7 Composing choices

It’s time to put everything you’ve learnt so far together into a complete and secure Daml model for asset issuance, management, transfer, and trading. This application will have capabilities similar to the one in IOU Quickstart Tutorial. In the process you will learn about a few more concepts:

  • Daml projects, packages and modules
  • Composition of transactions
  • Observers and stakeholders
  • Daml’s execution model
  • Privacy

The model in this section is not a single Daml file, but a Daml project consisting of several files that depend on each other.

Hint

Remember that you can load all the code for this section into a folder called intro7 by running daml new intro7 --template daml-intro-7

Daml projects

Daml is organized in projects, packages and modules. A Daml project is specified using a single daml.yaml file, and compiles into a package in Daml’s intermediate language, or bytecode equivalent, Daml-LF. Each Daml file within a project becomes a Daml module, which is a bit like a namespace. Each Daml project has a source root specified in the source parameter in the project’s daml.yaml file. The package will include all modules specified in *.daml files beneath that source directory.

You can start a new project with a skeleton structure using daml new project-name in the terminal. A minimal project would contain just a daml.yaml file and an empty directory of source files.

Take a look at the daml.yaml for the chapter 7 project:
sdk-version: __VERSION__
name: __PROJECT_NAME__
source: daml
version: 1.0.0
dependencies:
  - daml-prim
  - daml-stdlib
  - daml-script
sandbox-options:
  - --wall-clock-time

You can generally set name and version freely to describe your project. dependencies does what the name suggests: It includes dependencies. You should always include daml-prim and daml-stdlib. The former contains internals of compiler and Daml Runtime, the latter gives access to the Daml Standard Library. daml-script contains the types and standard library for Daml Script.

You compile a Daml project by running daml build from the project root directory. This creates a dar file in .daml/dist/dist/${project_name}-${project_version}.dar. A dar file is Daml’s equivalent of a JAR file in Java: it’s the artifact that gets deployed to a ledger to load the package and its dependencies. dar files are fully self-contained in that they contain all dependencies of the main package. More on all of this in 9 Working with Dependencies.

Project structure

This project contains an asset holding model for transferable, fungible assets and a separate trade workflow. The templates are structured in three modules: Intro.Asset, Intro.Asset.Role, and Intro.Asset.Trade.

In addition, there are tests in modules Test.Intro.Asset, Test.Intro.Asset.Role, and Test.Intro.Asset.Trade.

All but the last .-separated segment in module names correspond to paths relative to the project source directory, and the last one to a file name. The folder structure therefore looks like this:

.
├── daml
│   ├── Intro
│   │   ├── Asset
│   │   │   ├── Role.daml
│   │   │   └── Trade.daml
│   │   └── Asset.daml
│   └── Test
│       └── Intro
│           ├── Asset
│           │   ├── Role.daml
│           │   └── Trade.daml
│           └── Asset.daml
└── daml.yaml

Each file contains a module header. For example, daml/Intro/Asset/Role.daml:

module Intro.Asset.Role where

You can import one module into another using the import keyword. The LibraryModules module imports all six modules:

import Intro.Asset

Imports always have to appear just below the module declaration. You can optionally add a list of names after the import to import only the selected names:

import DA.List (sortOn, groupOn)

If your module contains any Daml Scripts, you need to import the corresponding functionality:

import Daml.Script

Project overview

The project both changes and adds to the Iou model presented in 6 Parties and authority:

  • Assets are fungible in the sense that they have Merge and Split choices that allow the owner to manage their holdings.

  • Transfer proposals now need the authorities of both issuer and newOwner to accept. This makes Asset safer than Iou from the issuer’s point of view.

    With the Iou model, an issuer could end up owing cash to anyone as transfers were authorized by just owner and newOwner. In this project, only parties having an AssetHolder contract can end up owning assets. This allows the issuer to determine which parties may own their assets.

  • The Trade template adds a swap of two assets to the model.

