5 Adding constraints to a contract

You will often want to constrain the data stored or the allowed data transformations in your contract models. In this section, you will learn about the two main mechanisms provided in DAML:

  • The ensure keyword.
  • The assert, abort and error keywords.

To make sense of the latter, you’ll also learn more about the Update and Scenario types and do blocks, which will be good preparation for 7 Composing choices, where you will use do blocks to compose choices into complex transactions.

Lastly, you will learn about time on the ledger and in scenarios.

Hint

Remember that you can load all the code for this section into a folder called 5_Restrictions by running daml new 5_Restrictions daml-intro-5

Template preconditions

The first kind of restriction you may want to put on the contract model are called template pre-conditions. These are simply restrictions on the data that can be stored on a contract from that template.

Suppose, for example, that the SimpleIou contract from A simple cash model should only be able to store positive amounts. You can enforce this using the ensure keyword:

template SimpleIou
  with
    issuer : Party
    owner : Party
    cash : Cash
  where
    signatory issuer

    ensure cash.amount > 0.0

The ensure keyword takes a single expression of type Bool. If you want to add more restrictions, use logical operators &&, || and not to build up expressions. The below shows the additional restriction that currencies are three capital letters:

        && T.length cash.currency == 3
        && T.isUpper cash.currency

Hint

The T here stands for the DA.Text standard library which has been imported using import DA.Text as T.

test_restrictions = scenario do
  alice <- getParty "Alice"
  bob <- getParty "Bob"
  dora <- getParty "Dora"

  -- Dora can't issue negative Ious.
  submitMustFail dora do
    create SimpleIou with
      issuer = dora
      owner = alice
      cash = Cash with
        amount = -100.0
        currency = "USD"

  -- Or even zero Ious.
  submitMustFail dora do
    create SimpleIou with
      issuer = dora
      owner = alice
      cash = Cash with
        amount = 0.0
        currency = "USD"

  -- Nor positive Ious with invalid currencies.
  submitMustFail dora do
    create SimpleIou with
      issuer = dora
      owner = alice
      cash = Cash with
        amount = 100.0
        currency = "Swiss Francs"

  -- But positive Ious still work, of course.
  iou <- submit dora do
    create SimpleIou with
      issuer = dora
      owner = alice
      cash = Cash with
        amount = 100.0
        currency = "USD"

Assertions

A second common kind of restriction is one on data transformations.

For example, the simple Iou in A simple cash model allowed the no-op where the owner transfers to themselves. You can prevent that using an assert statement, which you have already encountered in the context of scenarios.

assert does not return an informative error so often it’s better to use the function assertMsg, which takes a custom error message:

    controller owner can
      Transfer
        : ContractId SimpleIou
        with
          newOwner : Party
        do
          assertMsg "newOwner cannot be equal to owner." (owner /= newOwner)
          create this with owner = newOwner
  -- Alice can't transfer to herself...
  submitMustFail alice do
    exercise iou Transfer with
      newOwner = alice

  -- ... but can transfer to Bob.
  iou2 <- submit alice do
    exercise iou Transfer with
      newOwner = bob

Similarly, you can write a Redeem choice, which allows the owner to redeem an Iou during business hours on weekdays. The choice doesn’t do anything other than archiving the SimpleIou. (This assumes that actual cash changes hands off-ledger.)

    controller owner can
      Redeem
        : ()
        do
          now <- getTime
          let
            today = toDateUTC now
            dow = dayOfWeek today
            timeofday = now `subTime` time today 0 0 0
            hrs = convertRelTimeToMicroseconds timeofday / 3600000000
          assertMsg
            ("Cannot redeem outside business hours. Current time: " <> show timeofday)
            (hrs >= 8 && hrs <= 18)
          case dow of
            Saturday -> abort "Cannot redeem on a Saturday."
            Sunday -> abort "Cannot redeem on a Sunday."
            _ -> return ()
  -- June 1st 2019 is a Saturday.
  passToDate (date 2019 Jun 1)
  -- Bob cannot redeem on a Saturday.
  submitMustFail bob do
    exercise iou2 Redeem

