The `wit` format

The Wasm Interface Type (WIT) format is an IDL to provide tooling for the WebAssembly Component Model in two primary ways:

WIT is a developer-friendly format to describe the imports and exports to a component. It is easy to read and write and provides the foundational basis for producing components from guest languages as well as consuming components in host languages.
WIT packages are the basis of sharing types and definitions in an ecosystem of components. Authors can import types from other WIT packages when generating a component, publish a WIT package representing a host embedding, or collaborate on a WIT definition of a shared set of APIs between platforms.

A WIT package is a collection of WIT interfaces and worlds defined in files in the same directory that all use the file extension wit, for example foo.wit. Files are encoded as valid utf-8 bytes. Types can be imported between interfaces within a package using unqualified names and additionally from other packages through namespace- and-package-qualified names.

This document will go through the purpose of the syntactic constructs of a WIT document, a pseudo-formal grammar specification, and additionally a specification of the package format of a WIT package suitable for distribution.

See Gated Features for an explanation of 🔧.

Package Names

All WIT packages are assigned a package name. Package names look like foo:bar@1.0.0 and have three fields:

A namespace field, for example foo in foo:bar. This namespace is intended to disambiguate between registries, top-level organizations, etc. For example WASI interfaces use the wasi namespace.
A package field, for example clocks in wasi:clocks. A "package" groups together a set of interfaces and worlds that would otherwise be named with a common prefix.
An optional version field, specified as full semver.

🪺 With "nested namespaces and packages", package names are generalized to look like foo:bar:baz/quux, where bar is a nested namespace of foo and quux is a nested package of baz. See the package declaration section for more details.

Package names are specified at the top of a WIT file via a package declaration:

package wasi:clocks;

or

package wasi:clocks@1.2.0;

WIT packages can be defined in a collection of files. At least one of these files must specify a package name. Multiple files can specify the package, though they must all agree on what the package name is.

Additionally, many packages can be declared consecutively in one or more files, if the following nested package notation is used:

package local:a {
  interface foo {}
}

package local:b {
  interface bar {}
}

It is worth noting that defining nested packages does not remove the need for the "root" package declaration above. These nested package definitions simply provide the contents of other packages inline so that they don't have to be otherwise resolved via the filesystem or a registry.

Package names are used to generate the names of imports and exports in the Component Model's representation of interfaces and worlds as described below.

WIT Interfaces

The concept of an "interface" is central in WIT as a collection of functions and types. An interface can be thought of as an instance in the WebAssembly Component Model, for example a unit of functionality imported from the host or implemented by a component for consumption on a host. All functions and types belong to an interface.

An example of an interface is:

package local:demo;

interface host {
  log: func(msg: string);
}

represents an interface called host which provides one function, log, which takes a single string argument. If this were imported into a component then it would correspond to:

(component
  (import "local:demo/host" (instance $host
    (export "log" (func (param "msg" string)))
  ))
  ;; ...
)

An interface can contain use statements, type definitions, and function definitions. For example:

package wasi:filesystem;

interface types {
  use wasi:clocks/wall-clock.{datetime};

  record stat {
    ino: u64,
    size: u64,
    mtime: datetime,
    // ...
  }

  stat-file: func(path: string) -> result<stat>;
}

More information about use and types are described below, but this is an example of a collection of items within an interface. All items defined in an interface, including use items, are considered as exports from the interface. This means that types can further be used from the interface by other interfaces. An interface has a single namespace which means that none of the defined names can collide.

A WIT package can contain any number of interfaces listed at the top-level and in any order. The WIT validator will ensure that all references between interfaces are well-formed and acyclic.

WIT Worlds

WIT packages can contain world definitions at the top-level in addition to interface definitions. A world is a complete description of both imports and exports of a component. A world can be thought of as an equivalent of a component type in the component model. For example this world:

package local:demo;

world my-world {
  import host: interface {
    log: func(param: string);
  }

  export run: func();
}

can be thought of as this component type:

(type $my-world (component
  (import "host" (instance
    (export "log" (func (param "param" string)))
  ))
  (export "run" (func))
))

Worlds describe a concrete component and are the basis of bindings generation. A guest language will use a world to determine what functions are imported, what they're named, and what functions are exported, in addition to their names.

Worlds can contain any number of imports and exports, and can be either a function or an interface.

package local:demo;

world command {
  import wasi:filesystem/filesystem;
  import wasi:random/random;
  import wasi:clocks/monotonic-clock;
  // ...

  export main: func(args: list<string>);
}

More information about the wasi:random/random syntax is available below in the description of use.

An imported or exported interface corresponds to an imported or exported instance in the component model. Functions are equivalent to bare component functions. Additionally interfaces can be defined inline with an explicit plain name that avoids the need to have an out-of-line definition.

package local:demo;

interface out-of-line {
  the-function: func();
}

world your-world {
  import out-of-line;
  // ... is roughly equivalent to ...
  import out-of-line: interface {
    the-function: func();
  }
}

The plain name of an import or export statement is used as the plain name of the final component import or export definition.

In the component model imports to a component either use a plain or interface name, and in WIT this is reflected in the syntax:

package local:demo;

interface my-interface {
  // ..
}

world command {
  // generates an import of the name `local:demo/my-interface`
  import my-interface;

  // generates an import of the name `wasi:filesystem/types`
  import wasi:filesystem/types;

  // generates an import of the plain name `foo`
  import foo: func();

  // generates an import of the plain name `bar`
  import bar: interface {
    // ...
  }
}

Each name must be case-insensitively unique in the scope in which it is declared. In the case of worlds, all imported names are in the same scope, but separate from all the export names, and thus the same name can not be imported twice, but can be both imported and exported.

Union of Worlds with `include`

A World can be created by taking the union of two or more worlds. This operation allows world builders to form larger worlds from smaller worlds.