Composed choices and scripts

This project showcases how you can put the Update and Script actions you learnt about in 6 Parties and authority to good use. For example, the Merge and Split choices each perform several actions in their consequences.

  • Two create actions in case of Split
  • One create and one archive action in case of Merge
    choice Split
      : SplitResult
      with
        splitQuantity : Decimal
      controller owner
      do
        splitAsset <- create this with
          quantity = splitQuantity
        remainder <- create this with
          quantity = quantity - splitQuantity
        return SplitResult with
          splitAsset
          remainder

    choice Merge
      : ContractId Asset
      with
        otherCid : ContractId Asset
      controller owner
      do
        other <- fetch otherCid
        assertMsg
          "Merge failed: issuer does not match"
          (issuer == other.issuer)
        assertMsg
          "Merge failed: owner does not match"
          (owner == other.owner)
        assertMsg
          "Merge failed: symbol does not match"
          (symbol == other.symbol)
        archive otherCid
        create this with
          quantity = quantity + other.quantity

The return function used in Split is available in any Action context. The result of return x is a no-op containing the value x. It has an alias pure, indicating that it’s a pure value, as opposed to a value with side-effects. The return name makes sense when it’s used as the last statement in a do block as its argument is indeed the “return”-value of the do block in that case.

Taking transaction composition a step further, the Trade_Settle choice on Trade composes two exercise actions:

    choice Trade_Settle
      : (ContractId Asset, ContractId Asset)
      with
        quoteAssetCid : ContractId Asset
        baseApprovalCid : ContractId TransferApproval
      controller quoteAsset.owner
      do
        fetchedBaseAsset <- fetch baseAssetCid
        assertMsg
          "Base asset mismatch"
          (baseAsset == fetchedBaseAsset with
            observers = baseAsset.observers)

        fetchedQuoteAsset <- fetch quoteAssetCid
        assertMsg
          "Quote asset mismatch"
          (quoteAsset == fetchedQuoteAsset with
            observers = quoteAsset.observers)

        transferredBaseCid <- exercise
          baseApprovalCid TransferApproval_Transfer with
            assetCid = baseAssetCid

        transferredQuoteCid <- exercise
          quoteApprovalCid TransferApproval_Transfer with
            assetCid = quoteAssetCid

        return (transferredBaseCid, transferredQuoteCid)

The resulting transaction, with its two nested levels of consequences, can be seen in the test_trade script in Test.Intro.Asset.Trade:

TX #15 1970-01-01T00:00:00Z (Test.Intro.Asset.Trade:77:23)
#15:0
│   known to (since): 'Alice' (#15), 'Bob' (#15)
└─> 'Bob' exercises Trade_Settle on #13:1 (Intro.Asset.Trade:Trade)
          with
            quoteAssetCid = #10:1; baseApprovalCid = #14:2
    children:
    #15:1
    │   known to (since): 'Alice' (#15), 'Bob' (#15)
    └─> fetch #11:1 (Intro.Asset:Asset)

    #15:2
    │   known to (since): 'Alice' (#15), 'Bob' (#15)
    └─> fetch #10:1 (Intro.Asset:Asset)

    #15:3
    │   known to (since): 'USD_Bank' (#15), 'Bob' (#15), 'Alice' (#15)
    └─> 'Alice',
        'Bob' exercises TransferApproval_Transfer on #14:2 (Intro.Asset:TransferApproval)
              with
                assetCid = #11:1
        children:
        #15:4
        │   known to (since): 'USD_Bank' (#15), 'Bob' (#15), 'Alice' (#15)
        └─> fetch #11:1 (Intro.Asset:Asset)

        #15:5
        │   known to (since): 'Alice' (#15), 'USD_Bank' (#15), 'Bob' (#15)
        └─> 'Alice', 'USD_Bank' exercises Archive on #11:1 (Intro.Asset:Asset)

