What is Candid?

Candid is an interface description language. Its primary purpose is to describe the public interface of a service, usually in the form of a program deployed as a canister that runs on the Internet Computer. One of the key benefits of Candid is that it is language-agnostic, and allows inter-operation between services and front-ends written in different programming languages, including Motoko, Rust, and JavaScript.

A typical interface description in Candid might look like this:

service counter : {
  add : (nat) -> ();
  subtract : (nat) -> ();
  get : () -> (int) query;
  subscribe : (func (int) -> ()) -> ();
}

In this example, the described service—counter—consists of the following public methods. - The add and subtract methods change the value of a counter. - The get method reads the current value of a counter. - The subscribe method can be used to invoke another function, for example, to invoke a notification callback method each time the counter value changes.

As this example illustrates, every method has a sequence of argument and result types. A method can also include annotations—like the query notation shown in this example—that are specific to the Internet Computer.

Given this simple interface description, it is possible for you to interact with this counter service directly from the command line or through a web-based front-end or programmatically from a Rust program or using through another programming or scripting language.

In addition to interoperability, Candid supports the evolution of service interfaces by precisely specifying the changes that can be made without breaking existing clients. For example, you can safely add new optional parameters to a service without losing compatibility for existing clients.

Why create a new IDL?

At first glance, you might think that other technologies, such as JSON, XML, or Protobuf, would suffice. However, Candid provides a unique combination of features that are not found in these other technologies. The features that make Candid particularly well-suited for developing applications for the Internet Computer include the following:

  • Many languages like JSON, XML, and Protobuf only describe how to map individual values to bytes or characters. These data description languages do not describe services as a whole. These languages focus on the data types you want to transfer instead of the methods that make use of those data types.

  • Candid implementations map the Candid value directly to types and values of the host language. With Candid, programmers do not construct or deconstruct some abstract Candid value.

  • Candid defines rules for how services and their interface can be upgraded in a sound and compositional way.

  • Candid is inherently a higher order languages. With Candid, you can pass more than plain data, including references to services and methods. Candid support for safe upgrades takes such higher-order use into account.

  • Candid has built-in support for specific Internet Computer features, such as the query annotation.

Candid types and values

Candid is a strongly typed system with a set of types that canonically cover most uses. It has:

  • Unbounded integral number types (nat, int).

  • Bounded integral number (nat8,nat16, nat32, nat64, int8,int16, int32, int64).

  • Floating point types (float32, float64).

  • The Boolean type (bool).

  • Types for textual (text) and binary (blob) data.

  • Container types, including variants (opt, vec, record, variant).

  • Reference types (service, func, principal).

  • The special null, reserved and empty types.

All types are described in detail in the Reference section.

The philosophy behind this set of types is that they are sufficient to describe the structure of data, so that information can be encoded, passed around and decoded, but intentionally do not describe semantic constraints beyond what’s needed to describe the representation. For example, there’s no way to express that a number should be even, that a vector has a certain length, or that the elements of a vector are sorted.

Candid supports this set of types to allow a natural mapping of data types based on reasonable, canonical choices suitable for each host language, whether you are writing your code in Motoko, Rust, JavaScript, or some other language.

Candid service descriptions

Once you are familiar with the Candid types, you can use them to describe a service. A Candid service description file—a .DID file—can either be written by hand or generated from a service implementation.

Before you explore how to generate service descriptions for a specific host language, let’s take a closer look at the structure of a sample service description and its constituent parts.

The simplest service description specifies a service with no public methods and would look like this:

service : {}

This service is not very useful, so let’s add a simple ping method:

service : {
  ping : () -> ();
}

This example describes a service that supports a single public method called ping. Method names can be arbitrary strings, and you can quote them ("method with spaces") if they are not plain identifiers.

Methods declare a sequence of arguments and result types. In the case of this ping method, no arguments are passed and no results are returned, so the empty sequence `() ` is used for both arguments and results.

Now that you’ve seen the simplest case, let’s consider a slightly more complex service description. This service consists of two methods—reverse and divMod—and each method include a sequence of argument and result types:

service : {
  reverse : (text) -> (text);
  divMod : (dividend : nat, divisor : nat) -> (div : nat, mod : nat);
}

The method reverse expects a single parameter of type text and returns one value of type text.

The method divMod expects and returns two values, all of type nat.

Naming arguments and results

In the previous example, the signature for the divMod method includes names for the argument and result values. Naming the arguments or results for a method is purely for documentation purposes. The name you use does not change the method’s type or the values being passed. Instead, arguments and results are identified by their position, independent of the name.

In particular, Candid does not prevent you from changing the type to:

  divMod : (dividend : nat, divisor : nat) -> (mod : nat, div : nat);

or passing the above divMod to a service expecting a method that returns mod first.

This is thus very different from named record fields, which are semantically relevant.

Reusing complex types

Often, multiple methods in a service may refer to the same complex type. In that case, the type can be named and reused multiple times. For example:

type address = record {
  street : text;
  city : text;
  zip_code : nat;
  country : text;
};
service address_book : {
  set_address: (name : text, addr : address) -> ();
  get_address: (name : text) -> (opt address) query;
}

These type definitions merely abbreviate an existing type, they do not define a new type. It does not matter whether you use address in the function signature, or write out the records. Also, two abbreviations with different names but equivalent definitions, describe the same type and are interchangeable. In other words, Candid uses structural typing.

