doing them?
For many changes, including CLTV the actual soft fork change is by far
the most natural way of implementing the change itself. One simply
takes an existing non-standard placeholder op code sequence and
assigns it the new VERIFY style meaning. It is clean, tidy, and the
result is nearly as if the system has had it all along. The only
complexity is around the activation and can be dropped in future code.
Beyond that, the primary upside is no forced industry wide "flag day"
where everyone is _forced_ to modify their software arises, taking
considerable cost. People who care about the new rule can use it,
people who don't don't. All the rules that you care about enforcing
remain in force-- you still prevent inflation, you still will not
tolerate the theft of your own coins (or those of most other people),
etc.. No one is necessarily caught by surprise since the block
versions communicate that something is happening, allowing network
participants to choose to act (or not).
For example, for years you stonewalled P2SH and multi-signature. You
didn't care about it. You didn't think it was valuable. You didn't add
it to your software, even after it was well specified and deployed in
production. Could it have been done as a hard-fork? Likely not: you
would have prevented it. But as a soft-fork you were free to ignore it
with no ill-effect for a long time existing for those who cared about
it, and not for you, until widespread use resulted in demand enough to
justify accepting a patch that permitted sending to it.
What if we'd needed a hard fork to enable CoinJoin or other privacy
features? I think would have blocked that too.
The relative ease of handling soft-forks which you are indifferent to
means that there is little reason to object to a compatible change
that gives other people flexibility they care about greatly but which
you are indifferent to; and it forces people who would oppose a
functionality because they don't want others to have some piece of
freedom to try to frame justifications in language other than "I don't
think it's worth the cost" since they have the nearly free option of
ignoring the change-- they're forced to actually argue against other
people having that freedom.
Soft-forks also allows us to deploy fixes to the Bitcoin protocol
which are more important-to-have but not urgently critical (like
height in coinbase), or sometimes to deploy fixes to critical
vulnerabilities without first handing everyone excruciatingly detailed
instructions on exploiting them, simply by closing off an pattern of
protocol which is obviously bad and risky.
The primary cost of a soft-fork for non-participants is simply some
risk of increased network instability around the change-- but short
lived forks happen every day, and longer lived ones happen from time
to time. Larger amounts of instability occur from time to time due to
network partitioning, misconfiguration, and software bugs-- and client
software must be prepared to cope with it; this is a fact of bitcoin
and decenteralized systems in general. Upgraded, change enforcing,
client software is not exposed to the this instability, and
non-upgraded software could choose to mitigate any exposure by
monitoring the block versions. This is a far better situation that the
natural instability that will happen from time to time in a
decentralized system.
By contrast, the programmed activation point of BIP101 at 75% almost
guarantees activation among considerable controversy, promising
network instability which BitcoinJ clients would experience upgraded
or not, even if the larger block side was ultimately the losing side
in the switch. I find it more than a little strange that you think the
instability of a 75%-version-hashpower cut is acceptable but the a
95%-version-hashpower compatible change is not.
Finally, there is the demonstrated track record: They work; they
deliver new features to people. Our experience in the half dozen or
more soft-forks in the system so far is that in practice do not cause
significant problems, including financial losses for SPV wallet users.
Even with that complete success there has been room for improvement,
which is why the process has evolved over time to feature things like
preemptive non-standardness, high switchover thresholds, etc. and
these will continue to evolve over time.
I hope that you can put aside your effort to force a blocksize
increase on others for a moment and add functionality, of the kind the
Bitcoin Core has had for years, to BitcoinJ to improve the experience
with soft-forks if you think it isn't good enough as is...
📝 Original message:On Wed, Sep 30, 2015 at 5:11 PM, Mike Hearn <hearn at vinumeris.com> wrote:
> Hi Gregory,
>
>>
>> I'm surprised to see this response
>
>
> Why? I have objected to the idea of soft forks many times. I wrote an entire
> article about it in August.
Yes, your article contained numerous factual and logical inaccuracies
which I corrected (many of which you had been previously corrected on
as well by others). (For example:
https://www.reddit.com/r/Bitcoin/comments/3griiv/on_consensus_and_forks_by_mike_hearn/cu0yv0r
)
I would have hoped that after so many corrections you would have
updated your beliefs.
> There's no such thing as a "real" hard fork - don't try and move the goal
> posts. SPV clients do not need any changes to do the right thing with BIP
> 101, they will follow the new chain automatically, so it needs no changes.
Yes, because what 101 does is not a hard-fork from the perspective of
BitcoinJ clients. Please do not conflate BitcoinJ with all of SPV; a
SPV client could validate the information received more extensively or
respond to alerts in reject rule violating blocks-- BitcoinJ does
not, but this is BitcoinJ's design decision to lack security in this
respect and not something inherent to SPV).
Directly fixing the time-warp attack, for example, would be a hard
fork from the perspective of BitcoinJ clients. Recovering the fixed 0
bits in the header for use as extra-nonce would be a hard fork from
the perspective of BitcoinJ clients. Changing the transaction format
to include an explicit nonce for ECDH (e.g. stealth addresses) would
be a hard fork from the perspective of BitcoinJ clients. Increasing
the precision of Bitcoin by 1000 would be a hard fork from the
perspective of BitcoinJ. As would adjusting the hashtree to commit to
fees, including fees under OP_CHECKSIGs hash, or switching to the
segregated witness commitment structure from elements alpha that
allows syncing the chain without fetching signatures... all that would
be hardforks from the perspective of BitcoinJ.
