Scilla in Depth¶
Structure of a Scilla Contract¶
The general structure of a Scilla contract is given in the code fragment below:
- The contract starts with the declaration of
scilla_version
, which indicates which major Scilla version the contract uses. - Then follows the declaration of a
library
that contains purely mathematical functions, e.g., a function to compute the BooleanAND
of two bits, or a function computing the factorial of a given natural number. - Then follows the actual contract definition declared using the
keyword
contract
. - Within a contract, there are then four distinct parts:
- The first part declares the immutable parameters of the contract.
- The second part describes the contract’s constraint, which must be valid when the contract is deployed.
- The third part declares the mutable fields.
- The fourth part contains all
transition
andprocedure
definitions.
(* Scilla contract structure *)
(***************************************************)
(* Scilla version *)
(***************************************************)
scilla_version 1
(***************************************************)
(* Associated library *)
(***************************************************)
library MyContractLib
(* Library code block follows *)
(***************************************************)
(* Contract definition *)
(***************************************************)
contract MyContract
(* Immutable contract parameters declaration *)
(vname_1 : vtype_1,
vname_2 : vtype_2)
(* Contract constraint *)
with
(* Constraint expression *)
=>
(* Mutable fields declaration *)
field vname_1 : vtype_1 = init_val_1
field vname_2 : vtype_2 = init_val_2
(* Transitions and procedures *)
(* Procedure signature *)
procedure firstProcedure (param_1 : type_1, param_2 : type_2)
(* Procedure body *)
end
(* Transition signature *)
transition firstTransition (param_1 : type_1, param_2 : type_2)
(* Transition body *)
end
(* Procedure signature *)
procedure secondProcedure (param_1 : type_1, param_2 : type_2)
(* Procedure body *)
end
transition secondTransition (param_1: type_1)
(* Transition body *)
end
Immutable Contract Parameters¶
Immutable parameters are the contract’s initial parameters whose values are defined when the contract is deployed, and cannot be modified afterwards.
Immutable parameters are declared using the following syntax:
(vname_1 : vtype_1,
vname_2 : vtype_2,
... )
Each declaration consists of a parameter name (an identifier) and
followed by its type, separated by :
. Multiple parameter
declarations are separated by ,
. The initialization values for
parameters are to be specified when the contract is deployed.
Note
In addition to the explicitly declared immutable parameters, a Scilla contract has the following implicitly declared immutable contract parameters
1.
_this_address
of typeByStr20
, which is initialised to the address of the contract when the contract is deployed.2.
_creation_block
of typeBNum
, which is initialized to the block number at which the contract is / was deployed.
These parameters can be freely read within the implementation without having to
dereference it using <-
and cannot be modified with :=
.
Contract Constraints¶
A contract constraint is a requirement placed on the contract’s immutable parameters. A contract constraint provides a way of establishing a contract invariant as soon as the contract is deployed, thus preventing the contract being deployed with nonsensical parameters.
A contract constraint is declared using the following syntax:
with
...
=>
The constraint must be an expression of type Bool
.
The constraint is checked when the contract is deployed. Contract
deployment only succeeds if the constraint evaluates to True
. If
it evaluates to False
, then the deployment fails.
Here is a simple example of using contract constraints to make sure a contract with a limited period of functioning is not deployed after that period:
contract Mortal(end_of_life : BNum)
with
builtin blt _creation_block end_of_life
=>
The snippet above uses the implicit contract parameter _creation_block
described in Immutable Contract Parameters.
Note
Declaring a contract constraint is optional. If no constraint is
declared, then the constraint is assumed to simply be True
.
Mutable Fields¶
Mutable fields represent the mutable state (mutable variables) of the
contract. They are declared after the immutable parameters, with each
declaration prefixed with the keyword field
.
field vname_1 : vtype_1 = expr_1
field vname_2 : vtype_2 = expr_2
...
Each expression here is an initialiser for the field in question. The definitions complete the initial state of the contract, at the time of creation. As the contract executes a transition, the values of these fields get modified.
Note
In addition to the explicitly declared mutable fields, a Scilla
contract has an implicitly declared mutable field _balance
of
type Uint128
, which is initialised to 0 when the contract is
deployed. The _balance
field keeps the amount of funds held by
the contract, measured in QA (1 ZIL = 1,000,000,000,000 QA). This
field can be freely read within the implementation, but can only be
modified by explicitly transferring funds to other accounts (using
send
), or by accepting money from incoming messages (using
accept
).
Note
Both mutable fields and immutable parameters must be of a storable type:
- Messages, events and the special
Unit
type are not storable. All other primitive types like integers and strings are storable. - Function types are not storable.
- Complex types involving uninstantiated type variables are not storable.
- Maps and ADT are storable if the types of their subvalues are storable. For maps this means that the key type and the value type must both be storable, and for ADTs this means that the type of every constructor argument must be storable.
Units¶
The Zilliqa protocol supports three basic tokens units - ZIL, LI (10^-6 ZIL) and QA (10^-12 ZIL).
The base unit used in Scilla smart contracts is QA. Hence, when using money variables, it is important to attach the trailing zeroes that are needed to represent it in QAs.
(* fee is 1 QA *) let fee = Uint128 1 (* fee is 1 LI *) let fee = Uint128 1000000 (* fee is 1 ZIL *) let fee = Uint128 1000000000000
Transitions¶
Transitions are a way to define how the state of the contract may change. The transitions of a contract define the public interface for the contract, since transitions may be invoked by sending a message to the contract.
Transitions are defined with the keyword transition
followed
by the parameters to be passed. The definition ends with the end
keyword.
transition foo (vname_1 : vtype_1, vname_2 : vtype_2, ...)
...
end
where vname : vtype
specifies the name and type of each parameter and
multiple parameters are separated by ,
.
Note
In addition to the parameters that are explicitly declared in the definition, each transition has the following implicit parameters:
_amount : Uint128
: Incoming amount, in QA (see section above on the units), sent by the sender. To transfer the money from the sender to the contract, the transition must explicitly accept the money using theaccept
instruction. The money transfer does not happen if the transition does not execute anaccept
._sender : ByStr20 with end
: The account address that triggered this transition. If the transition was called by a contract account instead of a user account, then_sender
is the address of the contract that called this transition. In a chain call, this is the contract that sent the message invoking the current transition._origin : ByStr20 with end
: The account address that initiated the current transaction (which can possibly be a chain call). This is always a user address, since contracts can never initiate transactions.
The type ByStr20 with end
is an address type. Address types are explained
in detail in the Addresses section.
Note
Transition parameters must be of a serialisable type.
Serialisable:
- Byte strings, integers, strings, and block numbers are serialisable.
- Addresses are serialisable only as
ByStr20
values. - ADT are serialisable if the types of their subvalues are serialisable. This means that the type of every constructor argument must be serialisable.
Not Serialisable:
- Messages, events and the special
Unit
type are not serialisable. - Function types and map types are not serialisable.
- Complex types involving uninstantiated type variables are not serialisable.
Procedures¶
Procedures are another way to define now the state of the contract may change, but in contrast to transitions, procedures are not part of the public interface of the contract, and may not be invoked by sending a message to the contract. The only way to invoke a procedure is to call it from a transition or from another procedure.
Procedures are defined with the keyword procedure
followed
by the parameters to be passed. The definition ends with the end
keyword.
procedure foo (vname_1 : vtype_1, vname_2 : vtype_2, ...)
...
end
where vname : vtype
specifies the name and type of each parameter and
multiple parameters are separated by ,
.
Once a procedure is defined it is available to be invoked from transitions and procedures in the rest of the contract file. It is not possible to invoke a procedure from transition or procedure defined earlier in the contract, nor is it possible for a procedure to call itself recursively.
Procedures are invoked using the name of the procedure followed by the actual arguments to the procedure:
v1 = ...;
v2 = ...;
foo v1 v2;
All arguments must be supplied when the procedure is invoked. A procedure does not return a result.
Note
The implicit transition parameters _sender
, _origin
and _amount
are
implicitly passed to all the procedures that a transition
calls. There is therefore no need to declare those parameters
explicitly when defining a procedure.
Note
Procedure parameters cannot be (or contain) maps. If a procedure needs to access a map, it is therefore necessary to either make the procedure directly access the contract field containing the map, or use a library function to perform the necessary computations on the map.
Expressions¶
Expressions handle pure operations. Scilla contains the following types of expressions:
let x = f
: Givef
the namex
in the contract. The binding ofx
tof
is global and extends to the end of the contract. The following code fragment defines a constantone
whose values is1
of typeInt32
throughout the contract.let one = Int32 1
let x = f in expr
: Bindf
to the namex
within expressionexpr
. The binding here is local toexpr
only. The following example binds the value ofone
to1
of typeInt32
andtwo
to2
of typeInt32
in the expressionbuiltin add one two
, which adds1
to2
and hence evaluates to3
of typeInt32
.let sum = let one = Int32 1 in let two = Int32 2 in builtin add one two
{ <entry>_1 ; <entry>_2 ... }
: Message or event expression, where each entry has the following form:b : x
. Hereb
is an identifier andx
a variable, whose value is bound to the identifier in the message.fun (x : T) => expr
: A function that takes an inputx
of typeT
and returns the value to which expressionexpr
evaluates.f x
: Apply the functionf
to the parameterx
.tfun 'T => expr
: A type function that takes'T
as a parametric type and returns the value to which expressionexpr
evaluates. These are typically used to build library functions. See the implementation of fst for an example.Note
Shadowing of type variables is not currently allowed. E.g.
tfun 'T => tfun 'T => expr
is not a valid expression.@x T
: Apply the type functionx
to the typeT
. This specialises the type functionx
by instantiating the first type variable ofx
toT
. Type applications are typically used when a library function is about to be applied. See the example application of fst for an example.builtin f x
: Apply the built-in functionf
onx
.match
expression: Matches a bound variable with patterns and evaluates the expression in that clause. Thematch
expression is similar to thematch
expression in OCaml. The pattern to be matched can be an ADT constructor (see ADTs) with subpatterns, a variable, or a wildcard_
. An ADT constructor pattern matches values constructed with the same constructor if the subpatterns match the corresponding subvalues. A variable matches anything, and binds the variable to the value it matches in the expression of that clause. A wildcard matches anything, but the value is then ignored.match x with | pattern_1 => expression_1 ... | pattern_2 => expression_2 ... | _ => (*Wildcard*) expression ... end
Note
A pattern-match must be exhaustive, i.e., every legal (type-safe) value of
x
must be matched by a pattern. Additionally, every pattern must be reachable, i.e., for each pattern there must be a legal (type-safe) value ofx
that matches that pattern, and which does not match any pattern preceding it.
