-
Notifications
You must be signed in to change notification settings - Fork 648
architecture
Graphene is designed to allow multiple applications to connect to the network. An application consists of a p2p node which receives blocks and optionally broadcasts new transactions on behalf of the application's users. Applications contain core logic which consists of core chain state and core indexes, which is the minimal amount of data necessary to reach core consensus, shared global agreement among all nodes participating in the p2p network about core features.
Virtual property consists of the things that can be owned. Traditional cryptocurrencies, such as Bitcoin, have a single type of virtual property: A digital token which is transferrible, fungible (e.g. all Bitcoins are identical), and (for practical purposes) divisible.
Graphene has two different types of digital property: Base property, and derived property. (Economists would call base property fiat, but this term is already used in the cryptocurrency space as a retronym for government-issued currency, used to differentiate it from cryptocurrency.)
The base property in Graphene core consists of:
- Accounts
- Core asset
- UIA's
Derived property in Graphene core consists of:
- BitAssets
- Bonds
- Options
(TODO: Other types of derived property, e.g. escrow?)
Applications which leverage the Graphene blockchain for consensus may implement their own base and/or derived virtual property. (TODO: explain how application-level virtual property may interact with core property.)
Derived property is created by core smart contracts which are hard-coded in Graphene. Smart contracts can be as diverse as real-world contracts, but in Graphene, all core smart contracts are collateralized, two-party contracts. One side of the contract must post collateral which is used to perform settlement. The party which posts collateral is referred to as the long side.
Contract type | Long name | Long fungible? | Long transferrable? | Short name | Short fungible? | Short transferrable? |
---|---|---|---|---|---|---|
BitAsset | Long holder | Yes | Yes | Short holder | No | No |
Bond | Lender | No | Yes | Borrower | No | Yes |
Option | Option holder | ? | Yes | Option writer | ? | Yes |
A word about the rationale for short BitAssets not being transferrable: There is little technical obstacle to making short positions transferrable. However, if Alice is a weakly collateralized short seeking to exit, and Bob is a new short seeking to enter at high leverage, Alice and Bob both have incentive for her to sell her position to him. It is desirable to force Alice to cover and force Bob to short at the minimum initial leverage, to re-capitalize shorts over time.
-
Shorting mechanics. Shorts have maintenance margin requirement and user-settable stop-loss. TODO: Document exactly how this works.
-
Incremental order matching. Orders match incrementally when entered. TODO: Document exactly how this works.
-
Traditional price splitting. If orders match at a spread, they split the surplus 50-50. TODO: Document more exactly, including example.
-
Fill-or-kill orders. TODO: How do these work?
-
BitAssets 3.0. TODO: Turn forum posts, etc. into detailed spec including examples.
-
TaPoS. TODO: Explain this.
-
Short expiration time. TODO: Explain this.
-
Multisig uses authority system.
TODO: Document this (sort of "multisig for humans")
TODO: Document this
TODO: Document this, including very exact semantics
A custom op is a no-op with data.
A custom object is an object with data and an owner, the owner can update the data. TODO: Is this actually a thing?
TODO: Write the list
TODO: Document limitations of relative ID's. They can only be used in some operations -- which operations? TODO: Fix this limitation
- This is
theoreticalbts
idea for an interesting feature
This is a feature implemented in Namecoin. It is a commit/reveal procedure to prevent front-running of name registration.
When registering a new name, you can commit (H(name + separator + salt), recipient_pubkey)
in one tx, then within 24 hours,
reveal salt
in another tx to claim the name. If multiple claims to the same name are submitted, the claim with the
earliest commit time is given priority. NB the recipient pubkey is given in the commit, not the reveal, so someone else
front-running your reveal pays a fee but doesn't gain the name.
Note, this can result in situations where account name is revoked (because it tried to claim a name that was revealed earlier). So the named object (e.g. account, but are account objects the only named objects in Graphene?) still exists, but just becomes nameless.
- This is
theoreticalbts
idea for an interesting feature
Many user-bases already exist, and some of these may have name collisions. Common names like dan
or nathan
are probably
already registered on Github, Linkedin, Twitter, Google, Yahoo, etc. and probably belong to different people on all these services.
If our business model is to convince online services to migrate their user bases, then we should give them a way to namespace these
accounts. E.g. github/dan
, github/nathan
etc., in general a registration of a/b
must be approved by account a
.
Should this reflect referral structure. For example if a
is your referrer, then your name is a/b
. New accounts are always a/b
,
but can be promoted to b
by buying out. Hmm, seems like the buyout should also give you an opportunity to change your name (since
root NS might have conflicts), and this change should be name-blinded.
- This is
theoreticalbts
idea for an interesting feature
In traditional exchanges, unfilled orders are free -- market fees are only charged on matched orders. We have to charge a minimal amount per unfilled order as anti-spam measure. However, we can imagine an e(x)change provider (X)avier who hosts orders on an external server. When Alice wants to place an order, she creates an order transaction with no fee, then uploads the order to Xavier's server; Xavier publishes it (and Xavier will need to implement alternative anti-spam measures to protect his server from abuse).
When Bob wants to match Alice's order, he provides the fee.