Because of an cheaply avoidable the lack of validation in BitcoinJ no
increase of the blocksize is a hard-fork from its perspective. Nor
would increasing the subsidy to miners, or allowing third parties to
confiscate coins. But other SPV clients could, if they wanted to,
reject blocks the violated most of these criteria.
The argument you are presenting against BIP65 is that it is bad
because it is silently accepted. But this applies no less to 101 for
SPV clients, and in 101's case it's a failure to enforce pre-existing
rules which the users might care a great deal about. Worse, counting
on this kind of behavior can build a dependence on weak security forms
of SPV and inhibits the use of full security SPV.
In truth, both of BIP101 and BIP65 are detectable even by the most
simplistic and pre-change clients due to the voluntarily use of block
version signaling. Any participant in the network is free to take
whatever action they choose to take in response to such an event.
Bitcoin Core's behavior is to issue alerts to the user when unexpected
block versions show up on the network. Users and implementer are free
to turn changes like BIP65 into hardforks from the perspective of
their own system, necessitating manual intervention, by simply forcing
the block version to be a particular value (or shutting down when
there are many blocks of a new version; until manually authorized to
continue).
> Several people have asked several times now: given the very real and widely
> acknowledged downsides that come with a soft fork, what is the specific
> benefit to end users of doing them?
For many changes, including CLTV the actual soft fork change is by far
the most natural way of implementing the change itself. One simply
takes an existing non-standard placeholder op code sequence and
assigns it the new VERIFY style meaning. It is clean, tidy, and the
result is nearly as if the system has had it all along. The only
complexity is around the activation and can be dropped in future code.
Beyond that, the primary upside is no forced industry wide "flag day"
where everyone is _forced_ to modify their software arises, taking
considerable cost. People who care about the new rule can use it,
people who don't don't. All the rules that you care about enforcing
remain in force-- you still prevent inflation, you still will not
tolerate the theft of your own coins (or those of most other people),
etc.. No one is necessarily caught by surprise since the block
versions communicate that something is happening, allowing network
participants to choose to act (or not).
For example, for years you stonewalled P2SH and multi-signature. You
didn't care about it. You didn't think it was valuable. You didn't add
it to your software, even after it was well specified and deployed in
production. Could it have been done as a hard-fork? Likely not: you
would have prevented it. But as a soft-fork you were free to ignore it
with no ill-effect for a long time existing for those who cared about
it, and not for you, until widespread use resulted in demand enough to
justify accepting a patch that permitted sending to it.
What if we'd needed a hard fork to enable CoinJoin or other privacy
features? I think would have blocked that too.
The relative ease of handling soft-forks which you are indifferent to
means that there is little reason to object to a compatible change
that gives other people flexibility they care about greatly but which
you are indifferent to; and it forces people who would oppose a
functionality because they don't want others to have some piece of
freedom to try to frame justifications in language other than "I don't
think it's worth the cost" since they have the nearly free option of
ignoring the change-- they're forced to actually argue against other
people having that freedom.
Soft-forks also allows us to deploy fixes to the Bitcoin protocol
which are more important-to-have but not urgently critical (like
height in coinbase), or sometimes to deploy fixes to critical
vulnerabilities without first handing everyone excruciatingly detailed
instructions on exploiting them, simply by closing off an pattern of
protocol which is obviously bad and risky.
The primary cost of a soft-fork for non-participants is simply some
risk of increased network instability around the change-- but short
lived forks happen every day, and longer lived ones happen from time
to time. Larger amounts of instability occur from time to time due to
network partitioning, misconfiguration, and software bugs-- and client
software must be prepared to cope with it; this is a fact of bitcoin
and decenteralized systems in general. Upgraded, change enforcing,
client software is not exposed to the this instability, and
non-upgraded software could choose to mitigate any exposure by
monitoring the block versions. This is a far better situation that the
natural instability that will happen from time to time in a
decentralized system.
By contrast, the programmed activation point of BIP101 at 75% almost
guarantees activation among considerable controversy, promising
network instability which BitcoinJ clients would experience upgraded
or not, even if the larger block side was ultimately the losing side
in the switch. I find it more than a little strange that you think the
instability of a 75%-version-hashpower cut is acceptable but the a
95%-version-hashpower compatible change is not.
Finally, there is the demonstrated track record: They work; they
deliver new features to people. Our experience in the half dozen or
more soft-forks in the system so far is that in practice do not cause
significant problems, including financial losses for SPV wallet users.
Even with that complete success there has been room for improvement,
which is why the process has evolved over time to feature things like
preemptive non-standardness, high switchover thresholds, etc. and
these will continue to evolve over time.
I hope that you can put aside your effort to force a blocksize
increase on others for a moment and add functionality, of the kind the
Bitcoin Core has had for years, to BitcoinJ to improve the experience
with soft-forks if you think it isn't good enough as is...