Application Binary Interface Specification

Basic Design

The Application Binary Interface is the standard way to interact with contracts in the Ethereum ecosystem, both from outside the blockchain and for contract-to-contract interaction. Data is encoded according to its type, as described in this specification. The encoding is not self describing and thus requires a schema in order to decode.

We assume the interface functions of a contract are strongly typed, known at compilation time and static. No introspection mechanism will be provided. We assume that all contracts will have the interface definitions of any contracts they call available at compile-time.

This specification does not address contracts whose interface is dynamic or otherwise known only at run-time. Should these cases become important they can be adequately handled as facilities built within the Ethereum ecosystem.

Function Selector

The first four bytes of the call data for a function call specifies the function to be called. It is the first (left, high-order in big-endian) four bytes of the Keccak (SHA-3) hash of the signature of the function. The signature is defined as the canonical expression of the basic prototype, i.e. the function name with the parenthesised list of parameter types. Parameter types are split by a single comma - no spaces are used.


The return type of a function is not part of this signature. In Solidity’s function overloading return types are not considered. The reason is to keep function call resolution context-independent. The JSON description of the ABI however contains both inputs and outputs. See (the JSON ABI)

Argument Encoding

Starting from the fifth byte, the encoded arguments follow. This encoding is also used in other places, e.g. the return values and also event arguments are encoded in the same way, without the four bytes specifying the function.


The following elementary types exist:

  • uint<M>: unsigned integer type of M bits, 0 < M <= 256, M % 8 == 0. e.g. uint32, uint8, uint256.
  • int<M>: two’s complement signed integer type of M bits, 0 < M <= 256, M % 8 == 0.
  • address: equivalent to uint160, except for the assumed interpretation and language typing. For computing the function selector, address is used.
  • uint, int: synonyms for uint256, int256 respectively. For computing the function selector, uint256 and int256 have to be used.
  • bool: equivalent to uint8 restricted to the values 0 and 1. For computing the function selector, bool is used.
  • fixed<M>x<N>: signed fixed-point decimal number of M bits, 8 <= M <= 256, M % 8 ==0, and 0 < N <= 80, which denotes the value v as v / (10 ** N).
  • ufixed<M>x<N>: unsigned variant of fixed<M>x<N>.
  • fixed, ufixed: synonyms for fixed128x19, ufixed128x19 respectively. For computing the function selector, fixed128x19 and ufixed128x19 have to be used.
  • bytes<M>: binary type of M bytes, 0 < M <= 32.
  • function: an address (20 bytes) folled by a function selector (4 bytes). Encoded identical to bytes24.

The following (fixed-size) array type exists:

  • <type>[M]: a fixed-length array of M elements, M > 0, of the given type.

The following non-fixed-size types exist:

  • bytes: dynamic sized byte sequence.
  • string: dynamic sized unicode string assumed to be UTF-8 encoded.
  • <type>[]: a variable-length array of elements of the given type.

Types can be combined to a tuple by enclosing a finite non-negative number of them inside parentheses, separated by commas:

  • (T1,T2,...,Tn): tuple consisting of the types T1, …, Tn, n >= 0

It is possible to form tuples of tuples, arrays of tuples and so on.


Solidity supports all the types presented above with the same names with the exception of tuples. The ABI tuple type is utilised for encoding Solidity structs.

Formal Specification of the Encoding

We will now formally specify the encoding, such that it will have the following properties, which are especially useful if some arguments are nested arrays:


  1. The number of reads necessary to access a value is at most the depth of the value inside the argument array structure, i.e. four reads are needed to retrieve a_i[k][l][r]. In a previous version of the ABI, the number of reads scaled linearly with the total number of dynamic parameters in the worst case.
  2. The data of a variable or array element is not interleaved with other data and it is relocatable, i.e. it only uses relative “addresses”

We distinguish static and dynamic types. Static types are encoded in-place and dynamic types are encoded at a separately allocated location after the current block.

Definition: The following types are called “dynamic”:

  • bytes
  • string
  • T[] for any T
  • T[k] for any dynamic T and any k > 0
  • (T1,...,Tk) if any Ti is dynamic for 1 <= i <= k

All other types are called “static”.

Definition: len(a) is the number of bytes in a binary string a. The type of len(a) is assumed to be uint256.

We define enc, the actual encoding, as a mapping of values of the ABI types to binary strings such that len(enc(X)) depends on the value of X if and only if the type of X is dynamic.