  -- Not even at mid-day.
  pass (hours 12)
  -- Bob cannot redeem on a Saturday.
  submitMustFail bob do
    exercise iou2 Redeem

  -- Bob also cannot redeem at 6am on a Monday.
  pass (hours 42)
  submitMustFail bob do
    exercise iou2 Redeem

  -- Bob can redeem at 8am on Monday.
  pass (hours 2)
  submit bob do
    exercise iou2 Redeem

There are quite a few new time-related functions from the DA.Time and DA.Date libraries here. Their names should be reasonably descriptive so how they work won’t be covered here, but given that DAML assumes it is run in a distributed setting, we will still discuss time in DAML.

There’s also quite a lot going on inside the do block of the Redeem choice, with several uses of the <- operator. do blocks and <- deserve a proper explanation at this point.

Time on DAML ledgers

Each transaction on a DAML ledger has two timestamps called the ledger time (LT) and the record time (RT). The ledger time is set by the participant, the record time is set by the ledger.

Each DAML ledger has a policy on the allowed difference between LT and RT called the skew. The participant has to take a good guess at what the record time will be. If it’s too far off, the transaction will be rejected.

getTime is an action that gets the LT from the ledger. In the above example, that time is taken apart into day of week and hour of day using standard library functions from DA.Date and DA.Time. The hour of the day is checked to be in the range from 8 to 18.

Consider the following example: Suppose that the ledger had a skew of 10 seconds. At 17:59:55, Alice submits a transaction to redeem an Iou. One second later, the transaction is assigned a LET of 17:59:56, but then takes 10 seconds to commit and is recorded on the ledger at 18:00:06. Even though it was committed after business hours, it would be a valid transaction and be committed successfully as getTime will return 17:59:56 so hrs == 17. Since the RT is 18:00:06, LT - RT <= 10 seconds and the transaction won’t be rejected.

Time therefore has to be considered slightly fuzzy in DAML, with the fuzziness depending on the skew parameter.

For details, see Background concepts - time.

Time in scenarios

In scenarios, record and ledger time are always equal. You can set them using the following functions:

  • passToDate, which takes a date and sets the time to midnight (UTC) of that date
  • pass, which takes a RelTime (a relative time) and moves the ledger by that much

Time on ledgers

On a distributed DAML ledger, there are no guarantees that ledger time or record time are strictly increasing. The only guarantee is that ledger time is increasing with causality. That is, if a transaction TX2 depends on a transaction TX1, then the ledger enforces that the LT of TX2 is greater than or equal to that of TX1:

  iou3 <- submit dora do
    create SimpleIou with
      issuer = dora
      owner = alice
      cash = Cash with
        amount = 100.0
        currency = "USD"

  pass (days (-3))
  submitMustFail alice do
    exercise iou3 Redeem

Actions and do blocks

You have come across do blocks and <- notations in two contexts by now: Scenario and Update. Both of these are examples of an Action, also called a Monad in functional programming. You can construct Actions conveniently using do notation.

Understanding Actions and do blocks is therefore crucial to being able to construct correct contract models and test them, so this section will explain them in some detail.

Pure expressions compared to Actions

Expressions in DAML are pure in the sense that they have no side-effects: they neither read nor modify any external state. If you know the value of all variables in scope and write an expression, you can work out the value of that expression on pen and paper.

However, the expressions you’ve seen that used the <- notation are not like that. For example, take getTime, which is an Action. Here’s the example we used earlier:

getTime is a good example of an Action. Here’s the example we used earlier

now <- getTime

You cannot work out the value of now based on any variable in scope. To put it another way, there is no expression expr that you could put on the right hand side of now = expr. To get the ledger time, you must be in the context of a submitted transaction, and then look at that context.

Similarly, you’ve come across fetch. If you have cid : ContractId Account in scope and you come across the expression fetch cid, you can’t evaluate that to an Account so you can’t write account = fetch cid. To do so, you’d have to have a ledger you can look that contract ID up on.

