UUID to identify chains (payment protocol and elsewhere)
12 messages, 20 to 22 May 2013. Started by Mark Friedenbach.
Jeff Garzik (4), Mark Friedenbach (2), Luke Dashjr (2), Melvin Carvalho (2), Mike Hearn, Gavin Andresen
At the developer round-table it was asked if the payment protocol would alt-chains, and Gavin noted that it has a UTF-8 encoded string identifying the network ("main" or "test"). As someone with two proposals in the works which also require chain/coin identification (one for merged mining, one for colored coins), I am opinionated on this. I believe that we need a standard mechanism for identifying chains, and one which avoids the trap of maintaining a standard registry of string-to-chain mappings.
Any chain can be uniquely identified by its genesis block, 122 random bits is more than sufficient for uniquely tagging chains/colored assets, and the low-order 16-bytes of the block's hash are effectively random. With these facts in mind, I propose that we identify chains by UUID.
So as to remain reasonably compliant with RFC 4122, I recommend that we use Version 4 (random) UUIDs, with the random bits extracted from the double-SHA256 hash of the genesis block of the chain. (For colored coins, the colored coin definition transaction would be used instead, but I will address that in a separate proposal and will say just one thing about it: adopting this method for identifying chains/coins will greatly assist in adopting the payment protocol to colored coins.)
The following Python code illustrates how to construct the chain identifier from the serialized genesis block:
from hashlib import sha256
from uuid import UUID
def chain_uuid(serialized_genesis_block):
h = sha256(serialized_genesis_block).digest()
h = sha256(h).digest()
h = h[:16]
h = ''.join([
h[:6],
chr(0x40 | ord(h[6]) & 0x0f),
h[7],
chr(0x80 | ord(h[8]) & 0x3f),
h[9:]
])
return UUID(bytes=h)
And some example chain identifiers:
mainnet: UUID('6fe28c0a-b6f1-4372-81a6-a246ae63f74f')
testnet3: UUID('43497fd7-f826-4571-88f4-a30fd9cec3ae')
namecoin: UUID('70c7a9f0-a2fb-4d48-a635-a70d5b157c80')
As for encoding the chain identifier, the simplest method is to give "network" the "bytes" type, but defining a "UUID" message type is also possible. In either case bitcoin mainnet would be the default, so the extra 12 bytes (vs: "main" or "test") would only be an issue for alt-chains or colored coins.
Kind regards, Mark Friedenbach
On Mon, May 20, 2013 at 7:59 PM, Mark Friedenbach <mark@monetize•io> wrote:
Quoted text (7 lines)
> So as to remain reasonably compliant with RFC 4122, I recommend that we > use Version 4 (random) UUIDs, with the random bits extracted from the > double-SHA256 hash of the genesis block of the chain. (For colored > coins, the colored coin definition transaction would be used instead, > but I will address that in a separate proposal and will say just one > thing about it: adopting this method for identifying chains/coins will > greatly assist in adopting the payment protocol to colored coins.)
This proposal seems closer to Version 5 than Version 4, in spirit. But given that useful content may be deduced from UUID, it is not truly applicable to either. A bitcoin-specific version 6, if you will.
Quoted text (5 lines)
> And some example chain identifiers:
>
> mainnet: UUID('6fe28c0a-b6f1-4372-81a6-a246ae63f74f')
> testnet3: UUID('43497fd7-f826-4571-88f4-a30fd9cec3ae')
> namecoin: UUID('70c7a9f0-a2fb-4d48-a635-a70d5b157c80')Note that, as this example unintentionally implies, humans are going to want a side-by-side mapping /anyway/, just to make it readable and usable to humans.
Almost all useful multi-chain software will require a readable shortname string anyway, the thing this proposal wishes to avoid.