Definition: For any ABI value X, we recursively define enc(X), depending on the type of X being

  • (T1,...,Tk) for k >= 0 and any types T1, …, Tk

    enc(X) = head(X(1)) ... head(X(k-1)) tail(X(0)) ... tail(X(k-1))

    where X(i) is the ith component of the value, and head and tail are defined for Ti being a static type as

    head(X(i)) = enc(X(i)) and tail(X(i)) = "" (the empty string)

    and as

    head(X(i)) = enc(len(head(X(0)) ... head(X(k-1)) tail(X(0)) ... tail(X(i-1)))) tail(X(i)) = enc(X(i))

    otherwise, i.e. if Ti is a dynamic type.

    Note that in the dynamic case, head(X(i)) is well-defined since the lengths of the head parts only depend on the types and not the values. Its value is the offset of the beginning of tail(X(i)) relative to the start of enc(X).

  • T[k] for any T and k:

    enc(X) = enc((X[0], ..., X[k-1]))

    i.e. it is encoded as if it were a tuple with k elements of the same type.

  • T[] where X has k elements (k is assumed to be of type uint256):

    enc(X) = enc(k) enc([X[1], ..., X[k]])

    i.e. it is encoded as if it were an array of static size k, prefixed with the number of elements.

  • bytes, of length k (which is assumed to be of type uint256):

    enc(X) = enc(k) pad_right(X), i.e. the number of bytes is encoded as a

    uint256 followed by the actual value of X as a byte sequence, followed by the minimum number of zero-bytes such that len(enc(X)) is a multiple of 32.

  • string:

    enc(X) = enc(enc_utf8(X)), i.e. X is utf-8 encoded and this value is interpreted as of bytes type and encoded further. Note that the length used in this subsequent encoding is the number of bytes of the utf-8 encoded string, not its number of characters.

  • uint<M>: enc(X) is the big-endian encoding of X, padded on the higher-order (left) side with zero-bytes such that the length is 32 bytes.

  • address: as in the uint160 case

  • int<M>: enc(X) is the big-endian two’s complement encoding of X, padded on the higher-order (left) side with 0xff for negative X and with zero bytes for positive X such that the length is 32 bytes.

  • bool: as in the uint8 case, where 1 is used for true and 0 for false

  • fixed<M>x<N>: enc(X) is enc(X * 10**N) where X * 10**N is interpreted as a int256.

  • fixed: as in the fixed128x19 case

  • ufixed<M>x<N>: enc(X) is enc(X * 10**N) where X * 10**N is interpreted as a uint256.

  • ufixed: as in the ufixed128x19 case

  • bytes<M>: enc(X) is the sequence of bytes in X padded with trailing zero-bytes to a length of 32 bytes.

Note that for any X, len(enc(X)) is a multiple of 32.

Function Selector and Argument Encoding

All in all, a call to the function f with parameters a_1, ..., a_n is encoded as

function_selector(f) enc((a_1, ..., a_n))

and the return values v_1, ..., v_k of f are encoded as

enc((v_1, ..., v_k))

i.e. the values are combined into a tuple and encoded.


Given the contract:

pragma solidity ^0.4.16;

contract Foo {
  function bar(bytes3[2]) public pure {}
  function baz(uint32 x, bool y) public pure returns (bool r) { r = x > 32 || y; }
  function sam(bytes, bool, uint[]) public pure {}

Thus for our Foo example if we wanted to call baz with the parameters 69 and true, we would pass 68 bytes total, which can be broken down into:

  • 0xcdcd77c0: the Method ID. This is derived as the first 4 bytes of the Keccak hash of the ASCII form of the signature baz(uint32,bool).
  • 0x0000000000000000000000000000000000000000000000000000000000000045: the first parameter, a uint32 value 69 padded to 32 bytes
  • 0x0000000000000000000000000000000000000000000000000000000000000001: the second parameter - boolean true, padded to 32 bytes

In total:


It returns a single bool. If, for example, it were to return false, its output would be the single byte array 0x0000000000000000000000000000000000000000000000000000000000000000, a single bool.

If we wanted to call bar with the argument ["abc", "def"], we would pass 68 bytes total, broken down into:

  • 0xfce353f6: the Method ID. This is derived from the signature bar(bytes3[2]).
  • 0x6162630000000000000000000000000000000000000000000000000000000000: the first part of the first parameter, a bytes3 value "abc" (left-aligned).
  • 0x6465660000000000000000000000000000000000000000000000000000000000: the second part of the first parameter, a bytes3 value "def" (left-aligned).