Below is a simple example of a world that includes two other worlds.

package local:demo;

// definitions of a, b, c, foo, bar, baz are omitted

world my-world-a {
    import a;
    import b;
    export c;
}

world my-world-b {
    import foo;
    import bar;
    export baz;
}

world union-my-world {
     include my-world-a;
     include my-world-b;
}

The include statement is used to include the imports and exports of another World to the current World. It says that the new World should be able to run all components that target the included worlds and more.

The union-my-world World defined above is equivalent to the following World:

world union-my-world {
    import a;
    import b;
    export c;
    import foo;
    import bar;
    export baz;
}

De-duplication of interfaces

If two worlds share an imported or exported interface name, then the union of the two worlds will only contain one copy of that imported or exported name. For example, the following two worlds union-my-world-a and union-my-world-b are equivalent:

package local:demo;

world my-world-a {
    import a1;
    import b1;
}

world my-world-b {
    import a1;
    import b1;
}

world union-my-world-a {
    include my-world-a;
    include my-world-b;
}

world union-my-world-b {
    import a1;
    import b1;
}

Name Conflicts and `with`

When two or more included Worlds have the same name for an import or export with a plain name, automatic de-duplication cannot be used (because the two same-named imports/exports might have different meanings in the different worlds) and thus the conflict has to be resolved manually using the with keyword.

The following example shows how to resolve name conflicts where union-my-world-a and union-my-world-b are equivalent:

package local:demo;

world world-one { import a: func(); }
world world-two { import a: func(); }

world union-my-world-a {
    include world-one;
    include world-two with { a as b }
}

world union-my-world-b {
  import a: func();
  import b: func();
}

with cannot be used to rename interface names, however, so the following world would be invalid:

package local:demo;

interface a {
    foo: func();
}

world world-using-a {
    import a;
}

world invalid-union-world {
    include my-using-a with { a as b }  // invalid: 'a', which is short for 'local:demo/a', is an interface name
}

A Note on Subtyping

In the future, when optional export is supported, the world author may explicitly mark exports as optional to make a component targeting an included World a subtype of the union World.

For now, we are not following the subtyping rules for the include statement. That is, the include statement does not imply any subtyping relationship between the included worlds and the union world.

WIT Packages and `use`

A WIT package represents a unit of distribution that can be published to a registry, for example, and used by other WIT packages. WIT packages are a flat list of interfaces and worlds defined in *.wit files. The current thinking for a convention is that projects will have a wit folder where all wit/*.wit files within describe a single package.

The purpose of the use statement is to enable sharing types between interfaces, even if they're defined outside of the current package in a dependency. The use statement can be used both within interfaces and worlds and at the top-level of a WIT file.

Interfaces, worlds, and `use`

A use statement inside of an interface or world block can be used to import types:

package local:demo;

interface types {
  enum errno { /* ... */ }

  type size = u32;
}

interface my-host-functions {
  use types.{errno, size};
}

The use target, types, is resolved within the scope of the package to an interface, in this case defined prior. Afterwards a list of types are provided as what's going to be imported with the use statement. The interface types may textually come either after or before the use directive's interface. Interfaces linked with use must be acyclic.

Names imported via use can be renamed as they're imported as well:

package local:demo;

interface my-host-functions {
  use types.{errno as my-errno};
}

This form of use is using a single identifier as the target of what's being imported, in this case types. The name types is first looked up within the scope of the current file, but it will additionally consult the package's namespace as well. This means that the above syntax still works if the interfaces are defined in sibling files:

// types.wit
interface types {
  enum errno { /* ... */ }

  type size = u32;
}

// host.wit
package local:demo;

interface my-host-functions {
  use types.{errno, size};
}

Here the types interface is not defined in host.wit but lookup will find it as it's defined in the same package, just instead in a different file. Since files are not ordered, but type definitions in the Component Model are ordered and acyclic, the WIT parser will perform an implicit topological sort of all parsed WIT definitions to find an acyclic definition order (or produce an error if there is none).

When importing or exporting an interface in a world the same syntax is used in import and export directives:

// a.wit
package local:demo;

world my-world {
  import host;

  export another-interface;
}

interface host {
  // ...
}

// b.wit
interface another-interface {
  // ...
}

When referring to an interface, a fully-qualified interface name can be used. For example, in this WIT document:

package local:demo;

world my-world {
  import wasi:clocks/monotonic-clock;
}

The monotonic-clock interface of the wasi:clocks package is being imported. This same syntax can be used in use as well:

package local:demo;

interface my-interface {
  use wasi:http/types.{request, response};
}

Top-level `use`

If a package being referred to has a version number, then using the above syntax so far it can get a bit repetitive to be referred to:

package local:demo;

interface my-interface {
  use wasi:http/types@1.0.0.{request, response};
}

world my-world {
  import wasi:http/handler@1.0.0;
  export wasi:http/handler@1.0.0;
}

To reduce repetition and to possibly help avoid naming conflicts the use statement can additionally be used at the top-level of a file to rename interfaces within the scope of the file itself. For example the above could be rewritten as:

package local:demo;

use wasi:http/types@1.0.0;
use wasi:http/handler@1.0.0;

interface my-interface {
  use types.{request, response};
}

world my-world {
  import handler;
  export handler;
}

The meaning of this and the previous world are the same, and use is purely a developer convenience for providing smaller names if necessary.

The interface referred to by a use is the name that is defined in the current file's scope:

package local:demo;

use wasi:http/types;   // defines the name `types`
use wasi:http/handler; // defines the name `handler`

Like with interface-level-use the as keyword can be used to rename the inferred name:

package local:demo;

use wasi:http/types as http-types;
use wasi:http/handler as http-handler;

Note that these can all be combined to additionally import packages with multiple versions and renaming as different WIT identifiers.

package local:demo;

use wasi:http/types@1.0.0 as http-types1;
use wasi:http/types@2.0.0 as http-types2;

// ...