Bitcoinj already has such chain id's and we use standard Java style reverse DNS names: org.bitcoin.main, etc. If we want a more global naming system that seems like a good compromise between uniqueness and readability. On 20 May 2013 19:45, "Jeff Garzik" <jgarzik@exmulti•com> wrote:
Quoted text (46 lines)
> On Mon, May 20, 2013 at 7:59 PM, Mark Friedenbach <mark@monetize•io>
> wrote:
> > So as to remain reasonably compliant with RFC 4122, I recommend that we
> > use Version 4 (random) UUIDs, with the random bits extracted from the
> > double-SHA256 hash of the genesis block of the chain. (For colored
> > coins, the colored coin definition transaction would be used instead,
> > but I will address that in a separate proposal and will say just one
> > thing about it: adopting this method for identifying chains/coins will
> > greatly assist in adopting the payment protocol to colored coins.)
>
> This proposal seems closer to Version 5 than Version 4, in spirit.
> But given that useful content may be deduced from UUID, it is not
> truly applicable to either. A bitcoin-specific version 6, if you
> will.
>
>
> > And some example chain identifiers:
> >
> > mainnet: UUID('6fe28c0a-b6f1-4372-81a6-a246ae63f74f')
> > testnet3: UUID('43497fd7-f826-4571-88f4-a30fd9cec3ae')
> > namecoin: UUID('70c7a9f0-a2fb-4d48-a635-a70d5b157c80')
>
> Note that, as this example unintentionally implies, humans are going
> to want a side-by-side mapping /anyway/, just to make it readable and
> usable to humans.
>
> Almost all useful multi-chain software will require a readable
> shortname string anyway, the thing this proposal wishes to avoid.
>
> --
> Jeff Garzik
> exMULTI, Inc.
> jgarzik@exmulti•com
>
>
> ------------------------------------------------------------------------------
> Try New Relic Now & We'll Send You this Cool Shirt
> New Relic is the only SaaS-based application performance monitoring service
> that delivers powerful full stack analytics. Optimize and monitor your
> browser, app, & servers with just a few lines of code. Try New Relic
> and get this awesome Nerd Life shirt! http://p.sf.net/sfu/newrelic_d2d_may
> _______________________________________________
> Bitcoin-development mailing list
> Bitcoin-development@lists•sourceforge•net
> https://lists.sourceforge.net/lists/listinfo/bitcoin-development
>This was meant to go to everyone:
On 5/20/13 7:45 PM, Jeff Garzik wrote:
Quoted text (12 lines)
> On Mon, May 20, 2013 at 7:59 PM, Mark Friedenbach <mark@monetize•io> wrote: >> So as to remain reasonably compliant with RFC 4122, I recommend that we >> use Version 4 (random) UUIDs, with the random bits extracted from the >> double-SHA256 hash of the genesis block of the chain. (For colored >> coins, the colored coin definition transaction would be used instead, >> but I will address that in a separate proposal and will say just one >> thing about it: adopting this method for identifying chains/coins will >> greatly assist in adopting the payment protocol to colored coins.) > This proposal seems closer to Version 5 than Version 4, in spirit. > But given that useful content may be deduced from UUID, it is not > truly applicable to either. A bitcoin-specific version 6, if you > will.
That is true, and perhaps we have enough clout to push an RFC specifying a double-SHA256 Version 6, or at least get it reserved. I proposed Version 4 (random) because any UUID library should allow you to specify the 122 supposedly random bits of that version, whereas conceivably there might exist UUID libraries that require a SHA1 pre-image to create a Version 5 UUID (I know of no examples though). Regardless, making an official double-SHA256 UUID version RFC is an option worth considering.
Quoted text (11 lines)
> And some example chain identifiers:
>
> mainnet: UUID('6fe28c0a-b6f1-4372-81a6-a246ae63f74f')
> testnet3: UUID('43497fd7-f826-4571-88f4-a30fd9cec3ae')
> namecoin: UUID('70c7a9f0-a2fb-4d48-a635-a70d5b157c80')
> Note that, as this example unintentionally implies, humans are going
> to want a side-by-side mapping /anyway/, just to make it readable and
> usable to humans.