Statements¶
Statements in Scilla are operations with effect, and hence not purely mathematical. Scilla contains the following types of statements:
x <- f
: Fetch the value of the contract fieldf
, and store it into the local variablex
.f := x
: Update the mutable contract fieldf
with the value ofx
.x
may be a local variable, or another contract field.x <- & BLOCKNUMBER
: Fetch the value of the blockchain state variableBLOCKNUMBER
, and store it into the local variablex
.x <- & c.f
: Remote fetch. Fetch the value of the contract fieldf
at addressc
, and store it into the local variablex
. Note that the type ofc
must be an address type containing the fieldf
. See the secion on Addresses for details on address types.x <- & c as t
: Address type cast. Check whetherc
satisfies the typet
. If it does, then store the valueSome v
intox
, wherev
is the same value asc
, but of typet
rather than the type ofc
. If it does not, then store the valueNone
intox
. Note thatc
must be of typeByStr20
or an address type, and thatt
must be an address type. See the section on Addresses for details on address types.v = e
: Evaluate the expressione
, and assign the value to the local variablev
.p x y z
: Invoke the procedurep
with the argumentsx
,y
andz
. The number of arguments supplied must correspond to the number of arguments the procedure takes.forall ls p
: Invoke procedurep
for each element in the listls
.p
should be defined to take exactly one argument whose type is equal to an element of the listls
.match
: Pattern-matching at statement level:match x with | pattern_1 => statement_11; statement_12; ... | pattern_2 => statement_21; statement_22; ... | _ => (*Wildcard*) statement_n1; statement_n2; ... end
accept
: Accept the QA of the message that invoked the transition. The amount is automatically added to the_balance
field of the contract. If a message contains QA, but the invoked transition does not accept the money, the money is transferred back to the sender of the message. Not accepting the incoming amount (when it is non-zero) is not an error.send
andevent
: Communication with the blockchain. See the next section for details.In-place map operations : Operations on contract fields of type
Map
. See the Maps section for details.
A sequence of statements must be separated by semicolons ;
:
transition T ()
statement_1;
statement_2;
...
statement_n
end
Notice that the final statement does not have a trailing ;
, since
;
is used to separate statements rather than terminate them.
Communication¶
A contract can communicate with other contract and user accounts
through the send
instruction:
send msgs
: Send a list of messagesmsgs
.The following code snippet defines a
msg
with four entries_tag
,_recipient
,_amount
andparam
.(*Assume contractAddress is the address of the contract being called and the contract contains the transition setHello*) msg = { _tag : "setHello"; _recipient : contractAddress; _amount : Uint128 0; param : Uint32 0 };
A message passed to send
must contain the mandatory fields
_tag
, _recipient
and _amount
.
The _recipient
field (of type ByStr20
) is the blockchain
address that the message is to be sent to, and the _amount
field
(of type Uint128
) is the number of QA to be transferred to that
account.
The _tag
field (of type String
) is only used when the value of
the _recipient
field is the address of a contract. In this case,
the value of the _tag
field is the name of the transition that is
to be invoked on the recipient contract. If the recipient is a user
account, the _tag
field is ignored.
Note
To make it possible to transfer funds from a contract to both contracts and
user accounts, use a standard transition name as per ZRC-5, i.e.
AddFunds
. Please make sure to check if a contract to which you intend to
send funds is implemented in adherence with ZRC-5 convention.
In addition to the compulsory fields the message may contain other
fields (of any type), such as param
above. However, if the message
recipient is a contract, the additional fields must have the same
names and types as the parameters of the transition being invoked on
the recipient contract.
Here’s an example that sends multiple messages.
msg1 = { _tag : "setFoo"; _recipient : contractAddress1; _amount : Uint128 0; foo : Uint32 101 }; msg2 = { _tag : "setBar"; _recipient : contractAddress2; _amount : Uint128 0; bar : Uint32 100 }; msgs = let nil = Nil {Message} in let m1 = Cons {Message} msg1 nil in Cons msg2 m1 ; send msgs
Note
A transition may execute a send
at any point during execution
(including during the execution of the procedures it invokes), but
the messages will not sent onwards until after the transition has
finished. More details can be found in the Chain Calls section.
Note
Messages sent to a user-defined library are ignored. They do not cause the transaction to fail, but they have no effect.
A contract can also communicate to the outside world by emitting events. An event is a signal that gets stored on the blockchain for everyone to see. If a user uses a client application invoke a transition on a contract, the client application can listen for events that the contract may emit, and alert the user.
event e
: Emit a messagee
as an event. The following code emits an event with namee_name
.
e = { _eventname : "e_name"; <entry>_2 ; <entry>_3 }; event e
An emitted event must contain the compulsory field _eventname
(of
type String
), and may contain other entries as well. The value of
the _eventname
entry must be a string literal. All events with the
same name must have the same entry names and types.
Note
As with the sending of messages, a transition may emit events at any point during execution (including during the execution of the procedures it invokes), but the event will not be visible on the blockchain before the transition has finished. More details can be found in the Chain Calls section.
Run-time Errors¶
A transition may encounter a run-time error during execution, such as out-of-gas errors, integer overflows, or deliberately thrown exceptions. A run-time error causes the transition to terminate abruptly, and the entire transaction to be aborted. However, gas is still charged up until the point of the error.
The syntax for throwing an exception is similar to that of events and messages.
e = { _exception : "InvalidInput"; <entry>_2; <entry>_3 };
throw e
Unlike that for event
or send
, The argument to throw
is optional
and can be omitted. An empty throw will result in an error that just conveys
the location of where the throw
happened without more information.
Note
If a run-time error occurs during the execution of a transition, then the entire transaction is aborted, and any state changes in both the current and other contracts are rolled back. (The state of other contracts may have changed due to a chain call).
In particular:
- All transferred funds are returned to their respective senders,
even if an
accept
was executed before the error. - The message queue is cleared, so that as yet unprocessed messages
will no longer be sent onwards even if a
send
was executed before the error. - The event list is cleared, so that no events are emitted even if
an
event
was executed before the error.
Gas is still charged for the transaction up until the point the run-time error occurs.
Note
Scilla does not have exception handlers. Throwing an exception always aborts the entire transaction.
Gas consumption in Scilla¶
Deploying contracts and executing transitions in them cost gas. The detailed cost mechanism is explained here.
The Nucleus Wallet page can be used to estimate gas costs for some transactions .
Primitive Data Types & Operations¶
Integer Types¶
Scilla defines signed and unsigned integer types of 32, 64, 128, and
256 bits. These integer types can be specified with the keywords
IntX
and UintX
where X
can be 32, 64, 128, or 256. For
example, the type of an unsigned integer of 32 bits is Uint32
.
The following code snippet declares a variable of type Uint32
:
let x = Uint32 43
Scilla supports the following built-in operations on integers. Each
operation takes two integers IntX
/ UintX
(of the same type) as
arguments. Exceptions are pow
whose second argument is always
Uint32
and isqrt
which takes in a single UintX
argument.
builtin eq i1 i2
: Isi1
equal toi2
? Returns aBool
.builtin add i1 i2
: Add integer valuesi1
andi2
. Returns an integer of the same type.builtin sub i1 i2
: Subtracti2
fromi1
. Returns an integer of the same type.builtin mul i1 i2
: Integer product ofi1
andi2
. Returns an integer of the same type.builtin div i1 i2
: Integer division ofi1
byi2
. Returns an integer of the same type.builtin rem i1 i2
: The remainder of integer division ofi1
byi2
. Returns an integer of the same type.builtin lt i1 i2
: Isi1
less thani2
? Returns aBool
.builtin pow i1 i2
:i1
raised to the power ofi2
. Returns an integer of the same type asi1
.builtin isqrt i
: Computes the integer square root ofi
, i.e. the largest integerj
such thatj * j <= i
. Returns an integer of the same type asi
.builtin to_nat i1
: Convert a value of typeUint32
to the equivalent value of typeNat
.builtin to_(u)int32/64/128/256
: Convert aUintX
/IntX
or aString
(that represents a decimal number) value to the result ofOption UintX
orOption IntX
type. ReturnsSome res
if the conversion succeeded andNone
otherwise. The conversion may fail when- there is not enough bits to represent the result;
- when converting a negative integer (or a string representing a negative integer) into a value of an unsigned type;
- the input string cannot be parsed as an integer.
Here is the list of concrete conversion builtins for better discoverability:
to_int32
,to_int64
,to_int128
,to_int256
,to_uint32
,to_uint64
,to_uint128
,to_uint256
.
Addition, subtraction, multiplication, pow, division and remainder operations may raise integer overflow, underflow and division_by_zero errors. This aborts the execution of the current transition and unrolls all the state changes made so far.
Note
Variables related to blockchain money, such as the _amount
entry
of a message or the _balance
field of a contract, are of type
Uint128
.
Strings¶
String
literals in Scilla are expressed using a sequence of
characters enclosed in double quotes. Variables can be declared by
specifying using keyword String
.