Transitive imports and worlds

A use statement is not implemented by copying type information around but instead retains that it's a reference to a type defined elsewhere. This representation is plumbed all the way through to the final component, meaning that used types have an impact on the structure of the final generated component.

For example this document:

package local:demo;

interface shared {
  record metadata {
    // ...
  }
}

world my-world {
  import host: interface {
    use shared.{metadata};

    get: func() -> metadata;
  }
}

would generate this component:

(component
  (import "local:demo/shared" (instance $shared
    (type $metadata (record (; ... ;)))
    (export "metadata" (type (eq $metadata)))
  ))
  (alias export $shared "metadata" (type $metadata_from_shared))
  (import "host" (instance $host
    (export $metadata_in_host "metadata" (type (eq $metadata_from_shared)))
    (export "get" (func (result $metadata_in_host)))
  ))
)

Here it can be seen that despite the world only listing host as an import the component additionally imports a local:demo/shared interface. This is due to the fact that the use shared.{ ... } implicitly requires that shared is imported into the component as well.

Note that the name "local:demo/shared" here is derived from the name of the interface plus the package name local:demo.

For exported interfaces, any transitively used interface is assumed to be an import unless it's explicitly listed as an export. For example, here w1 is equivalent to w2:

interface a {
  resource r;
}
interface b {
  use a.{r};
  foo: func() -> r;
}

world w1 {
  export b;
}
world w2 {
  import a;
  export b;
}

Note: It's planned in the future to have "power user syntax" to configure this on a more fine-grained basis for exports, for example being able to configure that a use'd interface is a particular import or a particular export.

WIT Functions

Functions are defined in an interface or are listed as an import or export from a world. Parameters to a function must all be named and have case-insensitively unique names:

package local:demo;

interface foo {
  a1: func();
  a2: func(x: u32);
  a3: func(y: u64, z: f32);
}

Functions can return at most one unnamed type:

package local:demo;

interface foo {
  a1: func() -> u32;
  a2: func() -> string;
}

And functions can also return multiple types by naming them:

package local:demo;

interface foo {
  a: func() -> (a: u32, b: f32);
}

Note that returning multiple values from a function is not equivalent to returning a tuple of values from a function. These options are represented distinctly in the component binary format.

WIT Types

Types in WIT files can only be defined in interfaces at this time. The types supported in WIT is the same set of types supported in the component model itself:

package local:demo;

interface foo {
  // "package of named fields"
  record r {
    a: u32,
    b: string,
  }

  // values of this type will be one of the specified cases
  variant human {
    baby,
    child(u32), // optional type payload
    adult,
  }

  // similar to `variant`, but no type payloads
  enum errno {
    too-big,
    too-small,
    too-fast,
    too-slow,
  }

  // a bitflags type
  flags permissions {
    read,
    write,
    exec,
  }

  // type aliases are allowed to primitive types and additionally here are some
  // examples of other types
  type t1 = u32;
  type t2 = tuple<u32, u64>;
  type t3 = string;
  type t4 = option<u32>;
  type t5 = result<_, errno>;           // no "ok" type
  type t6 = result<string>;             // no "err" type
  type t7 = result<char, errno>;        // both types specified
  type t8 = result;                     // no "ok" or "err" type
  type t9 = list<string>;
  type t10 = t9;
}

The record, variant, enum, and flags types must all have names associated with them. The list, option, result, tuple, and primitive types do not need a name and can be mentioned in any context. This restriction is in place to assist with code generation in all languages to leverage language-builtin types where possible while accommodating types that need to be defined within each language as well.

WIT Identifiers

Identifiers in WIT can be defined with two different forms. The first is the kebab-case label production in the Component Model text format.

foo: func(bar: u32);

red-green-blue: func(r: u32, g: u32, b: u32);

resource XML { ... }
parse-XML-document: func(s: string) -> XML;

This form can't lexically represent WIT keywords, so the second form is the same syntax with the same restrictions as the first, but prefixed with '%':

%foo: func(%bar: u32);

%red-green-blue: func(%r: u32, %g: u32, %b: u32);

// This form also supports identifiers that would otherwise be keywords.
%variant: func(%enum: s32);

Lexical structure

The wit format is a curly-braced-based format where whitespace is optional (but recommended). A wit document is parsed as a unicode string, and when stored in a file is expected to be encoded as utf-8.

Additionally, wit files must not contain any bidirectional override scalar values, control codes other than newline, carriage return, and horizontal tab, or codepoints that Unicode officially deprecates or strongly discourages.

The current structure of tokens are:

token ::= whitespace
        | operator
        | keyword
        | integer
        | identifier

Whitespace and comments are ignored when parsing structures defined elsewhere here.

Whitespace

A whitespace token in wit is a space, a newline, a carriage return, a tab character, or a comment:

whitespace ::= ' ' | '\n' | '\r' | '\t' | comment

Comments

A comment token in wit is either a line comment preceded with // which ends at the next newline (\n) character or it's a block comment which starts with /* and ends with */. Note that block comments are allowed to be nested and their delimiters must be balanced

comment ::= '//' character-that-isnt-a-newline*
          | '/*' any-unicode-character* '*/'

Operators

There are some common operators in the lexical structure of wit used for various constructs. Note that delimiters such as { and ( must all be balanced.

operator ::= '=' | ',' | ':' | ';' | '(' | ')' | '{' | '}' | '<' | '>' | '*' | '->' | '/' | '.' | '@'