>
> Almost all useful multi-chain software will require a readable
> shortname string anyway, the thing this proposal wishes to avoid.I think there are perhaps two issues being conflated here (and in Mike's response): the UI identifying the network/coin to the user, and the matching of the protocol-supplied value to the underlying network/coin by the client/daemon. The former necessarily involves manual adjustments (e.g, localization), but it's preferable for the latter to be a self-validating reference to the block chain. This is a trivial difference for multi-chain wallets (what are you doing receiving requests for coins in chains you don't know about?), but is important for colored coins. Let me explain:
I will be proposing soon a colored coin architecture that allows issuance of new coins by anyone for a fee, by means of a special category of transaction. The hash of that issuing transaction would then be used to generate a UUID identifying the asset for the payment protocol and other purposes as well, analogous to how the hash of the genesis block identifies the host currency, bitcoin. It is expected that there will be many such coins issued, as they can be used to represent individual loans or lines of credit. In this context, any colored-coin aware client could scan the block chain (or lookup a maintained index) to discover the UUID -> coin mapping with absolute certainty. However the mechanism for mapping the text "mtgoxUSD" to a specific coin is not clear, and using some sort of DNS-resolution system adds huge external dependencies. IMHO it is much better to have the identifier derived from block chain data directly (and therefore accessible and trusted by all nodes), and then carry out optional UI mappings like UUID(...) -> "mtgoxUSD" at a higher level.
Does that make sense?
Bitcoin currently uses raw hashes extensively as UUIDs; whether the payment protocol should be influence by that or not, I've not given thought to yet.
Some alt coins may share a blockchain, or even merely the genesis block (two currently do; despite one of those being a scamcoin, I think the possibility should not be dismissed). Because of this, requiring a 1:1 mapping between genesis block and chain or coin seems non-ideal.
On Monday, May 20, 2013 11:59:39 PM Mark Friedenbach wrote:
Quoted text (65 lines)
> At the developer round-table it was asked if the payment protocol would
> alt-chains, and Gavin noted that it has a UTF-8 encoded string
> identifying the network ("main" or "test"). As someone with two
> proposals in the works which also require chain/coin identification (one
> for merged mining, one for colored coins), I am opinionated on this. I
> believe that we need a standard mechanism for identifying chains, and
> one which avoids the trap of maintaining a standard registry of
> string-to-chain mappings.
>
> Any chain can be uniquely identified by its genesis block, 122 random
> bits is more than sufficient for uniquely tagging chains/colored assets,
> and the low-order 16-bytes of the block's hash are effectively random.
> With these facts in mind, I propose that we identify chains by UUID.
>
> So as to remain reasonably compliant with RFC 4122, I recommend that we
> use Version 4 (random) UUIDs, with the random bits extracted from the
> double-SHA256 hash of the genesis block of the chain. (For colored
> coins, the colored coin definition transaction would be used instead,
> but I will address that in a separate proposal and will say just one
> thing about it: adopting this method for identifying chains/coins will
> greatly assist in adopting the payment protocol to colored coins.)
>
> The following Python code illustrates how to construct the chain
> identifier from the serialized genesis block:
>
> from hashlib import sha256
> from uuid import UUID
> def chain_uuid(serialized_genesis_block):
> h = sha256(serialized_genesis_block).digest()
> h = sha256(h).digest()
> h = h[:16]
> h = ''.join([
> h[:6],
> chr(0x40 | ord(h[6]) & 0x0f),
> h[7],
> chr(0x80 | ord(h[8]) & 0x3f),
> h[9:]
> ])
> return UUID(bytes=h)
>
> And some example chain identifiers:
>
> mainnet: UUID('6fe28c0a-b6f1-4372-81a6-a246ae63f74f')
> testnet3: UUID('43497fd7-f826-4571-88f4-a30fd9cec3ae')
> namecoin: UUID('70c7a9f0-a2fb-4d48-a635-a70d5b157c80')
>
> As for encoding the chain identifier, the simplest method is to give
> "network" the "bytes" type, but defining a "UUID" message type is also
> possible. In either case bitcoin mainnet would be the default, so the
> extra 12 bytes (vs: "main" or "test") would only be an issue for
> alt-chains or colored coins.