The following code snippet declares a variable of type String
:
let x = "Hello"
Scilla supports the following built-in operations on strings:
builtin eq s1 s2
: Iss1
equal tos2
? Returns aBool
.s1
ands2
must be of typeString
.builtin concat s1 s2
: Concatenate strings1
with strings2
. Returns aString
.builtin substr s idx len
: Extract the substring ofs
of lengthlen
starting from positionidx
.idx
andlen
must be of typeUint32
. Character indices in strings start from0
. Returns aString
or fails with a runtime error if the combination of the input parameters results in an invalid substring.builtin to_string x
: Convertx
to a string literal. Valid types ofx
areIntX
,UintX
,ByStrX
andByStr
. Returns aString
. Byte strings are converted to textual hexadecimal representation.builtin strlen s
: Calculate the length ofs
(of typeString
). Returns aUint32
.builtin strrev s
: Returns the reverse of the strings
.builtin to_ascii h
: Reinterprets a byte string (ByStr
orByStrX
) as a printable ASCII string and returns an equivalentString
value. If the byte string contains any non-printable characters, a runtime error is raised.
Byte strings¶
Byte strings in Scilla are represented using the types ByStr
and
ByStrX
, where X
is a number. ByStr
refers to a byte string of
arbitrary length, whereas for any X
, ByStrX
refers to a byte
string of fixed length X
. For instance, ByStr20
is the type of
byte strings of length 20, ByStr32
is the type of byte strings of
length 32, and so on.
Byte strings literals in Scilla are written using hexadecimal
characters prefixed with 0x
. Note that it takes 2 hexadecimal
characters to specify 1 byte, so a ByStrX
literal requires 2 *
X
hexadecimal characters. The following code snippet declares a
variable of type ByStr32
:
let x = 0x123456789012345678901234567890123456789012345678901234567890abff
Scilla supports the following built-in operations for computing on and converting between byte string types:
builtin to_bystr h
: Convert a valueh
of typeByStrX
(for some knownX
) to one of arbitrary length of typeByStr
.builtin to_bystrX h
: (note thatX
is a numerical paratemeter here and not a part of the builtin name, see the examples below)- if the argument
h
is of typeByStr
: Convert an arbitrary size byte string valueh
(of typeByStr
) to a fixed sized byte string of typeByStrX
, with lengthX
. The result is of typeOption ByStrX
in this case: the builtin returnsSome res
if the length of the argument is equal toX
andNone
otherwise. E.g.builtin to_bystr42 bs
returnsSome bs'
if the length ofbs
is 42. - if the argument
h
is of typeUint(32/64/128/256)
: Convert unsigned integers to their big endian byte representation, returning aByStr(4/8/16/32)
value (notice it’s not an optional type in this case). For instance,builtin to_bystr4 x
(this only typechecks ifx
has typeUint32
) orbuiltin to_bystr16 x
(this only typechecks ifx
is of typeUint128
).
- if the argument
builtin to_uint(32/64/128/256) h
: Convert a fixed sized byte string valueh
to an equivalent value of typeUint(32/64/128/256)
.h
must be of typeByStrX
for some knownX
less than or equal to (4/8/16/32). A big-endian representation is assumed.builtin concat h1 h2
: Concatenate byte stringsh1
andh2
.- If
h1
has typeByStrX
andh2
has typeByStrY
, then the result will have typeByStr(X+Y)
. - If the arguments are of type
ByStr
, the result is also of typeByStr
.
- If
builtin strlen h
: The length of byte string (ByStr
)h
. ReturnsUint32
.eq a1 a2
: Isa1
equal toa2
? Returns aBool
.
Addresses¶
Addresses on the Zilliqa network are strings of 20 bytes, and raw
addresses are therefore represented by values of type ByStr20
.
Additionally, Scilla supports structured address types, i.e., types
that are equivalent to ByStr20
, but which, when interpreted as an
address on the network, provide additional information about the
contents of that address. Address types are written
using the form ByStr20 with <address contents> end
, where
<address contents>
refers to what the address contains.
The hierarchy of address types is as follows:
ByStr20
: A raw byte string of length 20. The type does not provide any guarantee as to what is located at the address. (Generally,ByStr20
is not regarded as an address type, because it can refer to any byte string of length 20, whether it is meant to represent an address or not.)ByStr20 with end
: AByStr20
which, when interpreted as a network address, refers to an address that is in use. An address is in use if it either contains a contract or a library, or if the balance or the nonce of the address is greater than 0. (The balance of an address is the number of Qa held by the address account. The nonce of an address is the number of transactions that have been initiated from that address).ByStr20 with contract end
: AByStr20
which, when interpreted as a network address, refers to the address of a contract.ByStr20 with contract field f1 : t1, field f2 : t2, ... end
: AByStr20
which, when interpreted as a network address, refers to the address of a contract containing the mutable fieldsf1
of typet1
,f2
of typet2
, and so on. The contract in question may define more fields than the ones specified in the type, but the fields specified in the type must be defined in the contract.
Note
All addresses in use, and therefore by extension all contract
addresses, implicitly define a mutable field _balance :
Uint128
. For user accounts the _balance
field refers to the
account balance.
Note
Address types specifying immutable parameters or transitions of a contract are not supported.
Address subtyping¶
The hierarchy of address types defines a subtype relation:
- Any address type
ByStr20 with ... end
is as subtype ofByStr20
. This means that any address type can be used in place of aByStr20
, e.g., for comparing equality usingbuiltin eq
, or as the_recipient
value of a message. - Any contract address type
ByStr20 with contract ... end
is a subtype ofByStr20 with end
. - Any contract address type specifying explict fields
ByStr20 with contract field f1 : t11, field f2 : t12, ... end
is a subtype of a contract address type specifying a subset of those fieldsByStr20 with contract field f1 : t21, field f2 : t22, ... end
, provided thatt11
is a subtype oft21
,t12
is a subtype oft22
, and so on for each field specified in both contract types. - For ADTs with type parameters such as
List
orOption
, an ADTT t1 t2 ...
is a subtype ofS s1 s2 ...
ifT
is the same asS
, andt1
is a subtype ofs1
,t2
is a subtype ofs2
, and so on. - A map with key type
kt1
and value typevt1
is a subtype of another map with key typekt2
and value typevt2
ifkt1
is a subtype ofkt2
andvt1
is a subtype ofvt2
.
Dynamic typecheck of addresses¶
In general, address types cannot be fully typechecked statically by the Scilla checker. This can happen, e.g., because a byte string is a transition parameter and thus not known statically, or because a byte string refers to an address that does not currently contain a contract, but which might contain a contract in the future.
For this reason immutable parameters (i.e., contract parameters supplied when the contract is deployed) and transition parameters of address types are typechecked dynamically, when the actual byte string is known.
For example, a contract might specify an immutable field
init_owner
as follows:
contract MyContract (init_owner : ByStr20 with end)
When the contract is deployed, the byte string supplied as
init_owner
is looked up as an address on the blockchain, and if
the contents of that address matches the address type (in this
case that the address is in use either by a user or by a
contract), then deployment continues, and init_owner
can be
treated as a ByStr20 with end
throughout the contract.
Similarly, a transition might specify a parameter
token_contract
as follows:
transition Transfer (
token_contract : ByStr20 with contract
field balances : Map ByStr20 Uint128
end
)
When the transition is invoked, the byte string supplied as the
token_contract
parameter is looked up as an address on the
blockchain, and if the contents of that address matches the address
type (in this case that the address contains a contract with a
field balances
of a type that is assignable to Map ByStr20
Uint128
), then the transition parameter is initialised
successfully, and token_contract
can be treated as a ByStr20
with contract field balances : Map ByStr20 Uint128 end
throughout
this transition invocation.
In either case, if the contents of the address does not match the specified type, then the dynamic typecheck is unsuccessful, causing deployment (for failed immutable parameters) or transition invocation (for transition parameters) to fail. A failed dynamic typecheck is considered a run-time error, causing the current transaction to abort. (For the purposes of dynamic typechecks of immutable fields the deployment of a contract is considered a transaction).
It is also possible to perform a dynamic typecheck as a statement, using the address type cast construct:
x <- & c as ByStr20 with ... end
If c
satisfies the address type, then x
is bound to Some
v
, where v
is the same value as c
but is treated as having
the address type specified in cast. On the other hand, if c
does
not satisfy the address type, then x
is bound to None
.
The advantage of using an address type cast instead of a dynamic
typecheck of transition parameters is that a type cast does not cause
a run-time failure in case of a failed cast - a failed cast just
results in a None
value.
The disadvantage of address type casts is that they cost more gas. The type cast itself costs the same as a typecheck of a transition parameter, but one then needs to pattern-match on the result of the type cast, which costs additional gas.
Note
It is not possible to specify a ByStr20
literal and have it
interpreted as an address. In other words, the following code
snippet will result in a static type error:
let x : ByStr20 with end = 0x1234567890123456789012345678901234567890
The only ways for a byte string to be validated against an address type is either to pass it as the value of an immutable field or as a transition parameter of the appropriate type, or to perform a cast statement to the appropriate type.
Remote fetches¶
To perform a remote fetch x <- & c.f
, the type of c
must be
some address type declaring the field f
. For instance, if c
has the type ByStr20 with contract field paused : Bool end
, then
the value of the field paused
at address c
can be fetched
using the statement x <- & c.paused
, whereas it is not possible to
fetch the value of an undeclared field (e.g., admin
) of c
,
even if the contract at address c
does actually contain a field
admin
. To be able to fetch the value of the admin
field, the
type of c
must contain the admin
field as well, e.g, ByStr20
with contract field paused : Bool, field admin : ByStr20 end
Remote fetches of map fields can be performed using in-place
operations in the same way as for locally declared map fields, i.e.,
x <- & c.m[key]
, x <- & c.m[key1][key2]
, x <- & exists
m[key]
, etc. As with remote fetches of map fields, the remote map
field must be declared in the type of c
, e.g., ByStr20 with
contract field m : Map Uint128 (Map Uint32 Bool) end
.