Keywords

Certain identifiers are reserved for use in WIT documents and cannot be used bare as an identifier. These are used to help parse the format, and the list of keywords is still in flux at this time but the current set is:

keyword ::= 'as'
          | 'bool'
          | 'borrow'
          | 'char'
          | 'constructor'
          | 'enum'
          | 'export'
          | 'f32'
          | 'f64'
          | 'flags'
          | 'from'
          | 'func'
          | 'future'
          | 'import'
          | 'include'
          | 'interface'
          | 'list'
          | 'option'
          | 'own'
          | 'package'
          | 'record'
          | 'resource'
          | 'result'
          | 's16'
          | 's32'
          | 's64'
          | 's8'
          | 'static'
          | 'stream'
          | 'string'
          | 'tuple'
          | 'type'
          | 'u16'
          | 'u32'
          | 'u64'
          | 'u8'
          | 'use'
          | 'variant'
          | 'with'
          | 'world'

Integers

Integers are currently only used for package versions and are a contiguous sequence of digits:

integer ::= [0-9]+

Top-level items

A wit document is a sequence of items specified at the top level. These items come one after another and it's recommended to separate them with newlines for readability but this isn't required.

Concretely, the structure of a wit file is:

wit-file ::= package-decl? (package-items | nested-package-definition)*

nested-package-definition ::= package-decl '{' package-items* '}'

package-items ::= toplevel-use-item | interface-item | world-item

Essentially, these top level items are worlds, interfaces, use statements and other package defintions.

Feature Gates

Various WIT items can be "gated", to reflect the fact that the item is part of an unstable feature, that the item was added as part of a minor version update and shouldn't be used when targeting an earlier minor version, or that a feature has been deprecated and should no longer be used.

For example, the following interface has 4 items, 3 of which are gated:

interface foo {
  a: func();

  @since(version = 0.2.1)
  b: func();

  @since(version = 0.2.2, feature = fancy-foo)
  c: func();

  @unstable(feature = fancier-foo)
  d: func();

  @since(version = 0.2.0)
  @deprecated(version = 0.2.2)
  e: func();
}

The @since gate indicates that b and c were added as part of the 0.2.1 and 0.2.2 releases, resp. Thus, when building a component targeting, e.g., 0.2.1, b can be used, but c cannot. An important expectation set by the @since gate is that, once applied to an item, the item is not modified incompatibly going forward (according to general semantic versioning rules).

In contrast, the @unstable gate on d indicates that d is part of the fancier-foo feature that is still under active development and thus d may change type or be removed at any time. An important expectation set by the @unstable gate is that toolchains will not expose @unstable features by default unless explicitly opted-into by the developer.

Finally, the @deprecated gate on e indicates that e should no longer be used starting version 0.2.2. Both toolchains and host runtimes may warn users if they detect an @deprecated API is being used. An @deprecated gate is required to always be paired up with either a @since or @deprecated gate.

Together, these gates support a development flow in which new features start with an @unstable gate while the details are still being hashed out. Then, once the feature is stable (and, in a WASI context, voted upon), the @unstable gate is switched to a @since gate.

Thus, c is enabled if the version is 0.2.2 or newer or the fancy-foo feature is explicitly enabled by the developer. The feature field can be removed once producer toolchains have updated their default version to enable the feature by default.

Feature gate syntax

The grammar that governs feature gate syntax is:

gate ::= gate-item*
gate-item ::= unstable-gate
            | since-gate
            | deprecated-gate

unstable-gate ::= '@unstable' '(' feature-field ')'
since-gate ::= '@since' '(' version-field ')'
deprecated-gate ::= '@deprecated' '(' version-field ')'

feature-field ::= 'feature' '=' id
version-field ::= 'version' '=' <valid semver>

Rules for feature gate usage

As part of WIT validation, any item that refers to another gated item must also be compatibly gated. For example, this is an error:

interface i {
  @since(version = 1.0.1)
  type t1 = u32;

  type t2 = t1; // error
}

Additionally, if an item is contained by a gated item, it must also be compatibly gated. For example, this is an error:

@since(version = 1.0.2)
interface i {
  foo: func();  // error: no gate

  @since(version = 1.0.1)
  bar: func();  // also error: weaker gate
}

The following rules apply to the use of feature gates:

Either @since or @unstable should be used, but not both (exclusive or).
If a package contains a feature gate, it's version must be specified (i.e. namespace:package@x.y.z)

Scenario: Stabilization of a new feature

This section lays out the basic flow and expected usage of feature gate machinery when stabilizing new features and deprecating old ones.

Assume the following WIT package as the initial interface:

package examples:fgates-calc@0.1.0;

@since(version = 0.1.0)
interface calc {
  @since(version = 0.1.0)
  variant calc-error {
    integer-overflow,
    integer-underflow,
    unexpected,
  }

  @since(version = 0.1.0)
  add: func(x: i32, y: i32) -> result<i32, calc-error>;
}

First, add new items under an @unstable annotation with a feature specified:

package examples:fgates-calc@0.1.1;

@since(version = 0.1.0)
interface calc {
  @since(version = 0.1.0)
  variant calc-error {
    integer-overflow,
    integer-underflow,
    unexpected,
  }

  @since(version = 0.1.0)
  add: func(x: i32, y: i32) -> result<i32, calc-error>;

  /// By convention, feature flags should be prefixed with package name to reduce chance of collisions
  ///
  /// see: https://github.com/WebAssembly/WASI/blob/main/Contributing.md#filing-changes-to-existing-phase-3-proposals
  @unstable(feature = fgates-calc-minus)
  sub: func(x: i32, y: i32) -> result<i32, calc-error>;
}

At this point, consumers of the WIT can enable feature fgates-calc-minus through their relevant tooling and get access to the sub function.

Note that, at least until subtyping is relaxed in the Component Model, if we had to add a new case to calc-error, this would be a breaking change and require either a new major version or adding a second, distinct variant definition used by new functions.