In total:


If we wanted to call sam with the arguments "dave", true and [1,2,3], we would pass 292 bytes total, broken down into:

  • 0xa5643bf2: the Method ID. This is derived from the signature sam(bytes,bool,uint256[]). Note that uint is replaced with its canonical representation uint256.
  • 0x0000000000000000000000000000000000000000000000000000000000000060: the location of the data part of the first parameter (dynamic type), measured in bytes from the start of the arguments block. In this case, 0x60.
  • 0x0000000000000000000000000000000000000000000000000000000000000001: the second parameter: boolean true.
  • 0x00000000000000000000000000000000000000000000000000000000000000a0: the location of the data part of the third parameter (dynamic type), measured in bytes. In this case, 0xa0.
  • 0x0000000000000000000000000000000000000000000000000000000000000004: the data part of the first argument, it starts with the length of the byte array in elements, in this case, 4.
  • 0x6461766500000000000000000000000000000000000000000000000000000000: the contents of the first argument: the UTF-8 (equal to ASCII in this case) encoding of "dave", padded on the right to 32 bytes.
  • 0x0000000000000000000000000000000000000000000000000000000000000003: the data part of the third argument, it starts with the length of the array in elements, in this case, 3.
  • 0x0000000000000000000000000000000000000000000000000000000000000001: the first entry of the third parameter.
  • 0x0000000000000000000000000000000000000000000000000000000000000002: the second entry of the third parameter.
  • 0x0000000000000000000000000000000000000000000000000000000000000003: the third entry of the third parameter.

In total:


Use of Dynamic Types

A call to a function with the signature f(uint,uint32[],bytes10,bytes) with values (0x123, [0x456, 0x789], "1234567890", "Hello, world!") is encoded in the following way:

We take the first four bytes of sha3("f(uint256,uint32[],bytes10,bytes)"), i.e. 0x8be65246. Then we encode the head parts of all four arguments. For the static types uint256 and bytes10, these are directly the values we want to pass, whereas for the dynamic types uint32[] and bytes, we use the offset in bytes to the start of their data area, measured from the start of the value encoding (i.e. not counting the first four bytes containing the hash of the function signature). These are:

  • 0x0000000000000000000000000000000000000000000000000000000000000123 (0x123 padded to 32 bytes)
  • 0x0000000000000000000000000000000000000000000000000000000000000080 (offset to start of data part of second parameter, 4*32 bytes, exactly the size of the head part)
  • 0x3132333435363738393000000000000000000000000000000000000000000000 ("1234567890" padded to 32 bytes on the right)
  • 0x00000000000000000000000000000000000000000000000000000000000000e0 (offset to start of data part of fourth parameter = offset to start of data part of first dynamic parameter + size of data part of first dynamic parameter = 4*32 + 3*32 (see below))

After this, the data part of the first dynamic argument, [0x456, 0x789] follows:

  • 0x0000000000000000000000000000000000000000000000000000000000000002 (number of elements of the array, 2)
  • 0x0000000000000000000000000000000000000000000000000000000000000456 (first element)
  • 0x0000000000000000000000000000000000000000000000000000000000000789 (second element)

Finally, we encode the data part of the second dynamic argument, "Hello, world!":

  • 0x000000000000000000000000000000000000000000000000000000000000000d (number of elements (bytes in this case): 13)
  • 0x48656c6c6f2c20776f726c642100000000000000000000000000000000000000 ("Hello, world!" padded to 32 bytes on the right)

All together, the encoding is (newline after function selector and each 32-bytes for clarity):



Events are an abstraction of the Ethereum logging/event-watching protocol. Log entries provide the contract’s address, a series of up to four topics and some arbitrary length binary data. Events leverage the existing function ABI in order to interpret this (together with an interface spec) as a properly typed structure.

Given an event name and series of event parameters, we split them into two sub-series: those which are indexed and those which are not. Those which are indexed, which may number up to 3, are used alongside the Keccak hash of the event signature to form the topics of the log entry. Those which are not indexed form the byte array of the event.

In effect, a log entry using this ABI is described as:

  • address: the address of the contract (intrinsically provided by Ethereum);
  • topics[0]: keccak(EVENT_NAME+"("",")+")") (canonical_type_of is a function that simply returns the canonical type of a given argument, e.g. for uint indexed foo, it would return uint256). If the event is declared as anonymous the topics[0] is not generated;
  • topics[n]: EVENT_INDEXED_ARGS[n - 1] (EVENT_INDEXED_ARGS is the series of EVENT_ARGS that are indexed);
  • data: abi_serialise(EVENT_NON_INDEXED_ARGS) (EVENT_NON_INDEXED_ARGS is the series of EVENT_ARGS that are not indexed, abi_serialise is the ABI serialisation function used for returning a series of typed values from a function, as described above).