Writing to a remote field is not allowed.
Crypto Built-ins¶
A hash in Scilla is declared using the data type ByStr32
. A
ByStr32
represents a hexadecimal byte string of 32 bytes (64
hexadecimal characters). A ByStr32
literal is prefixed with
0x
.
Scilla supports the following built-in operations on hashes and other cryptographic primitives, including byte sequences.
builtin eq h1 h2
: Ish1
equal toh2
? Both inputs are of the same typeByStrX
(or both are of typeByStr
). Returns aBool
.builtin sha256hash x
: Convertx
of any non-closure type to its SHA256 hash. Returns aByStr32
.builtin keccak256hash x
: Convertx
of any non-closure type to its Keccak256 hash. Returns aByStr32
.builtin ripemd160hash x
: Convertx
of any non-closure type to its RIPEMD-160 hash. Returns aByStr20
.builtin substr h idx len
: Extract the sub-byte-string ofh
of lengthlen
starting from positionidx
.idx
andlen
must be of typeUint32
. Character indices in byte strings start from0
. Returns aByStr
or fails with a runtime error.builtin strrev h
: Reverse byte string (eitherByStr
orByStrX
). Returns a value of the same type as the argument.builtin schnorr_verify pubk data sig
: Verify a signaturesig
of typeByStr64
against a byte stringdata
of typeByStr
with the Schnorr public keypubk
of typeByStr33
.builtin schnorr_get_address pubk
: Given a public key of typeByStr33
, returns theByStr20
Zilliqa address that corresponds to that public key.builtin ecdsa_verify pubk data sig
: Verify a signaturesig
of typeByStr64
against a byte stringdata
of typeByStr
with the ECDSA public keypubk
of typeByStr33
.builtin ecdsa_recover_pk data sig recid
: Recoverdata
(of typeByStr
), having signaturesig
(of typeByStr64
) and aUint32
recovery integerrecid
, whose value is restricted to be 0, 1, 2 or 3, the uncompressed public key, returning aByStr65
value.builtin bech32_to_bystr20 prefix addr
. The builtin takes a network specific prefix ("zil"
/"tzil"
) of typeString
and an input bech32 string (of typeString
) and if the inputs are valid, converts it to a raw byte address (ByStr20). The return type isOption ByStr20
. On success,Some addr
is returned and on invalid inputsNone
is returned.builtin bystr20_to_bech32 prefix addr
. The builtin takes a network specific prefix ("zil"
/"tzil"
) of typeString
and an inputByStr20
address, and if the inputs are valid, converts it to a bech32 address. The return type isOption String
. On success,Some addr
is returned and on invalid inputsNone
is returned.builtin alt_bn128_G1_add p1 p2
. The builtin takes two pointsp1
,p2
on thealt_bn128
curve and returns the sum of the points in the underlying group G1. The input points are each aPair {Bystr32 ByStr32}
. The return type isOption (Pair (ByStr32) (ByStr32))
. On success,Some point
is returned and on invalid inputsNone
is returned. Each scalar componentByStr32
of a point is a big-endian encoded number. Also see https://github.com/ethereum/EIPs/blob/master/EIPS/eip-196.mdbuiltin alt_bn128_G1_mul p s
. The builtin takes a pointp
on thealt_bn128
curve (as described previously), and a scalarByStr32
values
and returns the sum of the pointp
takens
times. The return type isOption (Pair (ByStr32) (ByStr32))
. On success,Some point
is returned and on invalid inputsNone
is returned.builtin alt_bn128_G1_neg p
. The builtin takes a pointp
on thealt_bn128
curve and returns the negation of the point. The result isOption (Pair (ByStr32) (ByStr32))
. On success,Some point
is returned and on invalid inputsNone
is returned.builtin alt_bn128_pairing_product pairs
. This builtin takes in a list of pairspairs
of points. Each pair consists of a point in group G1 (Pair {Bystr32 ByStr32}
) as the first component and a point in group G2 (Pair {Bystr64 ByStr64}
) as the second component. Hence the argument has typeList {(Pair (Pair ByStr32 ByStr32) (Pair ByStr64 ByStr64)) }
. The function applies a pairing function on each point to check for equality and returnsTrue
orFalse
depending on whether the pairing check succeeds or fails. Also see https://github.com/ethereum/EIPs/blob/master/EIPS/eip-197.md
Maps¶
A value of type Map kt vt
provides a key-value store where kt
is the
type of keys and vt
is the type of values (in some other programming
languages datatypes like Scilla’s Map
are called associative arrays, symbol
tables, or dictionaries). The type of map keys kt
may be any one of the
following primitive types: String
, IntX
, UintX
, ByStrX
,
ByStr
or BNum
. The type of values vt
may be any type except a
function type, this includes both builtin and user-defined algebraic datatypes.
Since compound types are not supported as map key types, the way to model, e.g. association of pairs of values to another value is by using nested maps. For instance, if one wants to associate with an account and a particular trusted user some money limit the trusted user is allowed to spend on behalf of the account, one can use the following nested map:
field operators: Map ByStr20 (Map ByStr20 Uint128)
= Emp ByStr20 (Map ByStr20 Unit)
The first and the second key are of type ByStr20
and represent accounts and
the trusted users correspondingly. We represent the money limits with the
Uint128
type.
Scilla supports a number of operations on map, which can be categorized as
- in-place operations which modify field maps without making any copies, hence they belong to the imperative fragment of Scilla. These operations are efficient and recommended to use in almost all of the cases;
- functional map operations are intended to use in pure functions, e.g. when
designing a Scilla library, because they never modify the original map they
are called on. These operations may incur significant performance overhead as
some of them create a new (modified) copy of the input map. Syntactically, the
copying operations are all prefixed with
builtin
keyword (see below). Note that to call the functional builtins on a field map one first needs to make a copy of the field map using a command like so:map_copy <- field_map
, which results in gas consumption proportional to the size offield_map
.
In-place map operations¶
m[k] := v
: In-place insert. Inserts a keyk
bound to a valuev
into a mapm
. Ifm
already contains keyk
, the old value bound tok
gets replaced byv
in the map.m
must refer to a mutable field in the current contract. Insertion into nested maps is supported with the syntaxm[k1][k2][...] := v
. If the intermediate keys do not exist in the nested maps, they are freshly created along with the map values they are associated with.x <- m[k]
: In-place local fetch. Fetches the value associated with the keyk
in the mapm
.m
must refer to a mutable field in the current contract. Ifk
has an associated valuev
inm
, then the results of the fetch isSome v
(see theOption
type below), otherwise the result isNone
. After the fetch, the result gets bound to the local variablex
. Fetching from nested maps is supported with the syntaxx <- m[k1][k2][...]
. If one or more of the intermediate keys do not exist in the corresponding map, the result of the fetch isNone
.x <- & c.m[k]
: In-place remote fetch. Works in the same way as the local fetch operation, except thatm
must refer to a mutable field in the contract at addressc
.x <- exists m[k]
: In-place local key existence check.m
must refer to a mutable field in the current contract. Ifk
has an associated value in the mapm
then the result (of typeBool
) of the check isTrue
, otherwise the result isFalse
. After the check, the result gets bound to the local variablex
. Existence checks through nested maps is supported with the syntaxx <- exists m[k1][k2][...]
. If one or more of the intermediate keys do not exist in the corresponding map, the result isFalse
.b <- & exists c.m[k]
: In-place remote key existence check. Works in the same way as the local key existence check, except thatm
must refer to a mutable field in the contract at addressc
.delete m[k]
: In-place remove. Removes a keyk
and its associated value from the mapm
. The identifierm
must refer to a mutable field in the current contract. Removal from nested maps is supported with the syntaxdelete m[k1][k2][...]
. If one or more of the intermediate keys do not exist in the corresponding map, then the operation has no effect. Note that in the case of a nested removaldelete m[k1][...][kn-1][kn]
, only the key-value association ofkn
is removed. The key-value bindings ofk1
tokn-1
will remain in the map.
Functional map operations¶
builtin put m k v
: Insert a keyk
bound to a valuev
into a mapm
. Returns a new map which is a copy of them
but withk
associated withv
. Ifm
already contains keyk
, the old value bound tok
gets replaced byv
in the result map. The value ofm
is unchanged. Theput
function is typically used in library functions. Note thatput
makes a copy ofm
before inserting the key-value pair.builtin get m k
: Fetch the value associated with the keyk
in the mapm
. Returns an optional value (see theOption
type below) – ifk
has an associated valuev
inm
, then the result isSome v
, otherwise the result isNone
. Theget
function is typically used in library functions.builtin contains m k
: Is the keyk
associated with a value in the mapm
? Returns aBool
. Thecontains
function is typically used in library functions.builtin remove m k
: Remove a keyk
and its associated value from the mapm
. Returns a new map which is a copy ofm
but withk
being unassociated with a value. The value ofm
is unchanged. Ifm
does not contain keyk
theremove
function simply returns a copy ofm
with no indication thatk
is missing. Theremove
function is typically used in library functions. Note thatremove
makes a copy ofm
before removing the key-value pair.builtin to_list m
: Convert a mapm
to aList (Pair kt vt)
wherekt
andvt
are key and value types, respectively (see theList
type below).builtin size m
: Return the number of bindings in mapm
. The result type isUint32
. Calling this builtin consumes a small constant amount of gas. But calling it directly on a field map is not supported, meaning that getting the size of field maps while avoiding expensive copying needs some more scaffolding, which you can find out about in the Field map size section.
Note
Builtin functions put
and remove
return a new map, which is
a possibly modified copy of the original map. This may affect performance!