Second, when the feature is ready to be stabilized, switch to a @since annotation:

package examples:fgates-calc@0.1.2;

@since(version = 0.1.0)
interface calc {
  @since(version = 0.1.0)
  variant calc-error {
    integer-overflow,
    integer-underflow,
    unexpected,
  }

  @since(version = 0.1.0)
  add: func(x: i32, y: i32) -> result<i32, calc-error>;

  @since(version = 0.1.2)
  sub: func(x: i32, y: i32) -> result<i32, calc-error>;
}

Scenario: Deprecation of an existing stable feature

This section lays out the basic flow and expected usage of feature gate machinery when stabilizing a new feature.

Assume the following WIT package as the initial interface:

package examples:fgates-deprecation@0.1.1;

@since(version = 0.1.0)
interface calc {
  @since(version = 0.1.0)
  variant calc-error {
    integer-overflow,
    integer-underflow,
    unexpected,
  }

  @since(version = 0.1.0)
  add-one: func(x: i32) -> result<i32, calc-error>;

  @since(version = 0.1.1)
  add: func(x: i32, y: i32) -> result<i32, calc-error>;
}

First: Add the @deprecated annotation to the relevant item in a new version

package examples:fgates-deprecation@0.1.2;

@since(version = 0.1.0)
interface calc {
  @since(version = 0.1.0)
  variant calc-error {
    integer-overflow,
    integer-underflow,
    unexpected,
  }

  @deprecated(version = 0.1.2)
  add-one: func(x: i32) -> result<i32, calc-error>;

  @since(version = 0.1.1)
  add: func(x: i32, y: i32) -> result<i32, calc-error>;
}

At this point, tooling consuming this WIT will be able to appropriately alert users to the now-deprecated add-one function.

Second: completely remove the deprecated item in some future SemVer-compliant major version

package examples:fgates-deprecation@0.2.0;

@since(version = 0.1.0)
interface calc {
  @since(version = 0.1.0)
  variant calc-error {
    integer-overflow,
    integer-underflow,
    unexpected,
  }

  @since(version = 0.1.1)
  add: func(x: i32, y: i32) -> result<i32, calc-error>;
}

In this new "major" version (this is considered a major version under SemVer 0.X rules) -- the add-one function can be fully removed.

Package declaration

WIT files optionally start with a package declaration which defines the name of the package.

package-decl        ::= 'package' ( id ':' )+ id ( '/' id )* ('@' valid-semver)?  ';'

The production valid-semver is as defined by Semantic Versioning 2.0 and optional.

Item: `toplevel-use`

A use statement at the top-level of a file can be used to bring interfaces into the scope of the current file and/or rename interfaces locally for convenience:

toplevel-use-item ::= 'use' use-path ('as' id)? ';'

use-path ::= id
           | id ':' id '/' id ('@' valid-semver)?
           | ( id ':' )+ id ( '/' id )+ ('@' valid-semver)? 🪺

Here use-path is an interface name. The bare form id refers to interfaces defined within the current package, and the full form refers to interfaces in package dependencies.

The as syntax can be optionally used to specify a name that should be assigned to the interface. Otherwise the name is inferred from use-path.

As a future extension, WIT, components and component registries may allow nesting both namespaces and packages, which would then generalize the syntax of use-path as suggested by the 🪺 suffixed rule.

Item: `world`

Worlds define a componenttype as a collection of imports and exports, all of which can be gated.

Concretely, the structure of a world is:

world-item ::= gate 'world' id '{' world-items* '}'

world-items ::= gate world-definition

world-definition ::= export-item
                   | import-item
                   | use-item
                   | typedef-item
                   | include-item

export-item ::= 'export' id ':' extern-type
              | 'export' use-path ';'
import-item ::= 'import' id ':' extern-type
              | 'import' use-path ';'

extern-type ::= func-type ';' | 'interface' '{' interface-items* '}'

Note that worlds can import types and define their own types to be exported from the root of a component and used within functions imported and exported. The interface item here additionally defines the grammar for IDs used to refer to interface items.

Item: `include`

A include statement enables the union of the current world with another world. The structure of an include statement is:

include wasi:io/my-world-1 with { a as a1, b as b1 };
include my-world-2;

include-item ::= 'include' use-path ';'
               | 'include' use-path 'with' '{' include-names-list '}'

include-names-list ::= include-names-item
                     | include-names-list ',' include-names-item

include-names-item ::= id 'as' id

Item: `interface`

Interfaces can be defined in a wit file. Interfaces have a name and a sequence of items and functions, all of which can be gated.

Specifically interfaces have the structure:

Note: The symbol ε, also known as Epsilon, denotes an empty string.

interface-item ::= gate 'interface' id '{' interface-items* '}'

interface-items ::= gate interface-definition

interface-definition ::= typedef-item
                       | use-item
                       | func-item

typedef-item ::= resource-item
               | variant-items
               | record-item
               | flags-items
               | enum-items
               | type-item

func-item ::= id ':' func-type ';'

func-type ::= 'func' param-list result-list

param-list ::= '(' named-type-list ')'

result-list ::= ϵ
              | '->' ty

named-type-list ::= ϵ
                  | named-type ( ',' named-type )*

named-type ::= id ':' ty

Item: `use`

A use statement enables importing type or resource definitions from other wit packages or interfaces. The structure of a use statement is:

use an-interface.{a, list, of, names}
use my:dependency/the-interface.{more, names as foo}

Specifically the structure of this is:

use-item ::= 'use' use-path '.' '{' use-names-list '}' ';'

use-names-list ::= use-names-item
                 | use-names-item ',' use-names-list?

use-names-item ::= id
                 | id 'as' id

Note: Here use-names-list? means at least one use-name-list term.

Items: type

There are a number of methods of defining types in a wit package, and all of the types that can be defined in wit are intended to map directly to types in the component model.