>
> Kind regards,
> Mark Friedenbach
>
> ---------------------------------------------------------------------------
> --- Try New Relic Now & We'll Send You this Cool Shirt
> New Relic is the only SaaS-based application performance monitoring service
> that delivers powerful full stack analytics. Optimize and monitor your
> browser, app, & servers with just a few lines of code. Try New Relic
> and get this awesome Nerd Life shirt! http://p.sf.net/sfu/newrelic_d2d_may
> _______________________________________________
> Bitcoin-development mailing list
> Bitcoin-development@lists•sourceforge•net
> https://lists.sourceforge.net/lists/listinfo/bitcoin-developmentOn 21 May 2013 01:59, Mark Friedenbach <mark@monetize•io> wrote:
Quoted text (52 lines)
> At the developer round-table it was asked if the payment protocol would
> alt-chains, and Gavin noted that it has a UTF-8 encoded string
> identifying the network ("main" or "test"). As someone with two
> proposals in the works which also require chain/coin identification (one
> for merged mining, one for colored coins), I am opinionated on this. I
> believe that we need a standard mechanism for identifying chains, and
> one which avoids the trap of maintaining a standard registry of
> string-to-chain mappings.
>
> Any chain can be uniquely identified by its genesis block, 122 random
> bits is more than sufficient for uniquely tagging chains/colored assets,
> and the low-order 16-bytes of the block's hash are effectively random.
> With these facts in mind, I propose that we identify chains by UUID.
>
> So as to remain reasonably compliant with RFC 4122, I recommend that we
> use Version 4 (random) UUIDs, with the random bits extracted from the
> double-SHA256 hash of the genesis block of the chain. (For colored
> coins, the colored coin definition transaction would be used instead,
> but I will address that in a separate proposal and will say just one
> thing about it: adopting this method for identifying chains/coins will
> greatly assist in adopting the payment protocol to colored coins.)
>
> The following Python code illustrates how to construct the chain
> identifier from the serialized genesis block:
>
> from hashlib import sha256
> from uuid import UUID
> def chain_uuid(serialized_genesis_block):
> h = sha256(serialized_genesis_block).digest()
> h = sha256(h).digest()
> h = h[:16]
> h = ''.join([
> h[:6],
> chr(0x40 | ord(h[6]) & 0x0f),
> h[7],
> chr(0x80 | ord(h[8]) & 0x3f),
> h[9:]
> ])
> return UUID(bytes=h)
>
> And some example chain identifiers:
>
> mainnet: UUID('6fe28c0a-b6f1-4372-81a6-a246ae63f74f')
> testnet3: UUID('43497fd7-f826-4571-88f4-a30fd9cec3ae')
> namecoin: UUID('70c7a9f0-a2fb-4d48-a635-a70d5b157c80')
>
> As for encoding the chain identifier, the simplest method is to give
> "network" the "bytes" type, but defining a "UUID" message type is also
> possible. In either case bitcoin mainnet would be the default, so the
> extra 12 bytes (vs: "main" or "test") would only be an issue for
> alt-chains or colored coins.
>This is essentially name spacing. As registries grow namespaces become more important. In bitcoin's quest for decentrality there's also the question of who maintains the registry.
Some out of band algo/hash could work so long as there was a one to one relationship between the described object and the UUID. In this case the gensis block may not uniquely identify a coin.
The normal way to namespace a registry on the internet is to allow it to be a URI. In this case an http style uri has the added bonus side effect that it can be dereferencable and both human and machine readable. So yes something like org.bitcoin.* is good, just simply growing things to http style uris is cleaner, imho
Quoted text (16 lines)
> > Kind regards, > Mark Friedenbach > > > ------------------------------------------------------------------------------ > Try New Relic Now & We'll Send You this Cool Shirt > New Relic is the only SaaS-based application performance monitoring service > that delivers powerful full stack analytics. Optimize and monitor your > browser, app, & servers with just a few lines of code. Try New Relic > and get this awesome Nerd Life shirt! http://p.sf.net/sfu/newrelic_d2d_may > _______________________________________________ > Bitcoin-development mailing list > Bitcoin-development@lists•sourceforge•net > https://lists.sourceforge.net/lists/listinfo/bitcoin-development >
On Wed, May 22, 2013 at 6:27 AM, Melvin Carvalho <melvincarvalho@gmail•com> wrote:
Quoted text (3 lines)
> Some out of band algo/hash could work so long as there was a one to one > relationship between the described object and the UUID. In this case the > gensis block may not uniquely identify a coin.