Note
Empty maps can be constructed using the Emp
keyword, specifying the key
and value types as its arguments. This is the way to initialise Map
fields to be empty. For example field foomap : Map Uint128 String = Emp Uint128 String
declares a Map
field with keys of type Uint128
and values of type
String
, which is initialized to be the empty map.
Block Numbers¶
Block numbers have a dedicated type BNum
in Scilla. Variables of
this type are specified with the keyword BNum
followed by an
integer value (for example BNum 101
).
Scilla supports the following built-in operations on block numbers:
eq b1 b2
: Isb1
equal tob2
? Returns aBool
.blt b1 b2
: Isb1
less thanb2
? Returns aBool
.badd b1 i1
: Addi1
of typeUintX
tob1
of typeBNum
. Returns aBNum
.bsub b1 b2
: Subtractb2
fromb1
, both of typeBNum
. Returns anInt256
.
Algebraic Datatypes¶
An algebraic datatype (ADT) is a composite type used commonly in functional programming. Each ADT is defined as a set of constructors. Each constructor takes a set of arguments of certain types.
Scilla is equipped with a number of built-in ADTs, which are described below. Additionally, Scilla allows users to define their own ADTs.
Boolean¶
Boolean values are specified using the type Bool
. The Bool
ADT
has two constructors True
and False
, neither of which take any
arguments. Thus the following code fragment constructs a value of type
Bool
by using the constructor True
:
x = True
Option¶
Optional values are specified using the type Option t
, where t
is some type. The Option
ADT has two constructors:
Some
represents the presence of a value. TheSome
constructor takes one argument (the value, of typet
).None
represents the absence of a value. TheNone
constructor takes no arguments.
The following code snippet constructs two optional values. The first
value is an absent string value, constructed using None
. The
second value is the Int32
value 10, which, because the value is
present, is constructed using Some
:
let none_value = None {String}
let some_value =
let ten = Int32 10 in
Some {Int32} ten
Optional values are useful for initialising fields where the value is not yet known:
field empty_bool : Option Bool = None {Bool}
Optional values are also useful for functions that might not have a
result, such as the get
function for maps:
getValue = builtin get m _sender;
match getValue with
| Some v =>
(* _sender was associated with v in m *)
v = v + v;
...
| None =>
(* _sender was not associated with a value in m *)
...
end
List¶
Lists of values are specified using the type List t
, where t
is some type. The List
ADT has two constructors:
Nil
represents an empty list. TheNil
constructor takes no arguments.Cons
represents a non-empty list. TheCons
constructor takes two arguments: The first element of the list (of typet
), and another list (of typeList t
) representing the rest of the list.
All elements in a list must be of the same type t
. In other
words, two values of different types cannot be added to the same list.
The following example shows how to build a list of Int32
values. First we create an empty list using the Nil
constructor. We then add four other values one by one using the
Cons
constructor. Notice how the list is constructed backwards by
adding the last element, then the second-to-last element, and so on,
so that the final list is [11; 10; 2; 1]
:
let one = Int32 1 in
let two = Int32 2 in
let ten = Int32 10 in
let eleven = Int32 11 in
let nil = Nil {Int32} in
let l1 = Cons {Int32} one nil in
let l2 = Cons {Int32} two l1 in
let l3 = Cons {Int32} ten l2 in
Cons {Int32} eleven l3
Scilla provides three structural recursion primitives for lists, which can be used to traverse all the elements of any list:
list_foldl: ('B -> 'A -> 'B) -> 'B -> (List 'A) -> 'B
: Recursively process the elements in a list from front to back, while keeping track of an accumulator (which can be thought of as a running total).list_foldl
takes three arguments, which all depend on the two type variables'A
and'B
:- The function processing the elements. This function takes two
arguments. The first argument is the current value of the
accumulator (of type
'B
). The second argument is the next list element to be processed (of type'A
). The result of the function is the next value of the accumulator (of type'B
). - The initial value of the accumulator (of type
'B
). - The list of elements to be processed (of type
(List 'A)
).
The result of applying
list_foldl
is the value of the accumulator (of type'B
) when all list elements have been processed.- The function processing the elements. This function takes two
arguments. The first argument is the current value of the
accumulator (of type
list_foldr: ('A -> 'B -> 'B) -> 'B -> (List 'A) -> 'B
: Similar tolist_foldl
, except the list elements are processed from back to front. Notice also that the processing function takes the list element and the accumulator in the opposite order from the order inlist_foldl
.list_foldk: ('B -> 'A -> ('B -> 'B) -> 'B) -> 'B -> (List 'A) -> 'B
: Recursively process the elements in a list according to a folding function, while keeping track of an accumulator.list_foldk
is a more general version of the left and right folds, which, by the way, can be both implemented in terms of it.list_foldk
takes three arguments, which all depend on the two type variables'A
and'B
:- The function describing the fold step. This function takes three
arguments. The first argument is the current value of the
accumulator (of type
'B
). The second argument is the next list element to be processed (of type'A
). The third argument represents the postponed recursive call (of type'B -> 'B
). The result of the function is the next value of the accumulator (of type'B
). The computation terminates if the programmer does not invoke the postponed recursive call. This is a major difference betweenlist_foldk
and the left and right folds which process their input lists from the beginning to the end unconditionally. - The initial value of the accumulator
z
(of type'B
). - The list of elements to be processed (of type
List 'A
).
- The function describing the fold step. This function takes three
arguments. The first argument is the current value of the
accumulator (of type
Note
When an ADT takes type arguments (such as List 'A
), and occurs
inside a bigger type (such as the type of list_foldl
), the ADT
and its arguments must be grouped using parentheses ( )
. This
is the case even when the ADT occurs as the only argument to
another ADT. For instance, when constructing an empty list of
optional values of type Int32
, one must instantiate the list
type using the syntax Nil {(Option Int32)}
.
To further illustrate the List
type in Scilla, we show a small
example using list_foldl
to count the number of elements in a
list. For an example of list_foldk
see list_find.
1 2 3 4 5 6 7 8 9 10 11 12 | let list_length : forall 'A. List 'A -> Uint32 =
tfun 'A =>
fun (l : List 'A) =>
let foldl = @list_foldl 'A Uint32 in
let init = Uint32 0 in
let one = Uint32 1 in
let iter =
fun (z : Uint32) =>
fun (h : 'A) =>
builtin add one z
in
foldl iter init l
|
list_length
defines a function that takes a type argument 'A
,
and a normal (value) argument l
of type List 'A
.
'A
is a type variable which must be instantiated by the code
that intends to use list_length
. The type variable is specified in
line 2.
In line 4 we instantiate the types for list_foldl
. Since we are
traversing a list of values of type 'A
, we pass 'A
as the
first type argument to list_foldl
, and since we are calculating
the length of the list (a non-negative integer), we pass Uint32
as
the accumulator type.
In line 5 we define the initial value of the accumulator. Since an
empty list has length 0, the initial value of the accumulator is 0 (of
type Uint32
, to match the accumulator type).
In lines 6-10 we specify the processing function iter
, which takes
the current accumulator value z
and the current list element
h
. In this case the processing function ignores the list element,
and increments the accumulator by 1. When all elements in the list
have been processed, the accumulator will have been incremented as
many times as there are elements in the list, and hence the final
value of the accumulator will be equal to the length of the list.
In line 12 we apply the type-instantiated version of list_foldl
from line 4 to the processing function, the initial accumulator, and
the list of values.
Common utilities for the List
type (including list_length
) are
provided in the ListUtils
library as part of the standard library
distribution for Scilla.
Pair¶
Pairs of values are specified using the type Pair t1 t2
, where
t1
and t2
are types. The Pair
ADT has one constructor:
Pair
represents a pair of values. ThePair
constructor takes two arguments, namely the two values of the pair, of typest1
andt2
, respectively.
Note
Pair
is both the name of a type and the name of a constructor
of that type. An ADT and a constructor typically only share their
names when the constructor is the only constructor of the ADT.
A Pair
value may contain values of different types. In other
words, t1
and t2
need not be the same type.
Below is an example where we declare a field pp
of type Pair
String Uint32
, which we then initialise by constructing a pair
consisting of a value of type String
and a value of type
Uint32
:
field pp: Pair String Uint32 =
let s1 = "Hello" in
let num = Uint32 2 in
Pair {String Uint32} s1 num
Notice the difference in how we specify the type of the field as
Pair A' B'
, and how we specify the types of values given to the
constructor as Pair { A' B' }
.
We now illustrate how pattern matching can be used to extract the
first element from a Pair
. The function fst
shown below
is defined in the PairUtils
library of the Scilla standard library.
let fst =
tfun 'A =>
tfun 'B =>
fun (p : Pair ('A) ('B)) =>
match p with
| Pair a b => a
end
To apply fst
to one must first instantiate the type variables
'A
and 'B
, which is done as follows:
p <- pp;
fst_specialised = @fst String Uint32;
p_fst = fst_specialised p
The value associated with the identifier p_fst
will be the string
"Hello"
.
Note
Using Pair
is generally discouraged. Instead, the programmer
should define an ADT which is specialised to the particular
type of pairs that is needed in the particular use case. See the
section on User-defined ADTs below.
Nat¶
Peano numbers are specified using the type Nat
. The Nat
ADT
has two constructors:
Zero
represents the number 0. TheZero
constructor takes no arguments.Succ
represents the successor of another Peano number. TheSucc
constructor takes one argument (of typeNat
) which represents the Peano number that is one less than the current number.
