Amend the BIP 123 process to include buried deployments
6 messages, 14 to 21 Feb 2018. Started by Marco Falke.
Marco Falke (2), Eric Voskuil (2), Luke Dashjr, Gregory Maxwell
I define a buried deployment as a consensus rule change that affects validity of blocks that are buried by a sufficiently large number of blocks in the current valid most-work chain, but the current block (and all its parents) remain valid.
BIP 123 suggests that BIPs in the consensus layer should be assigned a label "soft fork" or "hard fork". However, I think the differentiation into soft fork or hard fork should not be made for BIPs that document buried deployments. In contrast to soft forks and hard forks, buried deployments do not require community and miner coordination for a safe deployment.
For a chain fork to happen due to a buried deployment, a massive chain reorganization must be produced off of a block in the very past. In the extremely unlikely event of such a large chain reorganization, Bitcoin's general security assumptions would be violated regardless of the presence of a buried deployment.
On Wednesday 14 February 2018 10:01:46 PM Marco Falke via bitcoin-dev wrote:
Quoted text (6 lines)
> BIP 123 suggests that BIPs in the consensus layer should be assigned a > label "soft fork" or "hard fork". However, I think the differentiation > into soft fork or hard fork should not be made for BIPs that document > buried deployments. In contrast to soft forks and hard forks, buried > deployments do not require community and miner coordination for a safe > deployment.
They also do not require software coordination. Therefore, why should there be BIPs at all? Seems to me that we should instead add these documents to https://github.com/bitcoin-core/docs
That being said, I'm also okay with just adding an Annex to the original softfork/hardfork BIP describing each shortcut. It just seems annoying to have two BIPs for every protocol change: one for the change itself, and then another for implementation-specific shortcuts taken.
Luke
On Wed, Feb 14, 2018 at 10:11 PM, Luke Dashjr via bitcoin-dev <bitcoin-dev@lists•linuxfoundation•org> wrote:
Quoted text (11 lines)
> On Wednesday 14 February 2018 10:01:46 PM Marco Falke via bitcoin-dev wrote: >> BIP 123 suggests that BIPs in the consensus layer should be assigned a >> label "soft fork" or "hard fork". However, I think the differentiation >> into soft fork or hard fork should not be made for BIPs that document >> buried deployments. In contrast to soft forks and hard forks, buried >> deployments do not require community and miner coordination for a safe >> deployment. > > They also do not require software coordination. Therefore, why should there be > BIPs at all? Seems to me that we should instead add these documents to > https://github.com/bitcoin-core/docs
In that sense, no but they help people understand the system (e.g. so they don't go look at implementations and confuse that the activations they expect are simply not there); and they aid other implementations in understanding what other people have already analyzed and concluded was safe. You could certainly get an analysis wrong for one of these things.
On 02/14/2018 02:01 PM, Marco Falke via bitcoin-dev wrote:
Quoted text (3 lines)
> I define a buried deployment as a consensus rule change that affects > validity of blocks that are buried by a sufficiently large number of > blocks in the current valid most-work chain,
Sufficient for what, specifically?
Quoted text (1 line)
> but the current block (and all its parents) remain valid.
Remain valid in the case where the depth assumption is "sufficient" to ensure that a chain split is not possible?
If this was true (which it is not), it would imply that there is no reason to validate any block deeper than the most recent 25,000. Presumably this means that people may continuously rely on some authority (like Bitcoin Core?) to determine the checkpoint for tip-25,000.
Quoted text (6 lines)
> BIP 123 suggests that BIPs in the consensus layer should be assigned a > label "soft fork" or "hard fork". However, I think the differentiation > into soft fork or hard fork should not be made for BIPs that document > buried deployments. In contrast to soft forks and hard forks, buried > deployments do not require community and miner coordination for a safe > deployment.
They can only avoid this requirement based on the assumption that the hard fork cannot result in a chain split. This is not the case.
Quoted text (2 lines)
> For a chain fork to happen due to a buried deployment, a massive chain > reorganization must be produced off of a block in the very past.