Actions and impurity

Actions are a way to handle such “impure” expressions. Action a is a type class with a single parameter a, and Update and Scenario are instances of Action. A value of such a type m a where m is an instance of Action can be interpreted as “a recipe for an action of type m, which, when executed, returns a value a”.

You can always write a recipe using just pen and paper, but you can’t cook it up unless you are in the context of a kitchen with the right ingredients and utensils. When cooking the recipe you have an effect – you change the state of the kitchen – and a return value – the thing you leave the kitchen with.

  • An Update a is “a recipe to update a DAML ledger, which, when committed, has the effect of changing the ledger, and returns a value of type a”. An update to a DAML ledger is a transaction so equivalently, an Update a is “a recipe to construct a transaction, which, when executed in the context of a ledger, returns a value of type a”.
  • A Scenario a is “a recipe for a test, which, when performed against a ledger, has the effect of changing the ledger in ways analogous to those available via the API, and returns a value of type a”.

Expressions like getTime, getParty party, pass time, submit party update, create contract and exercise choice should make more sense in that light. For example:

  • getTime : Update Time is the recipe for an empty transaction that also happens to return a value of type Time.
  • pass (days 10) : Scenario () is a recipe for a transaction that doesn’t submit any transactions, but has the side-effect of changing the LET of the test ledger. It returns (), also called Unit and can be thought of as a zero-tuple.
  • create iou : Update (ContractId Iou), where iou : Iou is a recipe for a transaction consisting of a single create action, and returns the contract id of the created contract if successful.
  • submit alice (create iou) : Scenario (ContractId Iou) is a recipe for a scenario in which Alice evaluates the result of create iou to get a transaction and a return value of type ContractId Iou, and then submits that transaction to the ledger.

Any DAML ledger knows how to perform actions of type Update a. Only some know how to run scenarios, meaning they can perform actions of type Scenario a.

Chaining up actions with do blocks

An action followed by another action, possibly depending on the result of the first action, is just another action. Specifically:

  • A transaction is a list of actions. So a transaction followed by another transaction is again a transaction.
  • A scenario is a list of interactions with the ledger (submit, getParty, pass, etc). So a scenario followed by another scenario is again a scenario.

This is where do blocks come in. do blocks allow you to build complex actions from simple ones, using the results of earlier actions in later ones.

sub_scenario1 : Scenario (ContractId SimpleIou) = scenario do
  alice <- getParty "Alice"
  dora <- getParty "Dora"

  submit dora do
    create SimpleIou with
      issuer = dora
      owner = alice
      cash = Cash with
        amount = 100.0
        currency = "USD"

sub_scenario2 : Scenario Int = scenario do
  getParty "Nobody"
  pass (days 1)
  pass (days (-1))
  return 42

sub_scenario3 : Scenario (ContractId SimpleIou) = scenario do
  bob <- getParty "Bob"
  dora <- getParty "Dora"

  submit dora do
    create SimpleIou with
      issuer = dora
      owner = bob
      cash = Cash with
        amount = 100.0
        currency = "USD"

main_scenario : Scenario () = scenario do
  dora <- getParty "Dora"

  iou1 <- sub_scenario1
  sub_scenario2
  iou2 <- sub_scenario3

  submit dora do
    archive iou1
    archive iou2

Above, we see do blocks in action for both Scenario and Update.

Wrapping values in actions

You may already have noticed the use of return in the redeem choice. return x is a no-op action which returns value x so return 42 : Update Int. Since do blocks always return the value of their last action, sub_scenario2 : Scenario Int.

Failing actions

Not only are Update and Scenario examples of Action, they are both examples of actions that can fail, e.g. because a transaction is illegal or the party retrieved via getParty doesn’t exist on the ledger.

Each has a special action abort txt that represents failure, and that takes on type Update () or Scenario () depending on context .

Transactions and scenarios succeed or fail atomically as a whole. So an occurrence of an abort action will always fail the entire evaluation of the current Scenario or Update.

The last expression in the do block of the Redeem choice is a pattern matching expression on dow. It has type Update () and is either an abort or return depending on the day of week. So during the week, it’s a no-op and on weekends, it’s the special failure action. Thanks to the atomicity of transactions, no transaction can ever make use of the Redeem choice on weekends, because it fails the entire transaction.