The following code shows how to build the Peano number corresponding to the integer 3:
let three =
let zero = Zero in
let one = Succ zero in
let two = Succ one in
Succ two
Scilla provides two structural recursion primitives for Peano numbers,
which can be used to traverse all the Peano numbers from a given
Nat
down to Zero
:
nat_fold: ('A -> Nat -> 'A) -> 'A -> Nat -> 'A
: Recursively process the succession of numbers from aNat
down toZero
, while keeping track of an accumulator.nat_fold
takes three arguments, two of which depend on the type variable'A
:- The function processing the numbers. This function takes two
arguments. The first argument is the current value of the
accumulator (of type
'A
). The second argument is the next Peano number to be processed (of typeNat
). Incidentally, the next number to be processed is the predecessor of the current number being processed. The result of the function is the next value of the accumulator (of type'A
). - The initial value of the accumulator (of type
'A
). - The first Peano number to be processed (of type
Nat
).
The result of applying
nat_fold
is the value of the accumulator (of type'A
) when all Peano numbers down toZero
have been processed.- The function processing the numbers. This function takes two
arguments. The first argument is the current value of the
accumulator (of type
nat_foldk: ('A -> Nat -> ('A -> 'A) -> 'A) -> 'A -> Nat -> 'A
: Recursively process the Peano numbers down to zero according to a folding function, while keeping track of an accumulator.nat_foldk
is a more general version of the left fold allowing for early termination. It takes three arguments, two depending on the type variable'A
.- The function describing the fold step. This function takes three
arguments. The first argument is the current value of the
accumulator (of type
'A
). The second argument is the predecessor of the Peano number being processed (of typeNat
). The third argument represents the postponed recursive call (of type'A -> 'A
). The result of the function is the next value of the accumulator (of type'A
). The computation terminates if the programmer does not invoke the postponed recursive call. Left folds inevitably process the whole list whereasnat_foldk
can differ in this regard. - The initial value of the accumulator
z
(of type'A
). - The Peano number to be processed (of type
Nat
).
- The function describing the fold step. This function takes three
arguments. The first argument is the current value of the
accumulator (of type
To better understand nat_foldk
, we explain how nat_eq
works.
nat_eq
checks to see if two Peano numbers are equivalent. Below
is the program, with line numbers and an explanation.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 | let nat_eq : Nat -> Nat -> Bool =
fun (n : Nat) => fun (m : Nat) =>
let foldk = @nat_foldk Nat in
let iter =
fun (n : Nat) => fun (ignore : Nat) => fun (recurse : Nat -> Nat) =>
match n with
| Succ n_pred => recurse n_pred
| Zero => m (* m is not zero in this context *)
end in
let remaining = foldk iter n m in
match remaining with
| Zero => True
| _ => False
end
|
Line 2 specifies that we take two Peano numbers m
and n
.
Line 3 instantiates the type of nat_foldk
, we give it Nat
because we will be passing a Nat
value as the fold accumulator.
Lines 4 to 8 specify the fold description, this is the first argument
that nat_foldk
takes usually of type 'A -> Nat -> ('A -> 'A) -> 'A
but we have specified that 'A
is Nat
in this case. Our function
takes the accumulator n
and ignore : Nat
is the predecessor of
the number being processed which we don’t care about in this particular case.
Essentially, we start accumulating the end result from n
and iterate
at most m
times (see line 10), decrementing both n
and m
at each recursive step (lines 4 - 9).
The m
variable gets decremented implicitly because this is how nat_foldk
works under the hood.
And we explicitly decrement n
using pattern matching (lines 6, 7).
To continue iteratively decrement both m
and n
we use recurse
on line 7.
If the two input numbers are equal, we will get the accumulator (n
) equal to
zero in the end.
We call the final value of the accumulator remaining
on line 10.
At the end we will be checking to see if our accumulator
ended up at Zero
to say if the input numbers are equal.
The last lines, return True
when the result of the fold is Zero
and False
otherwise as described above.
In the case when accumulator n
reaches zero (line 8) while m
still has not been fully processed, we stop iteration
(hence no recurse
on that line) and return a non-zero natural number
to indicate inequality.
Any number (e.g. Succ Zero
) would do, but to make the code concise
we return the original input number m
because we know iter
gets called on m
only if it’s not zero.
In the symmetrical case when m
reaches zero while the accumulator n
is still strictly positive, we indicate inequality, because remaining
gets this final value of n
.
User-defined ADTs¶
In addition to the built-in ADTs described above, Scilla supports user-defined ADTs.
ADT definitions may only occur in the library parts of a program, either in the library part of the contract, or in an imported library. An ADT definition is in scope in the entire library in which it is defined, except that an ADT definition may only refer to other ADT definitions defined earlier in the same library, or in imported libraries. In particular, an ADT definition may not refer to itself in an inductive/recursive manner.
Each ADT defines a set of constructors. Each constructor specifies a number of types which corresponds to the number and types of arguments that the constructor takes. A constructor may be specified as taking no arguments.
The ADTs of a contract must have distinct names, and the set of all
constructors of all ADTs in a contract must also have distinct
names. Both the ADT and constructor names must begin with a capital letter
(‘A’ - ‘Z’). However, a constructor and an ADT may have the same name, as
is the case with the Pair
type whose only constructor is also called
Pair
.
As an example of user-defined ADTs, consider the following type declarations from a contract implementing a chess-like game called Shogi or Japanese Chess (https://en.wikipedia.org/wiki/Shogi). When in turn, a player can choose to either move one of his pieces, place a previously captured piece back onto the board, or resign and award the victory to the opponent.
The pieces of the game can be defined using the following type
Piece
:
type Piece =
| King
| GoldGeneral
| SilverGeneral
| Knight
| Lance
| Pawn
| Rook
| Bishop
Each of the constructors represents a type of piece in the game. None of the constructors take any arguments.
The board is represented as a set of squares, where each square has two coordinates:
type Square =
| Square of Uint32 Uint32
The type Square
is an example of a type where a constructor has
the same name as the type. This usually happens when a type has only
one constructor. The constructor Square
takes two arguments, both
of type Uint32
, which are the coordinates (the row and the column)
of the square on the board.
Similar to the definition of the type Piece
, we can define the
type of direction of movement using a constructor for each of the
legal directions as follows:
type Direction =
| East
| SouthEast
| South
| SouthWest
| West
| NorthWest
| North
| NorthEast
We are now in a position to define the type of possible actions that a user may choose to perform when in turn:
type Action =
| Move of Square Direction Uint32 Bool
| Place of Piece Square
| Resign
If a player chooses to move a piece, she should use the constructor
Move
, and provide four arguments:
- An argument of type
Square
, indicating the current position of the piece she wants to move. - An argument of type
Direction
, indicating the direction of movement. - An argument of type
Uint32
, indicating the distance the piece should move. - An argument of type
Bool
, indicating whether the moved piece should be promoted after being moved.
If instead the player chooses to place a previously captured piece
back onto the board, she should use the constructor Place
, and
provide two arguments:
- An argument of type
Piece
, indicating which piece to place on the board. - An argument of type
Square
, indicating the position the piece should be placed in.
Finally, if the player chooses to resign and award the victory to her
opponent, she should use the constructor Resign
. Since Resign
does not take any arguments, no arguments should be provided.
To check which action a player has chosen we use a match statement or a match expression:
transition PlayerAction (action : Action)
...
match action with
| Resign =>
...
| Place piece square =>
...
| Move square direction distance promote =>
...
end;
...
end
Type identity for user-defined ADTs¶
Note
Due to a bug in the Scilla implementation the information in this section is only valid from Scilla version 0.10.0 and forwards. Contracts written in Scilla versions prior to 0.10.0 and which exploit this bug will have to be rewritten and redeployed, as they will no longer work from version 0.10.0 and onwards.
Each type declaration defines a unique type. In particular this means that even if two libraries both define identical types, the types are considered different.
As an example, consider the following two contracts:
library C1Lib
type T =
| C1 of Uint32
| C2 of Bool
contract Contract1()
field contract2_address : ByStr20 = 0x1234567890123456789012345678901234567890
transition Sending ()
c2 <- contract2_address;
x = Uint32 0;
out = C1 x;
msg = { _tag : "Receiving" ; _recipient : c2 ; _amount : Uint128 0 ;
param : out };
no_msg = Nil {Message};
msgs = Cons {Message} msg no_msg;
send msgs
end
(* ******************************* *)
(* Contract2 is deployed at address 0x1234567890123456789012345678901234567890 *)
library C2Lib
type T =
| C1 of Uint32
| C2 of Bool
contract Contract2()
transition Receiving (param : T)
match param with
| C1 v =>
| C2 b =>
end
end
Even though both contracts define identical types T
, the two types
are considered different in Scilla. In particlar this means that the
message sent from Contract1
to Contract2
will not trigger the
Receiving
transition, because the value sent as the param
message field has the type T
from Contract1
, whereas the type
T
from Contract2
is expected.
In order to pass a value of a user-defined ADT as a parameter to a transition, the type must be defined in a user-defined library, which both the sending and the receiving contract must import:
library MutualLib
type T =
| C1 of Uint32
| C2 of Bool
(* ******************************* *)
import MutualLib
library C1Lib
contract Contract1()
field contract2_address : ByStr20 = 0x1234567890123456789012345678901234567890
transition Sending ()
c2 <- contract2_address;
x = Uint32 0;
out = C1 x;
msg = { _tag : "Receiving" ; _recipient : c2 ; _amount : Uint128 0 ;
param : out };
no_msg = Nil {Message};
msgs = Cons {Message} msg no_msg;
send msgs
end
(* ******************************* *)
(* Contract2 is deployed at address 0x1234567890123456789012345678901234567890 *)
scilla_version 0
import MutualLib
library C2Lib
contract Contract2()
transition Receiving (param : T)
match param with
| C1 v =>
| C2 b =>
end
end
The section User-defined Libraries has more information on how to define and use libraries.
More ADT examples¶
To further illustrate how ADTs can be used, we provide some more
examples and describe them in detail. Versions of both the functions
described below can be found in the ListUtils
part of the
Scilla standard library.
