Add 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 Script types and do blocks, which will be good preparation for 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 Daml Script.

Hint

Remember that you can load all the code for this section into a folder called intro5 by running daml new intro5 --template 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
    observer owner

    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 = do
  alice <- allocateParty "Alice"
  bob <- allocateParty "Bob"
  dora <- allocateParty "Dora"

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

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

  -- Nor positive Ious with invalid currencies.
  submitMustFail dora do
    createCmd 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
    createCmd 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 scripts.

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

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

  -- ... but can transfer to Bob.
  iou2 <- submit alice do
    exerciseCmd 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 Redeem choice implementation below confirms that getTime returns a value that is during business hours on weekdays. If all those checks pass, the choice does not do anything other than archive the SimpleIou. (This assumes that actual cash changes hands off-ledger:)

    choice Redeem
      : ()
      controller owner
      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 ()

In the above example, the 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. The day of week is checked to not be Saturday or Sunday.

The following example shows how the Redeem choice is exercised in a script:

  -- June 1st 2019 is a Saturday.
  setTime (time (date 2019 Jun 1) 0 0 0)
  -- Bob cannot redeem on a Saturday.
  submitMustFail bob do
    exerciseCmd iou2 Redeem

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

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

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

For the purposes of testing the Redeem choice, the above code sets and advances the ledger time with the setTime and passTime functions respectively. Exercising the choice should fail or should not fail depending on the day of week and the time of day. While that is straightforward, the issue of time on a Daml ledger is worthy of more discussion.

Time on Daml Ledgers

Each transaction on a Daml ledger has two timestamps: the ledger time (LT) and the record time (RT).

Ledger time (LT) is the time associated with a transaction in the ledger model, as determined by the participant. It is the time of a transaction from a business and application perspective. When you call getTime, it is the LT that is returned. The LT is used when reasoning about related transactions and commits. The LT can be compared with other LTs to guarantee model consistency. For example, LTs are used to enforce that no transaction depends on a contract that does not exist. This is the requirement known as “causal monotonicity.”

Record time (RT) is the time assigned by the persistence layer. It represents the time that the transaction is “physically” recorded. For example, “The backing database ledger has assigned the timestamp of such-and-such time to this transaction.” The only purpose of the RT is to ensure that transactions are being recorded in a timely manner.

Each Daml ledger has a policy on the allowed difference between LT and RT called the skew. A consistent zero-skew is not feasible because this is a distributed system. If it is too far off, the transaction will be rejected. This is the requirement known as “bounded skew.” The RT is not relevant beyond this determination of skew.

Returning to the theme of business hours, consider the following example: Suppose that the ledger had a skew of 10 seconds. At 17:59:55, just before the end of business hours, Alice submits a transaction to redeem an Iou. One second later, the transaction is assigned an LT of 17:59:56. However, there still may be a few seconds before the transaction is persisted to the underlying storage. For example, the transaction might be written in the underlying backing store at 18:00:06, after business hours. Because LT is within business hours and LT - RT <= 10 seconds, the transaction will not be rejected.

For details, see Background concepts - time.

Time in Test Scripts

For tests, you can set time using the following functions:

  • setTime, which sets the ledger time to the given time.
  • passTime, which takes a RelTime (a relative time) and moves the ledger by that much.

On a distributed Daml ledger, there are no guarantees that LT or RT 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.

The following script illustrates that idea by moving the logical time back by three days and then trying to exercise a choice on a contract that hasn’t been created yet. That fails, as you would hope.

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

  passTime (days (-3))
  submitMustFail alice do
    exerciseCmd iou3 Redeem

Actions and do Blocks

You have come across do blocks and <- notations in two contexts by now: Script 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:

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 Script 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 Script 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, allocateParty party, passTime time, submit party commands, 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.
  • passTime (days 10) : Script () is a recipe for a transaction that doesn’t submit any transactions, but has the side-effect of changing the LT 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 (createCmd iou) : Script (ContractId Iou) is a recipe for a script in which Alice sends the command createCmd iou to the ledger which produces a transaction and a return value of type ContractId Iou and returns that back to Alice.

Commands is a bit more restricted than Script and Update as it represents a list of independent commands sent to the ledger. You can still use do blocks but if you have more than one command in a single do block you need to enable the ApplicativeDo extension at the beginning of your file. In addition to that, the last statement in such a do block must be of the form return expr or pure expr. Applicative is a more restricted version of Action that enforces that there are no dependencies between commands. If you do have dependencies between commands, you can always wrap it in a choice in a helper template and call that via createAndExerciseCmd just like we did to call fetchByKey. Alternatively, if you do not need them to be part of the same transaction, you can make multiple calls to submit:

{-# LANGUAGE ApplicativeDo #-}
module Restrictions where

Chain 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 script is a list of interactions with the ledger (submit, allocateParty, passTime, etc). So a script followed by another script is again a script.

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_script1 (alice, dora) = do
  submit dora do
    createCmd SimpleIou with
      issuer = dora
      owner = alice
      cash = Cash with
        amount = 100.0
        currency = "USD"

sub_script2 = do
  passTime (days 1)
  passTime (days (-1))
  return 42

sub_script3 (bob, dora) = do
  submit dora do
    createCmd SimpleIou with
      issuer = dora
      owner = bob
      cash = Cash with
        amount = 100.0
        currency = "USD"

main_: Script () = do
  dora <- allocateParty "Dora"
  alice <- allocateParty "Alice"
  bob <- allocateParty "Bob"

  iou1 <- sub_script1 (alice, dora)
  sub_script2
  iou2 <- sub_script3 (bob, dora)

  submit dora do
    archiveCmd iou1
    archiveCmd iou2
    pure ()

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

Wrap 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_script2 : Script Int.

Failing Actions

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

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

Transactions succeed or fail atomically as a whole. Scripts on the other hand do not fail atomically: while each submit is atomic, if a submit succeeded and the script fails later, the effects of that submit will still be applied to the ledger.

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 calculated:

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

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

flipCoin : CoinGame Face
getCoin : Script 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 Script 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 = do
  -- The coin is pseudo-random on LT so change the parameter to change the game.
  setTime (time (date 2019 Jun 1) 0 0 0)
  passTime (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 Script context you can get a Coin using the getCoin action, which uses the LT 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. See The Haskell Connection for some suggestions.

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 script succeeds or fails depending on the value of abortScript:

nonPerformedAbort = do
  let abortScript = False
  let failingAction : Script () = abort "Foo"
  let successfulAction : Script () = return ()
  if abortScript 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 preferable 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 script will fail, because failingComputation is evaluated:

nonPerformedError = script 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 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.