Specifying a query method

In the last example, you might have noticed the use of the query annotation for the get_address method. For example:

service address_book : {
  set_address: (name : text, addr : address) -> ();
  get_address: (name : text) -> (opt address) query;
}

This annotation indicates that the get_address method can be invoked as an Internet Computer query call. As discussed in Query and update methods, a query provides an efficient way to retrieve information from a canister without going through consensus, so being able to identify a method as a query one of the key benefits of using Candid to interact with the Internet Computer.

Encoding and decoding

The point of Candid is to allow seamless invocation of service methods, passing arguments encoded to a binary format and transferred by an underlying transportation method (such as messages into or within the Internet Computer), and decoded on the other side.

As a Candid user, you do not have to worry about the details of this binary format. If you plan to implement Candid yourself (for example, to support a new host language), you can consult the Candid specification for details. However, some aspects of the format are worth knowing:

  • The Candid binary format starts with DIDL… (or, in hex, 4449444c…). If you see this in some low-level log output, you are very likely observing a Candid-encoded value.

  • The Candid binary format always encodes sequences of values, because methods parameters and results are sequences of types.

  • The binary format is quite compact. A (vec nat64) with 125 000 entries takes 1 000 007 bytes.

  • The binary is self-describing, and includes a (condensed) description of type of the values therein. This allows the receiving side to detect if a message was sent at a different, incompatible type.

  • As long as the sender serializes the arguments at the same type that the receiving side expects them, deserialization will succeed.

Service upgrades

Services evolve over time: They gain new methods, existing methods return more data, or expect additional arguments. Usually, service authors want to do that without breaking existing clients.

Candid supports such evolution by defining precise rules that indicate when the new service type will still be able to communicate with all other parties that are using the previous interface description. The underlying formalism is that of subtyping.

Services can safely evolve in the following ways:

  • New methods can be added.

  • Existing methods can return additional values, that is, the sequence of result types can be extended. Old clients will simply ignore additional values.

  • Existing methods can shorten their parameter list. Old clients may still send the extra arguments, but they will be ignored.

  • Existing methods can extend their parameter list with optional arguments (type opt …). When reading messages from old clients, who do not pass that argument, a null values is assumed.

  • Existing parameter types may be changed, but only to a supertype of the previous type.

  • Existing result types may be changed, but only to a subtype of the previous type.

For information about the supertypes and subtypes of a given type, see the corresponding reference section for that type.

Let’s look at a concrete example of how a service might evolve. Consider a service with the following API:

service counter : {
  add : (nat) -> ();
  subtract : (nat) -> ();
  get : () -> (int) query;
  subscribe : (func (int) -> ()) -> ();
}

This service can evolve to the following interface:

type timestamp = nat;
service counter : {
  set : (nat) -> ();
  add : (int) -> (new_val : nat);
  subtract : (nat, trap_on_underflow : opt bool) -> (new_val : nat);
  get : () -> (nat, last_change : timestamp) query;
  subscribe : (func (nat) -> (unregister : opt bool)) -> ();
}

Candid textual values

The main purpose of Candid is to connect programs written in some host language—Motoko, Rust, or JavaScript, for example—with the Internet Computer. In most cases, therefore, you do not have to deal with your program data as Candid values. Instead, you work with a host language like JavaScript using familiar JavaScript values then rely on Candid to transparently transport those values to a canister written in Rust or Motoko. The canister receive the values treats them as native Rust or Motoko values.

However, there are some cases—for example, when logging, debugging, or interacting with a service on the command-line—where it is useful to see the Candid values directly in a human-readable form. In these scenarios, you can use the textual presentation for Candid values.

The syntax is similar to that for Candid types. For example, a typical textual presentation for a Candid value might look like this:

(record {
  first_name = "John";
  last_name = "Doe";
  age = 14;
  membership_status = variant { active };
  email_addresses =
    vec { "[email protected]"; "[email protected]" };
})

The Candid binary format does not include the actual field names, merely numeric hashes. So pretty-printing such a value without knowledge of the expected type will not include the field names of records and variants. The above value might then be printed as follows:

(record {
   4846783 = 14;
   456245371 = variant {373703110};
   1443915007 = vec {"[email protected]"; "[email protected]"};
   2797692922 = "John"; 3046132756 = "Doe"
})

Generating service descriptions

In the section above, you learned how to write a Candid service description from scratch. But often, that is not even needed! Depending on the language you use to implement your service, you can get the Candid service description generated from your code.

For example, in Motoko, you can write a canister like this:

actor {
  var v : Int = 0;
  public func add(d : Nat) : async () { v += d; };
  public func subtract(d : Nat) : async () { v -= d; };
  public query func get() : async Int { v };
  public func subscribe(handler : func (Int) -> async ()) { … }
}

When you compile this program, the Motoko compiler automatically generates a Candid service description file with the interface shown above.

In other languages, like Rust or C, you can still develop your service using the types that are native to that language, for example, using native Rust types. After you develop a service in a language like Rust, however, there’s currently no way to automatically generate the service description in Candid. Therefore, if you write a program for a service in Rust or C, you need to write the Candid interface description manually following the conventions described in the Candid specification.

For examples of how to write Candid service descriptions for Rust programs, see the Rust CDK examples and the Rust tutorials.

Regardless of the host language you use, it is important to know the mapping between host language types and Candid types. In the Supported types reference section, you’ll find Candid type mapping described for Motoko, Rust, and JavaScript.