A sample Action

If the above didn’t make complete sense, here’s another example to explain what actions are more generally, by creating a new type that is also an action. CoinGame a is an Action a in which a Coin is flipped. The Coin is a pseudo-random number generator and each flip has the effect of changing the random number generator’s state. Based on the Heads and Tails results, a return value of type a is calulated.

data Face = Heads | Tails
  deriving (Eq, Show, Enum)

data CoinGame a = CoinGame with
  play : Coin -> (Coin, a)

flipCoin : CoinGame Face
getCoin : Scenario Coin

A CoinGame a exposes a function play which takes a Coin and returns a new Coin and a result a. More on the -> syntax for functions later.

Coin and play are deliberately left obscure in the above. All you have is an action getCoin to get your hands on a Coin in a Scenario context and an action flipCoin which represents the simplest possible game: a single coin flip resulting in a Face.

You can’t play any CoinGame game on pen and paper as you don’t have a coin, but you can write down a script or recipe for a game:

coin_test = scenario do
  -- The coin is pseudo-random on LET so change the parameter to change the game.
  passToDate (date 2019 Jun 1)
  pass (seconds 2)
  coin <- getCoin
  let
    game = do
      f1r <- flipCoin
      f2r <- flipCoin
      f3r <- flipCoin

      if all (== Heads) [f1r, f2r, f3r]
        then return "Win"
        else return "Loss"
    (newCoin, result) = game.play coin

  assert (result == "Win")

The game expression is a CoinGame in which a coin is flipped three times. If all three tosses return Heads, the result is "Win", or else "Loss".

In a Scenario context you can get a Coin using the getCoin action, which uses the LET to calculate a seed, and play the game.

Somehow the Coin is threaded through the various actions. If you want to look through the looking glass and understand in-depth what’s going on, you can look at the source file to see how the CoinGame action is implemented, though be warned that the implementation uses a lot of DAML features we haven’t introduced yet in this introduction.

More generally, if you want to learn more about Actions (aka Monads), we recommend a general course on functional programming, and Haskell in particular. For example:

Errors

Above, you’ve learnt about assertMsg and abort, which represent (potentially) failing actions. Actions only have an effect when they are performed, so the following scenario succeeds or fails depending on the value of abortScenario:

nonPerformedAbort = scenario do
  let abortScenario = False
  let failingAction : Scenario () = abort "Foo"
  let successfulAction : Scenario () = return ()
  if abortScenario then failingAction else successfulAction

However, what about errors in contexts other than actions? Suppose we wanted to implement a function pow that takes an integer to the power of another positive integer. How do we handle that the second parameter has to be positive?

One option is to make the function explicitly partial by returning an Optional:

optPow : Int -> Int -> Optional Int
optPow base exponent
 | exponent == 0 = Some 1
 | exponent > 0 =
   let Some result = optPow base (exponent - 1)
   in Some (base * result)
 | otherwise = None

This is a useful pattern if we need to be able to handle the error case, but it also forces us to always handle it as we need to extract the result from an Optional. We can see the impact on convenience in the definition of the above function. In cases, like division by zero or the above function, it can therefore be preferrable to fail catastrophically instead:

errPow : Int -> Int -> Int
errPow base exponent
 | exponent == 0 = 1
 | exponent > 0 = base * errPow base (exponent - 1)
 | otherwise = error "Negative exponent not supported"

The big downside to this is that even unused errors cause failures. The following scenario will fail, because failingComputation is evaluated:

nonPerformedError = scenario do
  let causeError = False
  let failingComputation = errPow 1 (-1)
  let successfulComputation = errPow 1 1
  return if causeError then failingComputation else successfulComputation

error should therefore only be used in cases where the error case is unlikely to be encountered, and where explicit partiality would unduly impact usability of the function.

Next up

You can now specify a precise data and data-transformation model for DAML ledgers. In 6 Parties and authority, you will learn how to properly involve multiple parties in contracts, how authority works in DAML, and how to build contract models with strong guarantees in contexts with mutually distrusting entities.