Computing the Head of a List¶
The function list_head
returns the first element of a list.
Since a list may be empty, list_head
may not always be able to
compute a result, and thus should return a value of the Option
type. If the list is non-empty, and the first element is h
, then
list_head
should return Some h
. Otherwise, if the list is
empty, list_head
should return None
.
The following code snippet shows the implementation of list_head
,
and how to apply it:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 | let list_head =
tfun 'A =>
fun (l : List 'A) =>
match l with
| Cons h t =>
Some {'A} h
| Nil =>
None {'A}
end
let int_head = @list_head Int32 in
let one = Int32 1 in
let two = Int32 2 in
let three = Int32 3 in
let nil = Nil {Int32} in
let l1 = Cons {Int32} three nil in
let l2 = Cons {Int32} two l1 in
let l3 = Cons {Int32} one l2 in
int_head l3
|
Line 2 specifies that 'A
is a type parameter to the function,
while line 3 specifies that l
is a (value) parameter of type
List 'A
. In other words, lines 1-3 specify a function
list_head
which can be instantiated for any type 'A
, and which
takes as an argument a value of type List 'A
.
The pattern-match in lines 4-9 matches on the value of l
. In line
5 we match on the list constructor Cons h t
, where h
is the
first element of the list, and t
is the rest of the list. If the
list is not empty then the match is successful, and we return the
first element as an optional value Some h
. In line 7 we match on
the list constructor Nil
. If the list is empty then the match is
successful, and we return the optional value None
indicating that
there was no head element of the list.
Line 11 instantiates the list_head
function for the type
Int32
, so that list_head
can be applied to values of type
List Int32
. Lines 13-20 build a list of type List Int32
, and
line 21 invokes the instantiated list_head
function on the list
that was built.
Computing a Left Fold¶
The function list_foldl
returns the result of a left fold given a function
f : 'B -> 'A -> 'B
, accumulator z : 'B
and list xs : List 'A
.
This can be implemented as a recursion primitive or a list utility function.
A left fold is a recursive application of an accumulator z
and next
list element x : 'A
with f
repetitively until there are no more list
elements. For example the left fold on [1,2,3]
using subtraction starting with
accumulator 0 would be ((0-1)-2)-3 = -6
. The left fold is explained in
pseudocode below, note that the result is always the accumulator type.
1 2 | list_foldl _ z [] = z
list_foldl f z (x:xs) = list_foldl f (f z x) xs
|
The same can be achieved with list_foldk
by partially applying a left fold
description; this avoids illegal direct recursion. Our fold description
left_f : 'B -> 'A -> ('B -> 'B) -> 'B
takes arguments accumulator,
next list element and recursive call. The recursive call will be supplied
by the list_foldk
function. An implementation is explained below.
1 2 3 4 5 6 7 8 | let list_foldl : forall 'A. forall 'B. ( 'B -> 'A -> 'B) -> 'B -> List 'A -> 'B =
tfun 'A => tfun 'B =>
fun (f : 'B -> 'A -> 'B) =>
let left_f = fun (z: 'B) => fun (x: 'A) =>
fun (recurse : 'B -> 'B) => let res = f z x in
recurse res in
let folder = @list_foldk 'A 'B in
folder left_f
|
On line 1, we declare the name and type signature as according to the first
paragraph. On the second line, we say that the function takes two types as arguments
'A
and 'B
. The third line says that we take some function f
to process the list element
and accumulator, as in paragraph two.
On line 4, we define the fold description using f
. The fold description does not
take a function but instead it should be implemented in terms of some function, as
according to the type signature, left_f : 'B -> 'A -> ('B -> 'B) -> 'B
.
left_f
takes arguments as described in paragraph two. We calculate the new
accumulator f z x
and call it res
. Then we recursively call with the new
accumulator.
On line 7, we instantiate an instance of list_foldk
that has the right types
for the job using a type application.
On line 8, we partially apply folder
with the left fold description.
. What is significant about list_foldk
is that when calling the description,
it provides a recursive call to itself, changing to the next element
in the list and respective tail each time. This results in a function that
just needs the user to provide the updated accumulator in the description.
Computing a Right Fold¶
The function list_foldr
returns the result of a right fold given some
function f : 'A -> 'B -> 'B
, accumulator z : 'B
and
list xs : List 'A
. Like list_foldl
, this can be a recursion primitive
or a list utility function.
A right fold is similar to a left fold but is reversed in a way.
The right fold applies a function f
with an accumulator z
starting from
the end and then combines with the second last element, third last element,
etc… until it reaches the beginning. For example a right fold on
the list [1,2,3]
with subtraction starting with accumulator 0 would
be equal to 1-(2-(3-0)) = 2
. It is listed below in pseudocode,
note that the result is always the accumulator type.
1 2 | list_foldr _ z [] = z
list_foldr f z (x:xs) = f x (list_foldr f z xs)
|
Like before, the same can be achieved with list_foldk
by partially
applying a right fold description. The fold description takes arguments
accumulator z : 'B
, next list element x : 'A
and recursive call
recurse : 'B -> 'B
. The recursive call will be supplied by the
list_foldk
function. An implementation is explained below.
1 2 3 4 5 6 7 | let list_foldr : forall 'A. forall 'B. ('A -> 'B -> 'B) -> 'B -> List 'A -> 'B =
tfun 'A => tfun 'B =>
fun (f : 'A -> 'B -> 'B) =>
let right_f = fun (z: 'B) => fun (x: 'A) =>
fun (recurse : 'B -> 'B) => let res = recurse z in f x res in
let folder = @list_foldk 'A 'B in
folder right_f
|
This is very similar to before. On line 1 we declare the name and type
signature, according to the first paragraph. On line 2, we take two
type arguments 'A
and 'B
. The third line says that we take some
function f
to process the list element x : 'A
and accumulator z
.
The argument order is necessarily different to that of a left fold.
Following that we write a fold description like before.
list_foldk
processes lists from left to right.
But we need list_foldr
to emulate the right-to-left traversal.
By calling recurse z
on line 5 as our first action, we postpone actual computation
with the combining function f
preserving the original accumulator until the very end.
Once the recursive call reaches an empty list it returns the original accumulator.
Then the function calls f x res
(line 5) will evaluate outwards combining
from the end to the beginning, see paragraph two.
The recursive call recurse z
on line 5 may seem to be the same each time but what is changing
is the list element we process.
On line 6, we instantiate list_foldk
by applying the types 'A
and 'B
to make
a type-specific function. The last line we partially apply folder
with the
right fold description. Like before what is special about list_foldk
is that it calls
this function with a recursive call to itself that each time slightly truncates the list;
this provides the recursion.
Checking for Existence in a List¶
The function list_exists
takes a predicate function and a list,
and returns a value indicating whether the predicate holds for at
least one element in the list.
A predicate function is a function returning a Boolean value, and since we want to apply it to elements in the list, the argument type of the function should be the same as the element type of the list.
list_exists
should return either True
(if the predicate holds
for at least one element) or False
(if the predicate does not hold
for any element in the list), so the return type of list_exists
should be Bool
.
The following code snippet shows the implementation of
list_exists
, and how to apply it:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 | let list_exists =
tfun 'A =>
fun (f : 'A -> Bool) =>
fun (l : List 'A) =>
let folder = @list_foldl 'A Bool in
let init = False in
let iter =
fun (z : Bool) =>
fun (h : 'A) =>
let res = f h in
match res with
| True =>
True
| False =>
z
end
in
folder iter init l
let int_exists = @list_exists Int128 in
let f =
fun (a : Int128) =>
let three = Int128 3 in
builtin lt a three
(* build list l3 similar to the previous example *)
...
(* check if l3 has at least one element satisfying f *)
int_exists f l3
|
As in the previous example 'A
is a type variable to the
function. The function takes two arguments:
- A predicate
f
, i.e., a function that returns aBool
. In this case,f
will be applied to elements of the list, so the argument type of the predicate should be'A
. Hence,f
should have the type'A -> Bool
. - A list of elements
l
of typeList 'A
, so that the type of the elements in the list matches the argument type off
.
To traverse the elements of the input list l
we use
list_foldl
. In line 5 we instantiate list_foldl
for lists with
elements of type 'A
and for the accumulator type Bool
. In
line 6 we set the initial accumulator value to False
to indicate
that no element satisfying the predicate has yet been seen.
The processing function iter
defined in lines 7-16 tests the
predicate on the current list element, and returns an updated
accumulator. If an element has been found which satisfies the
predicate, the accumulator is set to True
and remains so for the
rest of the traversal.
The final value of the accumulator is either True
, indicating that
f
returned True
for at least one element in the list, or
False
, indicating that f
returned False
for all elements
in the list.
In line 20 we instantiate list_exists
to work on lists of type
Int128
. In lines 21-24 we define the predicate, which returns
True
if its argument is less than 3, and returns False
otherwise.
Omitted in line 27 is building the same list l3
as in the previous
example. In line 30 we apply the instantiated list_exists
to the
predicate and the list.
Finding the first occurrence satisfying a predicate¶
The function list_find
searches for the first occurrence in a
list that satisfies some predicate p : 'A -> Bool
. It takes
the predicate and the list, returning Some {'A} x :: Option 'A
if
x
is the first element such that p x
and None {'A} :: Option 'A
otherwise.
Below we have an implementation of list_find
that illustrates
how to use list_foldk
.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 | let list_find : forall 'A. ('A -> Bool) -> List 'A -> Option 'A =
tfun 'A =>
fun (p : 'A -> Bool) =>
let foldk = @list_foldk 'A (Option 'A) in
let init = None {'A} in
(* continue fold on None, exit fold when Some compare st. p(compare) *)
let predicate_step =
fun (ignore : Option 'A) => fun (x : 'A) =>
fun (recurse: Option 'A -> Option 'A) =>
let p_x = p x in
match p_x with
| True => Some {'A} x
| False => recurse init
end in
foldk predicate_step init
|
Like before, we take a type variable 'A
on line 2 and take the predicate
on the next line. We begin by using this type variable to instantiate foldk
,
by giving it our processing type and return type. The processing type being
the list element type and the result type being Option 'A
. The next line
is our accumulator, we assume that at the start of the search there is no
satisfier.