For all fixed-length Solidity types, the EVENT_INDEXED_ARGS array contains the 32-byte encoded value directly. However, for types of dynamic length, which include string, bytes, and arrays, EVENT_INDEXED_ARGS will contain the Keccak hash of the encoded value, rather than the encoded value directly. This allows applications to efficiently query for values of dynamic-length types (by setting the hash of the encoded value as the topic), but leaves applications unable to decode indexed values they have not queried for. For dynamic-length types, application developers face a trade-off between fast search for predetermined values (if the argument is indexed) and legibility of arbitrary values (which requires that the arguments not be indexed). Developers may overcome this tradeoff and achieve both efficient search and arbitrary legibility by defining events with two arguments — one indexed, one not — intended to hold the same value.


The JSON format for a contract’s interface is given by an array of function and/or event descriptions. A function description is a JSON object with the fields:

  • type: "function", "constructor", or "fallback" (the unnamed “default” function);
  • name: the name of the function;
  • inputs: an array of objects, each of which contains:
    • name: the name of the parameter;
    • type: the canonical type of the parameter (more below).
    • components: used for tuple types (more below).
  • outputs: an array of objects similar to inputs, can be omitted if function doesn’t return anything;
  • payable: true if function accepts ether, defaults to false;
  • stateMutability: a string with one of the following values: pure (specified to not read blockchain state), view (specified to not modify the blockchain state), nonpayable and payable (same as payable above).
  • constant: true if function is either pure or view

type can be omitted, defaulting to "function".

Constructor and fallback function never have name or outputs. Fallback function doesn’t have inputs either.

Sending non-zero ether to non-payable function will throw. Don’t do it.

An event description is a JSON object with fairly similar fields:

  • type: always "event"
  • name: the name of the event;
  • inputs: an array of objects, each of which contains:
    • name: the name of the parameter;
    • type: the canonical type of the parameter (more below).
    • components: used for tuple types (more below).
    • indexed: true if the field is part of the log’s topics, false if it one of the log’s data segment.
  • anonymous: true if the event was declared as anonymous.

For example,

pragma solidity ^0.4.0;

contract Test {
  function Test() public { b = 0x12345678901234567890123456789012; }
  event Event(uint indexed a, bytes32 b);
  event Event2(uint indexed a, bytes32 b);
  function foo(uint a) public { Event(a, b); }
  bytes32 b;

would result in the JSON:

"inputs": [{"name":"a","type":"uint256","indexed":true},{"name":"b","type":"bytes32","indexed":false}],
}, {
"inputs": [{"name":"a","type":"uint256","indexed":true},{"name":"b","type":"bytes32","indexed":false}],
}, {
"inputs": [{"name":"a","type":"uint256"}],
"outputs": []

Handling tuple types

Despite that names are intentionally not part of the ABI encoding they do make a lot of sense to be included in the JSON to enable displaying it to the end user. The structure is nested in the following way:

An object with members name, type and potentially components describes a typed variable. The canonical type is determined until a tuple type is reached and the string description up to that point is stored in type prefix with the word tuple, i.e. it will be tuple followed by a sequence of [] and [k] with integers k. The components of the tuple are then stored in the member components, which is of array type and has the same structure as the top-level object except that indexed is not allowed there.

As an example, the code

pragma solidity ^0.4.19;
pragma experimental ABIEncoderV2;

contract Test {
  struct S { uint a; uint[] b; T[] c; }
  struct T { uint x; uint y; }
  function f(S s, T t, uint a) public { }
  function g() public returns (S s, T t, uint a) {}

would result in the JSON:

    "name": "f",
    "type": "function",
    "inputs": [
        "name": "s",
        "type": "tuple",
        "components": [
            "name": "a",
            "type": "uint256"
            "name": "b",
            "type": "uint256[]"
            "name": "c",
            "type": "tuple[]",
            "components": [
                "name": "x",
                "type": "uint256"
                "name": "y",
                "type": "uint256"
        "name": "t",
        "type": "tuple",
        "components": [
            "name": "x",
            "type": "uint256"
            "name": "y",
            "type": "uint256"
        "name": "a",
        "type": "uint256"
    "outputs": []

Non-standard Packed Mode

Solidity supports a non-standard packed mode where:

  • no function selector is encoded,
  • types shorter than 32 bytes are neither zero padded nor sign extended and
  • dynamic types are encoded in-place and without the length.

As an example encoding int1, bytes1, uint16, string with values -1, 0x42, 0x2424, "Hello, world!" results in

  ^^                                 int1(-1)
    ^^                               bytes1(0x42)
      ^^^^                           uint16(0x2424)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^ string("Hello, world!") without a length field

More specifically, each statically-sized type takes as many bytes as its range has and dynamically-sized types like string, bytes or uint[] are encoded without their length field. This means that the encoding is ambiguous as soon as there are two dynamically-sized elements.