        #15:6
        │   referenced by #17:0
        │   known to (since): 'Bob' (#15), 'USD_Bank' (#15), 'Alice' (#15)
        └─> create Intro.Asset:Asset
            with
              issuer = 'USD_Bank'; owner = 'Bob'; symbol = "USD"; quantity = 100.0; observers = []

    #15:7
    │   known to (since): 'EUR_Bank' (#15), 'Alice' (#15), 'Bob' (#15)
    └─> 'Bob',
        'Alice' exercises TransferApproval_Transfer on #12:1 (Intro.Asset:TransferApproval)
                with
                  assetCid = #10:1
        children:
        #15:8
        │   known to (since): 'EUR_Bank' (#15), 'Alice' (#15), 'Bob' (#15)
        └─> fetch #10:1 (Intro.Asset:Asset)

        #15:9
        │   known to (since): 'Bob' (#15), 'EUR_Bank' (#15), 'Alice' (#15)
        └─> 'Bob', 'EUR_Bank' exercises Archive on #10:1 (Intro.Asset:Asset)

        #15:10
        │   referenced by #16:0
        │   known to (since): 'Alice' (#15), 'EUR_Bank' (#15), 'Bob' (#15)
        └─> create Intro.Asset:Asset
            with
              issuer = 'EUR_Bank'; owner = 'Alice'; symbol = "EUR"; quantity = 90.0; observers = []

Similar to choices, you can see how the scripts in this project are built up from each other:

test_issuance = do
  setupResult@(alice, bob, bank, aha, ahb) <- setupRoles

  assetCid <- submit bank do
    exerciseCmd aha Issue_Asset
      with
        symbol = "USD"
        quantity = 100.0

  Some asset <- queryContractId bank assetCid
  assert (asset == Asset with
      issuer = bank
      owner = alice
      symbol = "USD"
      quantity = 100.0
      observers = []
        )

  return (setupResult, assetCid)

In the above, the test_issuance script in Test.Intro.Asset.Role uses the output of the setupRoles script in the same module.

The same line shows a new kind of pattern matching. Rather than writing setupResult <- setupRoles and then accessing the components of setupResult using _1, _2, etc., you can give them names. It’s equivalent to writing

setupResult <- setupRoles
case setupResult of
  (alice, bob, bank, aha, ahb) -> ...

Just writing (alice, bob, bank, aha, ahb) <- setupRoles would also be legal, but setupResult is used in the return value of test_issuance so it makes sense to give it a name, too. The notation with @ allows you to give both the whole value as well as its constituents names in one go.

Daml’s execution model

Daml’s execution model is fairly easy to understand, but has some important consequences. You can imagine the life of a transaction as follows:

Command Submission
A user submits a list of Commands via the Ledger API of a Participant Node, acting as a Party hosted on that Node. That party is called the requester.
Interpretation
Each Command corresponds to one or more Actions. During this step, the Update corresponding to each Action is evaluated in the context of the ledger to calculate all consequences, including transitive ones (consequences of consequences, etc.). The result of this is a complete Transaction. Together with its requestor, this is also known as a Commit.
Blinding
On ledgers with strong privacy, projections (see Privacy) for all involved parties are created. This is also called projecting.
Transaction Submission
The Transaction/Commit is submitted to the network.
Validation
The Transaction/Commit is validated by the network. Who exactly validates can differ from implementation to implementation. Validation also involves scheduling and collision detection, ensuring that the transaction has a well-defined place in the (partial) ordering of Commits, and no double spends occur.
Commitment
The Commit is actually committed according to the commit or consensus protocol of the Ledger.
Confirmation
The network sends confirmations of the commitment back to all involved Participant Nodes.
Completion
The user gets back a confirmation through the Ledger API of the submitting Participant Node.

The first important consequence of the above is that all transactions are committed atomically. Either a transaction is committed as a whole and for all participants, or it fails.