On line 7, we write a fold description for foldk
. This embodies the order of
the recursion and conditions for recursion. predicate_step
has the
type Option 'A -> 'A -> (Option 'A -> Option 'A) -> Option 'A
.
The first argument is the accumulator, the second x
is the next element to
process and the third recurse
is the recursive call. We do not care what
the accumulator ignore
is since if it mattered we will have already
terminated.
On lines 10 to 12 check for p x
and if so return Some {'A} x
. In the case
that p x
does not hold, try again from scratch with the next element and
so on via recursion. recurse init
is in pseudo-code equal to
λk. foldk predicate_step init k xs
where xs
is the tail of our list of
to be processed elements.
With the final line we partially apply foldk
so that it just takes a list
argument and gives us our final answer. The first argument of foldk
gives
us the specific fold we want, for example if you wanted a left fold you
would replace predicate_step
with something else.
Standard Libraries¶
The Scilla standard library is documented separately.
User-defined Libraries¶
In addition to the standard library provided by Scilla, users are allowed
to deploy library code on the blockchain. Library files are allowed to only
contain pure Scilla code (which is the same restriction that in-contract
library code has). Library files must use the .scillib
file extension.
Below is an example of a user-defined library that defines a single function
add_if_equal
that adds to Uint128
values if they are equal and returns
0
otherwise.
import IntUtils
library ExampleLib
let add_if_equal =
fun (a : Uint128) => fun (b : Uint128) =>
let eq = uint128_eq a b in
match eq with
| True => builtin add a b
| False => Uint128 0
The structure of a library file is similar to the structure of the library part of a Scilla contract. A library file contains definitions of variables and pure library functions, but does not contain an actual contract definition with parameters, fields, transitions and so on.
Of particular importance is that a library cannot declare fields. Therefore, all libraries are stateless and can only contain pure code.
Similar to how contracts can import libraries, a library can import other libraries
(including user-defined libraries) too. The scope of variables in an imported library
is restricted to the immediate importer. So if X
imports library Y
which in
turn imports library Z
, then the names in Z
are not in scope in X`, but only
in Y
. Cyclic dependencies in imports are not allowed and flagged as errors
during the checking phase.
Local Development with User-defined Libraries¶
To use variables and functions declared in an external (user-defined) library module,
the command line argument to the Scilla executables must include a -libdir
option,
along with a list of directories as an argument. If the Scilla file imports a library ALib
,
then the Scilla executable will search for a library file called ALib.scillib
in the directories provided. If more than one directory contains a file with the correct name,
then the directories are given priority in the same order as they are provided to the Scilla executable.
Alternatively, the environment variable SCILLA_STDLIB_PATH
can be set to a list of library directories.
scilla-checker
typechecks library modules in the same way as contract modules. Similarly,
scilla-runner
can deploy libraries. Note that scilla-runner
takes a blockhain.json
as
argument (the way it does for Contract Creation) to be command
line argument compatible with contract creation.
User-defined Libraries on the Blockchain¶
While the Zilliqa blockchain is designed to provide the standard Scilla libraries to an executing contract, it must be provided with extra information to support user-defined libraries.
The init.json
of a library must include a Bool
entry named _library
, set to
True
. Additionally,
A contract or a library that imports user-defined libraries must include in its init.json
an entry named _extlibs
, of Scilla type List (Pair String ByStr20)
. Each entry in
the list maps an imported library’s name to its address in the blockchain.
Continuing the previous example, a contract or library that imports Examplelib
should have
the following entry in its init.json
:
[
...,
{
"vname" : "_library",
"type" : "Bool",
"value": { "constructor": "True", "argtypes": [], "arguments": [] }
}
{
"vname" : "_extlibs",
"type" : "List(Pair String ByStr20)",
"value" : [
{
"constructor" : "Pair",
"argtypes" : ["String", "ByStr20"],
"arguments" : ["ExampleLib", "0x986556789012345678901234567890123456abcd"]
},
...
]
}
]
Namespaces¶
Import statements can be used to define separate namespaces for imported names.
To push the names from a library Foo
into the namespace Bar
, use the statement
import Foo as Bar
. Accessing a variable v
in Foo must now be done using the qualified
name Bar.v
. This is useful when importing multiple libraries that define the same name.
The same variable name must not be defined more than once in the same namespace, so if multiple imported libraries define the same name, then at most one of the libraries may reside in the default (unqualified) namespace. All other conflicting libraries must be pushed to separate namespaces.
Extending our previous example, shown below is a contract that imports ExampleLib
in namespace Bar
, to use the function add_if_equal
.
scilla_version 0
import ExampleLib as Bar
library MyContract
let adder = fun (a : Uint128) => fun (b : Uint128) =>
Bar.add_if_equal a b
contract MyContract ()
...
Scilla versions¶
Major and Minor versions¶
Scilla releases have a major version, minor version and a patch
number, denoted as X.Y.Z
where X
is the major version,
Y
is the minor version, and Z
the patch number.
- Patches are usually bug fixes that do not impact the behaviour of existing contracts. Patches are backward compatible.
- Minor versions typically include performance improvements and feature additions that do not affect the behaviour of existing contracts. Minor versions are backward compatible until the latest major version.
- Major versions are not backward compatible. It is expected that miners have access to implementations of each major version of Scilla for running contracts set to that major version.
Within a major version, miners are advised to use the latest minor revision.
The command scilla-runner -version
will print major, minor and patch versions
of the interpreter being invoked.
Contract Syntax¶
Every Scilla contract must begin with a major version declaration. The syntax is shown below:
(***************************************************)
(* Scilla version *)
(***************************************************)
scilla_version 0
(***************************************************)
(* Associated library *)
(***************************************************)
library MyContractLib
...
(***************************************************)
(* Contract definition *)
(***************************************************)
contract MyContract
...
When deploying a contract the output of the interpreter contains the
field scilla_version : X.Y.Z
, to be used by the blockchain code to
keep track of the version of the contract. Similarly,
scilla-checker
also reports the version of the contract on a
successful check.
The init.json
file¶
In addition to the version specified in the contract source code, it
is also required that the contract’s init.json
specifies the same
version when the contract is deployed and when the contract’s
transitions are invoked. This eases the process for the blockchain
code to decide which interpreter to invoke.
A mismatch in the versions specified in init.json
and the source code
will lead to a gas-charged error by the interpreter.
An example init.json
:
[
{
"vname" : "_creation_block",
"type" : "BNum",
"value" : "1"
},
{
"vname" : "_scilla_version",
"type" : "Uint32",
"value" : "1",
}
]
Chain Calls¶
When a user invokes a transition on a contract by sending a message with the contract’s address as the recipient, then that transition may then send one or more messages onwards, possibly invoking other transitions on other contracts. The resulting collection of messages, fund transfers, transition invocations, and contract state changes are referred to as a transaction.
A transition that sends a message invoking another transition (typically on another contract) is referred to as a chain call.
During a transaction a LIFO queue (i.e., a stack) of unprocessed messages is maintained. Initially, the message queue contains only the single message sent by the original user, but additional messages may be added to the queue when a transition performs a chain call.
When a transition finishes, its outgoing messages are added to the message queue. The first message in the queue is then removed from the queue for processing. If there are no messages left to process, then the transaction finishes, and all state changes and fund transfers are committed to the blockchain.
When a transition sends multiple messages, the messages are added to the queue in the following order:
- If multiple
send
statements are executed, then the messages of the lastsend
are added first. This means that the messages of the firstsend
get processed first. - If a
send
statement is given a list with multiple messages in it, then the messages of the tail of the list are added to the queue before the head of the list is added. This means that the first message in the list gets processed first.
Any run-time failure during the execution of a transaction causes the entire transaction to be aborted, with no further statements being executed, no further messages being processed, all state changes being rolled back, and all transferred funds returned to their respective senders. However, gas is still charged for the transcaction up until the point of the failure.
The total number of messages that can be sent in a single transaction is currently set at 20. The number is subject to revision in the future.
Contracts of different Scilla versions may invoke transitions on each other. The semantics of message passing between contracts is guaranteed to be backward compatible between major versions.
Accounting¶
For the transfer of native ZIL funds, Scilla follows an acceptance
semantics. For a transfer to take place between contracts the funds
must explicitly be accepted by the recipient by executing an
accept
statement - it is not sufficient that the sender executes a
send
statement.
When a contract executes a send
statement the _amount
values
of the outgoing messages are deducted from the _balance
field. However, the funds are not yet added to the recipient’s
_balance
field, because they have not yet been explicitly accepted
by the recipient.
When the recipient executes an accept
statement the _amount
of
the incoming message is added to the contract’s _balance
field.
If the recipient does not execute and accept
statement, the
_amount
is transferred back to the sender before the next message
is processed.
While the funds in a message are “in transit”, i.e., between the point at which they is sent by the sender and the point at which they are accepted by the recipient (or, in the case of non-acceptance, the point at which the recipient transition finishes executing), the funds are stored away in the message itself, and therefore do not appear in neither the sender’s nor the recipient’s balances.
Note
A user account (i.e., an address that does not hold a contract) implicitly accepts all incoming funds, but does not do so until the message carrying the funds is processed.
If a contract executes a send
statement on messages for which the
total amount is more than the contract’s current balance, then a
run-time error occurs, and the entire transaction is aborted. In other
words, no account balance may drop below 0 at any point during a
transaction.