Here's my idea for how to implement this without substantially re-working the fee structure. We create a special "community" account (TODO: better name) with a special flag which signals that no authority is needed to withdraw funds from it. Alice signs her tx paying the fee from the community account, now the only reason her tx is invalid is because the community account has no funds. Now Bob can create a wrapping transaction containing his matching order, funding for the community account, and Alice's tx. The wrapped tx is signed by Bob. Crucially, doing it this way means no one can insert a tx taking the money from the community account in between Bob's operation funding the community account and Alice's transaction paying it.
Can we do this with proposed tx's? We have to think very carefully about the exact semantics of proposed tx's.
- This is
theoreticalbts
idea for an interesting feature
This can also be used for third-party "account porting gateways". E.g. let's say we have email
account and want to
set up a way where anyone with an email address can claim the email address in BitShares form. So for example
email/user_at_example_dot_com
would be given to someone who proves they control [email protected]
. The holder of
the email
BTS account, and not core witnesses / validation, is responsible for checking this proof (which may
require arbitrary off-blockchain actions that cannot be validated in a non-decentralized way, like sending confirmation
emails). The validation consists of signature checking ("user email
confirmed pubkey p
owns email/user_at_example_dot_com
")
combined with the email
account's policy ("we require successful response to registration mail before giving out names")
combined with trust in email
account (by sending money to account email/user_at_example_dot_com
, user is trusting email
account honestly associated user_at_example_dot_com
with the correct person).
This method can also be applied to any website which has a userbase that has a login API (Github / Twitter / LinkedIn / etc.), or even merely the ability for members to post content (e.g. in forum profile) -- if you give a user a challenge and they successfully post it in their profile or other publication area, they've successfully confirmed access to that account.
With wrapped transactions, the account porting gateway can create a no-fee tx assigning the name to the user, the user then adds the fee to actually register the account. This allows the account porting gateway to avoid having to solve the economic problem of determining which registrations will result in profitable CLV, and focus solely on the technical problem of verifying the owner of an existing name in a third-party system like email, DNS, Google accounts, etc., while still claiming referrer fees. Admittedly there's no "free lunch," the UX is a little more rocky because the user needs to provide their own funding.
- This is
theoreticalbts
idea for an interesting feature
There needs to be an "I lost my email address and private keys" button which allows email
to revoke the name email/user_at_example_dot_com
.
However, the underlying account should still exist, it just needs to be unlinked from the name (this way if user
later finds
their private keys, they still have access to funds). The name should be unable to be reassigned until a long enough delay which
at least allows transactions with TaPoS before the revocation block become invalid (otherwise if Alice sends to user_at_example_dot_com
and Eve controls email
and a single witness, Eve can have her witness censor Alice's tx inclusion in the block, instead including a tx
assigning email/user_at_example_dot_com
to herself, and taking the funds when the tx appears in a later block).
NB, this attack may be more difficult in practice, because clients do lookup of name-to-account-ID mapping locally. It still makes sense
to have a revocation period -- this way if you've heard from someone in the last 30 days that their address is email/user_at_example_dot_com
,
then you know you'll either send the funds to the right person or get an error. Also, wallets should warn if one of your contacts
has been revoked and reassigned.
It also complicates account history, as the name displayed will be determined by the mapping for the name at the time the tx was performed.
- This is
theoreticalbts
idea for an interesting feature
An assertion op is an operation which invalidates a transaction unless the asserted condition holds. So far we have:
- TaPoS assertion. Asserts that a particular block hash exists in the history. All transactions have this assertion, it prevents transactions from migrating to forks where ID's have different objects.
These may need to be included:
- Data object assertion. We may be required to check a data object which has an ID, owner and custom content. We may assert a data object exists with the given ID and owner, and some function of the content is true.
Examples of data object assertion functions: Assert that the content follows a certain byte pattern (including gaps), assert the content is a Merkle / Patricia proof that x is in S, assert some hash of the content is a given value. These can be combined to construct an AXCT smart contract, and the latter is also useful for name-blinding schemes.
-
Date assertions. Not-valid-before or not-valid-after. Every tx has expiration (not-valid-after) in the near future, and can be not-valid-before some point in the recent past via TaPoS. However it may be useful to extend these windows, e.g. create a transaction not-valid-before some time in the future (this is
nLockTime
in Bitcoin). Or create a transaction not-valid-after some time in the far future. A mechanism for more flexible date assertions than the default TaPoS / expiration should be provided. It is acceptable to require the use of proposed transactions for this mechanism. -
Authority assertions. An authority assertion is an op which adds the given authority to the list of authority required to make the transaction valid. Authority assertion is useful to make platform actions and app actions atomic. E.g. Alice and Bob want to create a transaction to trade her AppCoin for Bob's BitUSD. The tx includes a custom op interpreted by the app layer as transfer of Alice's AppCoins to Bob, which the app doesn't honor without Alice's signature; and a regular (platform) transfer op sending Bob's BitUSD to Alice. Without an authority assertion, Alice's signature also needs to be a custom op, which also means Alice has to sign first. (If Alice used a regular signature, Eve would be able to play the role of Alice, strip her own signature and apply Bob's signature, taking Bob's BitUSD without compensating him with her AppCoins.) Authority assertions mean that app signatures don't have to be wrapped in this way, and allows tx to be signed in any order.