That’s important in the context of the Trade_Settle choice shown above. The choice transfers a baseAsset one way and a quoteAsset the other way. Thanks to transaction atomicity, there is no chance that either party is left out of pocket.

The second consequence is that the requester of a transaction knows all consequences of their submitted transaction – there are no surprises in Daml. However, it also means that the requester must have all the information to interpret the transaction. We also refer to this as Principle 2 a bit later on this page.

That’s also important in the context of Trade. In order to allow Bob to interpret a transaction that transfers Alice’s cash to Bob, Bob needs to know both about Alice’s Asset contract, as well as about some way for Alice to accept a transfer – remember, accepting a transfer needs the authority of issuer in this example.

Observers

Observers are Daml’s mechanism to disclose contracts to other parties. They are declared just like signatories, but using the observer keyword, as shown in the Asset template:

template Asset
  with
    issuer : Party
    owner : Party
    symbol : Text
    quantity : Decimal
    observers : [Party]
  where
    signatory issuer, owner
    ensure quantity > 0.0

    observer observers

The Asset template also gives the owner a choice to set the observers, and you can see how Alice uses it to show her Asset to Bob just before proposing the trade. You can try out what happens if she didn’t do that by removing that transaction.

  usdCid <- submit alice do
    exerciseCmd usdCid SetObservers with
      newObservers = [bob]

Observers have guarantees in Daml. In particular, they are guaranteed to see actions that create and archive the contract on which they are an observer.

Since observers are calculated from the arguments of the contract, they always know about each other. That’s why, rather than adding Bob as an observer on Alice’s AssetHolder contract, and using that to authorize the transfer in Trade_Settle, Alice creates a one-time authorization in the form of a TransferAuthorization. If Alice had lots of counterparties, she would otherwise end up leaking them to each other.

Controllers declared in the choice syntax are not automatically made observers, as they can only be calculated at the point in time when the choice arguments are known. On the contrary, controllers declared via the controller cs can syntax are automatically made observers, but this syntax is deprecated and will be removed in a future version of Daml.

Privacy

Daml’s privacy model is based on two principles:

Principle 1. Parties see those actions that they have a stake in. Principle 2. Every party that sees an action sees its (transitive) consequences.

Principle 2 is necessary to ensure that every party can independently verify the validity of every transaction they see.

A party has a stake in an action if

  • they are a required authorizer of it
  • they are a signatory of the contract on which the action is performed
  • they are an observer on the contract, and the action creates or archives it

What does that mean for the exercise tradeCid Trade_Settle action from test_trade?

Alice is the signatory of tradeCid and Bob a required authorizer of the Trade_Settled action, so both of them see it. According to rule 2. above, that means they get to see everything in the transaction.

The consequences contain, next to some fetch actions, two exercise actions of the choice TransferApproval_Transfer.

Each of the two involved TransferApproval contracts is signed by a different issuer, which see the action on “their” contract. So the EUR_Bank sees the TransferApproval_Transfer action for the EUR Asset and the USD_Bank sees the TransferApproval_Transfer action for the USD Asset.

Some Daml ledgers, like the script runner and the Sandbox, work on the principle of “data minimization”, meaning nothing more than the above information is distributed. That is, the “projection” of the overall transaction that gets distributed to EUR_Bank in step 4 of Daml’s execution model would consist only of the TransferApproval_Transfer and its consequences.

Other implementations, in particular those on public blockchains, may have weaker privacy constraints.

Divulgence

Note that Principle 2 of the privacy model means that sometimes parties see contracts that they are not signatories or observers on. If you look at the final ledger state of the test_trade script, for example, you may notice that both Alice and Bob now see both assets, as indicated by the Xs in their respective columns:

The table as described above.

This is because the create action of these contracts are in the transitive consequences of the Trade_Settle action both of them have a stake in. This kind of disclosure is often called “divulgence” and needs to be considered when designing Daml models for privacy sensitive applications.

Next up

In 8 Exception Handling, we will learn about how errors in your model can be handled in Daml.