In other words a "buried deployment" is a hard fork that is not likely to cause a chain split. This is a subjective subcategory of hard fork, not an independent category - unless maybe you can show that there is the 25,000 blocks number is an objective threshold.
Quoted text (3 lines)
> In the extremely unlikely event of such a large chain reorganization, > Bitcoin's general security assumptions would be violated regardless of > the presence of a buried deployment.
This is untrue. The "security assumptions" of Bitcoin do not preclude deep reorganizations.
e
Quoted text (3 lines)
> They also do not require software coordination. Therefore, why should there be > BIPs at all? Seems to me that we should instead add these documents to > https://github.com/bitcoin-core/docs
Consensus is not trivial. I think documentation is important, even if it seems simple to some. Personally, I don't care too much where to place the documentation, but the BIPs repo seems a good place, since it also hosts other informational documents.
To prevent "two BIPs for every protocol change", related buried deployments could be bundled. E.g. the ISM BIP 90 change.
On Wed, Feb 14, 2018 at 6:57 PM, Eric Voskuil <eric@voskuil•org> wrote:
Quoted text (6 lines)
> On 02/14/2018 02:01 PM, Marco Falke via bitcoin-dev wrote: >> I define a buried deployment as a consensus rule change that affects >> validity of blocks that are buried by a sufficiently large number of >> blocks in the current valid most-work chain, > > Sufficient for what, specifically?
Sufficiently large to prevent potential bike-shedding. The expected number of blocks in two weeks could be considered a lower bound. Then multiply that by 10 or 20.
Quoted text (11 lines)
> >> but the current block (and all its parents) remain valid. > > Remain valid in the case where the depth assumption is "sufficient" to > ensure that a chain split is not possible? > > If this was true (which it is not), it would imply that there is no > reason to validate any block deeper than the most recent 25,000. > Presumably this means that people may continuously rely on some > authority (like Bitcoin Core?) to determine the checkpoint for tip-25,000. >
Note that a checkpoint *freezes* the chain completely at a given height. Buried deployments are *not* checkpoints.
Also note that buried deployments only make sense after a protocol upgrade has happened (i.e. a soft fork or hard fork). If a miner has the resources to cause a chain split, they could trivially do that even in the complete absence of buried deployments. Buried deployments are *not* a solution to 50% attacks.
Quoted text (17 lines)
>> BIP 123 suggests that BIPs in the consensus layer should be assigned a >> label "soft fork" or "hard fork". However, I think the differentiation >> into soft fork or hard fork should not be made for BIPs that document >> buried deployments. In contrast to soft forks and hard forks, buried >> deployments do not require community and miner coordination for a safe >> deployment. > > They can only avoid this requirement based on the assumption that the > hard fork cannot result in a chain split. This is not the case. > >> For a chain fork to happen due to a buried deployment, a massive chain >> reorganization must be produced off of a block in the very past. > > In other words a "buried deployment" is a hard fork that is not likely > to cause a chain split. This is a subjective subcategory of hard fork, > not an independent category - unless maybe you can show that there is > the 25,000 blocks number is an objective threshold.
Please note that a buried deployment can very well be a soft fork. I think this makes it even clearer, that such a label makes no sense for buried deployments.
Quoted text (8 lines)
>> In the extremely unlikely event of such a large chain reorganization, >> Bitcoin's general security assumptions would be violated regardless of >> the presence of a buried deployment. > > This is untrue. The "security assumptions" of Bitcoin do not preclude > deep reorganizations. > e >
On 02/18/2018 10:57 AM, Marco Falke via bitcoin-dev wrote:
Quoted text (12 lines)
>> They also do not require software coordination. Therefore, why should there be >> BIPs at all? Seems to me that we should instead add these documents to >> https://github.com/bitcoin-core/docs > > Consensus is not trivial. I think documentation is important, even if > it seems simple to some. > Personally, I don't care too much where to place the documentation, > but the BIPs repo seems a good place, since it also hosts other > informational documents. > > To prevent "two BIPs for every protocol change", related buried > deployments could be bundled. E.g. the ISM BIP 90 change.