Item: `type` (alias)

A type statement declares a new named type in the wit document. This name can be later referred to when defining items using this type. This construct is similar to a type alias in other languages

type my-awesome-u32 = u32;
type my-complicated-tuple = tuple<u32, s32, string>;

Specifically the structure of this is:

type-item ::= 'type' id '=' ty ';'

Item: `record` (bag of named fields)

A record statement declares a new named structure with named fields. Records are similar to a struct in many languages. Instances of a record always have their fields defined.

record pair {
    x: u32,
    y: u32,
}

record person {
    name: string,
    age: u32,
    has-lego-action-figure: bool,
}

Specifically the structure of this is:

record-item ::= 'record' id '{' record-fields '}'

record-fields ::= record-field
                | record-field ',' record-fields?

record-field ::= id ':' ty

Item: `flags` (bag-of-bools)

A flags represents a bitset structure with a name for each bit. The flags type is represented as a bit flags representation in the canonical ABI.

flags properties {
    lego,
    marvel-superhero,
    supervillan,
}

Specifically the structure of this is:

flags-items ::= 'flags' id '{' flags-fields '}'

flags-fields ::= id
               | id ',' flags-fields?

Item: `variant` (one of a set of types)

A variant statement defines a new type where instances of the type match exactly one of the variants listed for the type. This is similar to a "sum" type in algebraic datatypes (or an enum in Rust if you're familiar with it). Variants can be thought of as tagged unions as well.

Each case of a variant can have an optional type associated with it which is present when values have that particular case's tag.

All variant type must have at least one case specified.

variant filter {
    all,
    none,
    some(list<string>),
}

Specifically the structure of this is:

variant-items ::= 'variant' id '{' variant-cases '}'

variant-cases ::= variant-case
                | variant-case ',' variant-cases?

variant-case ::= id
               | id '(' ty ')'

Item: `enum` (variant but with no payload)

An enum statement defines a new type which is semantically equivalent to a variant where none of the cases have a payload type. This is special-cased, however, to possibly have a different representation in the language ABIs or have different bindings generated in for languages.

enum color {
    red,
    green,
    blue,
    yellow,
    other,
}

Specifically the structure of this is:

enum-items ::= 'enum' id '{' enum-cases '}'

enum-cases ::= id
             | id ',' enum-cases?

Item: `resource`

A resource statement defines a new abstract type for a resource, which is an entity with a lifetime that can only be passed around indirectly via handle values. Resource types are used in interfaces to describe things that can't or shouldn't be copied by value.

For example, the following Wit defines a resource type and a function that takes and returns a handle to a blob:

resource blob;
transform: func(blob) -> blob;

As syntactic sugar, resource statements can also declare any number of methods, which are functions that implicitly take a self parameter that is a handle. A resource statement can also contain any number of static functions, which do not have an implicit self parameter but are meant to be lexically nested in the scope of the resource type. Lastly, a resource statement can contain at most one constructor function, which is syntactic sugar for a function returning a handle of the containing resource type.

For example, the following resource definition:

resource blob {
    constructor(init: list<u8>);
    write: func(bytes: list<u8>);
    read: func(n: u32) -> list<u8>;
    merge: static func(lhs: borrow<blob>, rhs: borrow<blob>) -> blob;
}

desugars into:

resource blob;
%[constructor]blob: func(init: list<u8>) -> blob;
%[method]blob.write: func(self: borrow<blob>, bytes: list<u8>);
%[method]blob.read: func(self: borrow<blob>, n: u32) -> list<u8>;
%[static]blob.merge: func(lhs: borrow<blob>, rhs: borrow<blob>) -> blob;

These %-prefixed names embed the resource type name so that bindings generators can generate idiomatic syntax for the target language or (for languages like C) fall back to an appropriately-prefixed free function name.

When a resource type name is used directly (e.g. when blob is used as the return value of the constructor above), it stands for an "owning" handle that will call the resource's destructor when dropped. When a resource type name is wrapped with borrow<...>, it stands for a "borrowed" handle that will not call the destructor when dropped. As shown above, methods always desugar to a borrowed self parameter whereas constructors always desugar to an owned return value.

Specifically, the syntax for a resource definition is:

resource-item ::= 'resource' id ';'
                | 'resource' id '{' resource-method* '}'
resource-method ::= func-item
                  | id ':' 'static' func-type ';'
                  | 'constructor' param-list ';'

The syntax for handle types is presented below.

Types

As mentioned previously the intention of wit is to allow defining types corresponding to the interface types specification. Many of the top-level items above are introducing new named types but "anonymous" types are also supported, such as built-ins. For example:

type number = u32;
type fallible-function-result = result<u32, string>;
type headers = list<string>;

Specifically the following types are available:

ty ::= 'u8' | 'u16' | 'u32' | 'u64'
     | 's8' | 's16' | 's32' | 's64'
     | 'f32' | 'f64'
     | 'char'
     | 'bool'
     | 'string'
     | tuple
     | list
     | option
     | result
     | handle
     | id

tuple ::= 'tuple' '<' tuple-list '>'
tuple-list ::= ty
             | ty ',' tuple-list?

list ::= 'list' '<' ty '>'
       | 'list' '<' ty ',' uint '>' 🔧

uint ::= [1-9][0-9]*

option ::= 'option' '<' ty '>'

result ::= 'result' '<' ty ',' ty '>'
         | 'result' '<' '_' ',' ty '>'
         | 'result' '<' ty '>'
         | 'result'

The tuple type is semantically equivalent to a record with numerical fields, but it frequently can have language-specific meaning so it's provided as a first-class type.