What does this mean? It seems extremely unlikely that two different genesis blocks will have the same hash.
On 22 May 2013 16:07, Jeff Garzik <jgarzik@exmulti•com> wrote:
Quoted text (9 lines)
> On Wed, May 22, 2013 at 6:27 AM, Melvin Carvalho > <melvincarvalho@gmail•com> wrote: > > Some out of band algo/hash could work so long as there was a one to one > > relationship between the described object and the UUID. In this case the > > gensis block may not uniquely identify a coin. > > What does this mean? It seems extremely unlikely that two different > genesis blocks will have the same hash. >
Two coin ecosystems could have the same genesis block
Quoted text (6 lines)
> > -- > Jeff Garzik > exMULTI, Inc. > jgarzik@exmulti•com >
On Wed, May 22, 2013 at 10:12 AM, Melvin Carvalho <melvincarvalho@gmail•com> wrote:
Quoted text (14 lines)
> On 22 May 2013 16:07, Jeff Garzik <jgarzik@exmulti•com> wrote: >> >> On Wed, May 22, 2013 at 6:27 AM, Melvin Carvalho >> <melvincarvalho@gmail•com> wrote: >> > Some out of band algo/hash could work so long as there was a one to one >> > relationship between the described object and the UUID. In this case >> > the >> > gensis block may not uniquely identify a coin. >> >> What does this mean? It seems extremely unlikely that two different >> genesis blocks will have the same hash. > > > Two coin ecosystems could have the same genesis block
That has really, really bad side effects. The whole point of the bitcoin consensus algorithm is to avoid situations like this.
We don't want to encourage that behavior with code.
On Wednesday, May 22, 2013 2:20:22 PM Jeff Garzik wrote:
Quoted text (21 lines)
> On Wed, May 22, 2013 at 10:12 AM, Melvin Carvalho > > <melvincarvalho@gmail•com> wrote: > > On 22 May 2013 16:07, Jeff Garzik <jgarzik@exmulti•com> wrote: > >> On Wed, May 22, 2013 at 6:27 AM, Melvin Carvalho > >> > >> <melvincarvalho@gmail•com> wrote: > >> > Some out of band algo/hash could work so long as there was a one to > >> > one relationship between the described object and the UUID. In this > >> > case the > >> > gensis block may not uniquely identify a coin. > >> > >> What does this mean? It seems extremely unlikely that two different > >> genesis blocks will have the same hash. > > > > Two coin ecosystems could have the same genesis block > > That has really, really bad side effects. The whole point of the > bitcoin consensus algorithm is to avoid situations like this. > > We don't want to encourage that behavior with code.
In some cases, multiple currencies can use the same blockchain (not just the singular genesis block). This use case *is* something we want to encourage - no reason for people to make an entirely new blockchain if their altcoin fits within the scope of Bitcoin or another existing altchain.
On Wed, May 22, 2013 at 10:29 AM, Luke-Jr <luke@dashjr•org> wrote:
Quoted text (4 lines)
> In some cases, multiple currencies can use the same blockchain (not just the > singular genesis block). This use case *is* something we want to encourage - > no reason for people to make an entirely new blockchain if their altcoin fits > within the scope of Bitcoin or another existing altchain.
OK, let me qualify. Layers on top are one thing, but we really do not want to support cases like the fork that leaves the genesis block intact, and leaves the subsidy at 50.0 BTC forever.
Getting back to the original proposal:
RE: uuid instead of "main" / "test" in the payment protocol:
I vote no.
The payment protocol will become at least 3 BIPs:
1) Protocol messages (current gist, essentially) 2) MIME type 3) bitcoin: URI extension
An alt coin will need its own version of (2) and (3), so when you click on a foocoin: link a foocoin-specific MIME type is fetched and foocoin.exe is launched to handle the request.
... or a specific MIME type is fetched and delivered to the HandlesLotsOfCoins application (... and it knows what MIME type it is getting, so can Do the Right Thing).
If a payment request is delivered via HTTP or email, it will be bundled up in an envelope of some sort with the MIME type attached.
So, after further thought, I've changed my mind: which coin would be encoded in the MIME type. Which chain for that coin would be encoded in PaymentDetails.network.