You seem to have missed the point. Either the "buried deployment" is a consensus rule, and requires a BIP, or it is not a consensus rule, and does not warrant a BIP.
You are arguing that it is not a consensus rule, yet requires a BIP. You also strongly imply that it is a consensus rule ("consensus is important").
If it is a consensus rule it is either a hard fork (valid tx set expansion) or a soft fork (valid tx set contraction). You are attempting to create an independent category that violates this clear engineering definition. The category you desire is actually a subcategory of hard fork (employing an arbitrary threshold for likelihood of causing a chain split).
Quoted text (11 lines)
> On Wed, Feb 14, 2018 at 6:57 PM, Eric Voskuil <eric@voskuil•org> wrote: >> On 02/14/2018 02:01 PM, Marco Falke via bitcoin-dev wrote: >>> I define a buried deployment as a consensus rule change that affects >>> validity of blocks that are buried by a sufficiently large number of >>> blocks in the current valid most-work chain, >> >> Sufficient for what, specifically? > > Sufficiently large to prevent potential bike-shedding. The expected > number of blocks in two weeks could be considered a lower bound. Then > multiply that by 10 or 20.
The arbitrary threshold. It seems it could be anything. Such a definition has no clear *engineering* usefulness.
Quoted text (12 lines)
>>> but the current block (and all its parents) remain valid. >> >> Remain valid in the case where the depth assumption is "sufficient" to >> ensure that a chain split is not possible? >> >> If this was true (which it is not), it would imply that there is no >> reason to validate any block deeper than the most recent 25,000. >> Presumably this means that people may continuously rely on some >> authority (like Bitcoin Core?) to determine the checkpoint for tip-25,000. >> > Note that a checkpoint *freezes* the chain completely at a given > height. Buried deployments are *not* checkpoints.
You are arguing a point that I did not make. The issue is that you argue a "buried deployment" hard fork cannot create a chain split. This itself implies that the chain is "frozen" at the depth below which the chain cannot be split. In other words, by accepting your logic, we must conclude there is no reason whatsoever to validate the chain prior to that depth. This would lead to the conclusion that check-pointing the chain to that depth is always sufficient validation.
Quoted text (5 lines)
> Also note that buried deployments only make sense after a protocol > upgrade has happened (i.e. a soft fork or hard fork). If a miner has > the resources to cause a chain split, they could trivially do that > even in the complete absence of buried deployments. Buried deployments > are *not* a solution to 50% attacks.
Not sure why you are making this obvious but seemingly-irrelevant point.
Quoted text (21 lines)
>>> BIP 123 suggests that BIPs in the consensus layer should be assigned a >>> label "soft fork" or "hard fork". However, I think the differentiation >>> into soft fork or hard fork should not be made for BIPs that document >>> buried deployments. In contrast to soft forks and hard forks, buried >>> deployments do not require community and miner coordination for a safe >>> deployment. >> >> They can only avoid this requirement based on the assumption that the >> hard fork cannot result in a chain split. This is not the case. >> >>> For a chain fork to happen due to a buried deployment, a massive chain >>> reorganization must be produced off of a block in the very past. >> >> In other words a "buried deployment" is a hard fork that is not likely >> to cause a chain split. This is a subjective subcategory of hard fork, >> not an independent category - unless maybe you can show that there is >> the 25,000 blocks number is an objective threshold. > > Please note that a buried deployment can very well be a soft fork. I > think this makes it even clearer, that such a label makes no sense for > buried deployments.
No, it cannot. Removal of an activated soft fork (valid tx set contraction) is a hard fork (valid tx set expansion), and a new activation rule for an active soft fork creates a path to that removal. Given this error you may want to reconsider your proposal.
Quoted text (8 lines)
>>> In the extremely unlikely event of such a large chain reorganization, >>> Bitcoin's general security assumptions would be violated regardless of >>> the presence of a buried deployment. >> >> This is untrue. The "security assumptions" of Bitcoin do not preclude >> deep reorganizations. >> >> e