🔧 A list with a fixed length provides the low-level memory representation of a homogeneous tuple of the same length, but with the dynamic indexing of a list. E.g., the following two functions have the same low-level (Core WebAssembly) representation, but will naturally produce different source-level bindings:

get-ipv4-address1: func() -> list<u8, 4>;
get-ipv4-address2: func() -> tuple<u8, u8, u8, u8>;

The option and result types are semantically equivalent to the variants:

variant option {
    none,
    some(ty),
}

variant result {
    ok(ok-ty),
    err(err-ty),
}

These types are so frequently used and frequently have language-specific meanings though so they're also provided as first-class types.

Finally the last case of a ty is simply an id which is intended to refer to another type or resource defined in the document. Note that definitions can come through a use statement or they can be defined locally.

Handles

There are two types of handles in Wit: "owned" handles and "borrowed" handles. Owned handles represent the passing of unique ownership of a resource between two components. When the owner of an owned handle drops that handle, the resource is destroyed. In contrast, a borrowed handle represents a temporary loan of a handle from the caller to the callee for the duration of the call.

The syntax for handles is:

handle ::= id
         | 'borrow' '<' id '>'

The id case denotes an owned handle, where id is the name of a preceding resource item. Thus, the "default" way that resources are passed between components is via transfer of unique ownership.

The resource method syntax defined above is syntactic sugar that expands into separate function items that take a first parameter named self of type borrow. For example, the compound definition:

resource file {
    read: func(n: u32) -> list<u8>;
}

is expanded into:

resource file
%[method]file.read: func(self: borrow<file>, n: u32) -> list<u8>;

where %[method]file.read is the desugared name of a method according to the Component Model's definition of name.

Name resolution

A wit document is resolved after parsing to ensure that all names resolve correctly. For example this is not a valid wit document:

type foo = bar;  // ERROR: name `bar` not defined

Type references primarily happen through the id production of ty.

Additionally names in a wit document can only be defined once:

type foo = u32;
type foo = u64;  // ERROR: name `foo` already defined

Names do not need to be defined before they're used (unlike in C or C++), it's ok to define a type after it's used:

type foo = bar;

record bar {
    age: u32,
}

Types, however, cannot be recursive:

type foo = foo;  // ERROR: cannot refer to itself

record bar1 {
    a: bar2,
}

record bar2 {
    a: bar1,    // ERROR: record cannot refer to itself
}

Package Format

Each top-level WIT definition can be compiled into a single canonical Component Model type definition that captures the result of performing the type resolution described above. These Component Model types can then be exported by a component along with other sorts of exports, allowing a single component to package both runtime functionality and development-time WIT interfaces. Thus, WIT does not need its own separate package format; WIT can be packaged as a component binary.

Using component binaries to package WIT in this manner has several advantages:

We get to reuse the binary format of components, especially the tricky type bits.
Downstream tooling does not need to replicate the resolution logic nor the resolution environment (directories, registries, paths, arguments, etc) of the WIT package producer; it can reuse the simpler compiled result.
Many aspects of the WIT syntax can evolve over time without breaking downstream tooling, similar to what has happened with the Core WebAssembly WAT text format over time.
When components are published in registries and assigned names (see the discussion of naming in Import and Export Definitions), WIT interfaces and worlds can be published with the same tooling and named using the same namespace:package/export naming scheme.
A single package can both contain an implementation and a collection of interface and world definitions that are imported by that implementation (e.g., an engine component can define and exports its own plugin world).

As a first example, the following WIT:

package local:demo;

interface types {
  resource file {
    read: func(off: u32, n: u32) -> list<u8>;
    write: func(off: u32, bytes: list<u8>);
  }
}

interface namespace {
  use types.{file};
  open: func(name: string) -> file;
}

can be packaged into a component as:

(component
  (type (export "types") (component
    (export "local:demo/types" (instance
      (export $file "file" (type (sub resource)))
      (export "[method]file.read" (func
        (param "self" (borrow $file)) (param "off" u32) (param "n" u32)
        (result (list u8))
      ))
      (export "[method]file.write" (func
        (param "self" (borrow $file))
        (param "bytes" (list u8))
      ))
    ))
  ))
  (type (export "namespace") (component
    (import "local:demo/types" (instance $types
      (export "file" (type (sub resource)))
    ))
    (alias export $types "file" (type $file))
    (export "local:demo/namespace" (instance
      (export "open" (func (param "name" string) (result (own $file))))
    ))
  ))
)

This example illustrates the basic structure of interfaces:

Each top-level WIT definition (in this example: types and namespace) turns into a type export of the same kebab-name.
Each WIT interface is mapped to a component-type that exports an instance with a fully-qualified interface name (in this example: local:demo/types and local:demo/namespace). Note that this nested scheme allows a single component to both define and implement a WIT interface without name conflict.
The wrapping component-type has an import for every use in the interface, bringing any used types into scope so that they can be aliased when building the instance-type. The component-type can be thought of as "parameterizing" the interface's compiled instance type (∀T.{instance type}). Note that there is always an outer wrapping component-type, even when the interface contains no uses.

One useful consequence of this encoding scheme is that each top-level definition is self-contained and valid (according to Component Model validation rules) independent of each other definition. This allows packages to be trivially split or unioned (assuming the result doesn't have to be a valid package, but rather just a raw list of non-exported type definitions).

Another expectation is that, when a component containing WIT definitions is published to a registry, the registry validates that the fully-qualified WIT interface names inside the component are consistent with the registry-assigned package name. For example, the above component would only be valid if published with package name local:demo; any other package name would be inconsistent with the internal local:demo/types and local:demo/namespace exported interface names.

Inter-package references are structurally no different than intra-package references other than the referenced WIT definition is not present in the component. For example, the following WIT:

package local:demo

interface foo {
  use wasi:http/types.{request};
  frob: func(r: request) -> request;
}

is encoded as:

(component
  (type (export "foo") (component
    (import "wasi:http/types" (instance $types
      (export "request" (type (sub resource)))
    ))
    (alias export $types "request" (type $request))
    (export "local:demo/foo" (instance
      (export "frob" (func (param "r" (own $request)) (result (own $request))))
    ))
  ))
)

Worlds are encoded similarly to interfaces, but replace the inner exported instance with an inner exported component. For example, this WIT:

package local:demo;

world the-world {
  export test: func();
  export run: func();
}

is encoded as:

(component
  (type (export "the-world") (component
    (export "local:demo/the-world" (component
      (export "test" (func))
      (export "run" (func))
    ))
  ))
)

In the current version of WIT, the outer wrapping component-type will only ever contain a single export and thus only serves to separate the kebab-name export from the inner exported interface name and to provide consistency with the encoding of interface shown above.

When a world imports or exports an interface, to produce a valid component-type, the interface's compiled instance-type ends up getting copied into the component-type. For example, the following WIT:

package local:demo;

world the-world {
  import console;
}

interface console {
  log: func(arg: string);
}

is encoded as:

(component
  (type (export "the-world") (component
    (export "local:demo/the-world" (component
      (import "local:demo/console" (instance
        (export "log" (func (param "arg" string)))
      ))
    ))
  ))
  (type (export "console") (component
    (export "local:demo/console" (instance
      (export "log" (func (param "arg" string)))
    ))
  ))
)

This duplication is useful in the case of cross-package references or split packages, allowing a compiled world definition to be fully self-contained and able to be used to compile a component without additional type information.

Putting this all together, the following WIT definitions:

// wasi-http repo

// wit/types.wit
interface types {
  resource request { ... }
  resource response { ... }
}

// wit/handler.wit
interface handler {
  use types.{request, response};
  handle: func(r: request) -> response;
}

// wit/proxy.wit
package wasi:http;

world proxy {
  import wasi:logging/logger;
  import handler;
  export handler;
}

are encoded as:

(component
  (type (export "types") (component
    (export "wasi:http/types" (instance
      (export "request" (type (sub resource)))
      (export "response" (type (sub resource)))
      ...
    ))
  ))
  (type (export "handler") (component
    (import "wasi:http/types" (instance $http-types
      (export "request" (type (sub resource)))
      (export "response" (type (sub resource)))
    ))
    (alias export $http-types "request" (type $request))
    (alias export $http-types "response" (type $response))
    (export "wasi:http/handler" (instance
      (export "handle" (func (param "r" (own $request)) (result (own $response))))
    ))
  ))
  (type (export "proxy") (component
    (export "wasi:http/proxy" (component
      (import "wasi:logging/logger" (instance
        ...
      ))
      (import "wasi:http/types" (instance $http-types
        (export "request" (type (sub resource)))
        (export "response" (type (sub resource)))
        ...
      ))
      (alias export $http-types "request" (type $request))
      (alias export $http-types "response" (type $response))
      (import "wasi:http/handler" (instance
        (export "handle" (func (param "r" (own $request)) (result (own $response))))
      ))
      (export "wasi:http/handler" (instance
        (export "handle" (func (param "r" (own $request)) (result (own $response))))
      ))
    ))
  ))
)

This examples shows how, in the context of concrete world (wasi:http/proxy), standalone interface definitions (such wasi:http/handler) are no longer in a "parameterized" form: there is no outer wrapping component-type and instead all uses are replaced by direct aliases to preceding type imports as determined by the WIT resolution process.

Unlike most other WIT constructs, the @since and @unstable gates are not represented in the component binary. Instead, they are considered "macro" constructs that take the place of maintaining two copies of a single WIT document. In particular, when encoding a collection of WIT documents into a binary, the target version and set of explicitly-enabled feature names determine whether individual gated features are included in the encoded type or not.

For example, the following WIT document:

package ns:p@1.1.0;

interface i {
  f: func();

  @since(version = 1.1.0)
  g: func();
}

is encoded as the following component when the target version is 1.0.0:

(component
  (type (export "i") (component
    (export "ns:p/i@1.0.0" (instance
      (export "f" (func))
    ))
  ))
)

If the target version was instead 1.1.0, the same WIT document would be encoded as:

(component
  (type (export "i") (component
    (export "ns:p/i@1.1.0" (instance
      (export "f" (func))
      (export "g" (func))
    ))
  ))
)

Thus, @since and @unstable gates are not part of the runtime semantics of components, just part of the source-level tooling for producing components.

Files

WIT.md

Latest commit

History

WIT.md

File metadata and controls

The wit format

Package Names

WIT Interfaces

WIT Worlds

Union of Worlds with include

De-duplication of interfaces

Name Conflicts and with

A Note on Subtyping

WIT Packages and use

Interfaces, worlds, and use

Top-level use

Transitive imports and worlds

WIT Functions

WIT Types

WIT Identifiers

Lexical structure

Whitespace

Comments

Operators

Keywords

Integers

Top-level items

Feature Gates

Feature gate syntax

Rules for feature gate usage

Scenario: Stabilization of a new feature

Scenario: Deprecation of an existing stable feature

Package declaration

Item: toplevel-use

Item: world

Item: include

Item: interface

Item: use

Items: type

Item: type (alias)

Item: record (bag of named fields)

Item: flags (bag-of-bools)

Item: variant (one of a set of types)

Item: enum (variant but with no payload)

Item: resource

Types

Handles

Name resolution

Package Format

The `wit` format

Union of Worlds with `include`

Name Conflicts and `with`

WIT Packages and `use`

Interfaces, worlds, and `use`

Top-level `use`

Item: `toplevel-use`

Item: `world`

Item: `include`

Item: `interface`

Item: `use`

Item: `type` (alias)

Item: `record` (bag of named fields)

Item: `flags` (bag-of-bools)

Item: `variant` (one of a set of types)

Item: `enum` (variant but with no payload)

Item: `resource`