Language basics

This guide introduces the Ballerina language to help you get started with basics that are common to all C-Family programming languages.

Familiar subset of Ballerina

Ballerina is a modern programming language focused on building applications for the cloud era. It has a familiar syntax resembling the C-family of languages such as C/C++, C#, Java, and JavaScript. It brings in many powerful concepts from existing programming languages and some unconventional abstraction mechanisms to make application development for the cloud more intuitive and less strenuous.

This guide presents the subset of the Ballerina language that is common to most modern programming languages. It covers the baseline features such as data types, control flow, and error handling, which should be very familiar to most programmers. These features are the foundation for some of the more distinctive features of Ballerina.

Programs and modules

Ballerina programs consist of modules and each module consists of one or more .bal files.

The most important thing that a module does is define named functions.

Here's what a simple Ballerina module looks like:

import ballerina/io; 

public function main() {
    io:println("Hello World!"); 

The import keyword is used to bind predefined library modules to the module prefix. Ballerina modules have two-part names represented by an organization name and a module name. In this case, the organization is ballerina and the module is io. The bound identifier that refers to this library module is set to the module name io. To override the default identifier binding for the module, you can use the as keyword as follows: import ballerina/io as oi.

Module hierarchies follow the / and . notations. For example, org/x.y.z represents a hierarchical module name. The z:f notation points to a symbol f (e.g., a function) defined within the module x.y.z.

The main function is the program entry point and the public keyword makes this function visible outside the module. This main function calls the println() function defined in the ballerina/io module. Every statement in Ballerina code must terminate with a semicolon.

Variables and types

Modules and functions can declare variables.

Variables have types. Ballerina defines several built-in types such as integers, floating point numbers, strings, and booleans.

Here is a typical way of declaring variables:

import ballerina/io;

string greeting = "Hello";

public function main() {
    string name = "Ballerina";
    io:println(greeting, " ", name);

greeting is a module-level variable of type string initialized to a value of "Hello". The name variable is a local variable of type string initialized to a value "Ballerina".

In Ballerina, variable assignments are statements and not expressions.


Like all high-level programming languages, Ballerina supports functions.

A function definition in Ballerina has a familiar syntax that includes parameters and a return type.

function add(int x, int y) returns int {
    int sum = x + y;
    return sum;

In the above code example, you see the definition for the add function. It accepts two parameters of type int, x, and y. It also returns a value of type int. The returns keyword specifies the type of return value.

The function performs a simple addition operation and stores the result in another variable sum of type int. The return statement returns the value assigned to sum.


The general syntax of a Ballerina program is as follows:

import ballerina/io;

// This is a comment.
int count = 0;

// You can have Unicode identifiers.
function พิมพ์ชื่อ(string ชื) {
    // Use \u{H} to specify character using Unicode code point in hex.

string 'string = "xyz";

The overall syntax is more or less C-like. Comments start with // and end at the end of the line. Module definitions/declarations and statements use braces or terminate with a semicolon.

You can also use Unicode characters in identifiers. Unicode literals can be defined using the \u{H} syntax to represent Unicode code points in the hex format.

Keywords such as int, function, and string are reserved keywords in Ballerina. You can use them as identifiers by prefixing them with a single quote.


Ballerina defines an integer data type of a 64-bit signed representation. The int keyword is used to declare an integer. Integer literals can be declared either in decimal or hexadecimal format.

int m = 1;

int n = 0xFFFF;

The integer data type supports all the basic arithmetic operators: +, -, *, /, and %, the comparison operators: ==, !=, <, >, <=, and >= as well as the compound assignment operators: += and -=. It also supports the usual bitwise operators (i.e., &, |, ^, ~, <<, and >>). The operator precedence rules are exactly as per C.

However, there are a couple of notable deviations in Ballerina. Firstly, increment and decrement operators (i.e., ++ and --) are not supported. Moreover, integer overflow results in a runtime error unlike in C where it rolls over.

Floating point numbers

In Ballerina, floating point numbers have an IEEE 64-bit binary representation similar to double in Java. Variables of the float type are declared with the float keyword and they support all operations similar to integers except bitwise operators.

int n = 2;
float x = 1.0;

float y = x + <float>n;

However, they cannot be implicitly converted to or from integers. You can use the <T> notation for explicit conversion.

In Ballerina, NaN, the IEEE defined terminology for Not a Number is equal to itself. Therefore, the == operator on two NaNs will return true. For floating point types, the == and != operators test for the same value and do not test for numerical equivalence as per the IEEE convention.

Booleans and conditionals

The boolean type has only two values: true and false.

boolean flag = true;

Ballerina supports the logical operators, !, ||, and && with boolean values. The || and && operators support the same short-circuit behavior as in C. Implicit conversion between boolean and integer is not supported.

Booleans are also used in conditional expressions:

int n = flag ? 1 : 2;

As usual, booleans can be used in if-else statements:

function foo() {
    if flag {
    } else {

If-else statements and all compound statements in Ballerina require curly braces. However, the parenthesis before the curly braces are optional.


Nil is another data type. It has a special significance in Ballerina as it denotes the equivalent of what is known as a null value in other C-family languages. If you want to declare a variable or a value of type nil, here is how it is done.

int? v = ();

The keyword int is followed by a ? notation. It means that v can take an integer value or can be nil. It is initialized to nil, which is denoted by (). Types do not allow their values to be set as nil implicitly. There is also no implicit conversion to boolean.

The == and != comparison operators operate on a value to test whether the value is nil.

int n = v == () ? 0 : v;

Similarly, you can also use the Elvis operator like x ?: y that returns x if it is not nil or y otherwise.

int n = v ?: 0;

Leaving off the return type of a function is the same as defining nil as the return type. Moreover, falling off the end of a function or return by itself is equivalent to return (). Therefore, you can write a function as follows:

function foo() returns () {
    return ();


Strings are immutable sequences of zero or more Unicode characters. To define a variable of type string, you use the string keyword in Ballerina.

string grin = "\u{1F600}";

String literals are enclosed in double quotes and support the usual C-style escape sequences as well as numeric escape characters for Unicode code points.

The string type supports the usual operators. The == operator checks for the same characters. The comparison operators (i.e., <, <=, =>, and >) work by comparing code points. The + operator is used for concatenation.

string greeting = "Hello" + grin;

The member access expression (e.g., s[i]) on a string value points to the character at index i of the string. The index positions on values of the string type start with zero.

Ballerina does not support the notion of a separate character type like in C. Instead, a character is considered a string of length one.

Langlib functions

Ballerina defines lang libraries to provide fundamental operations on built-in datatypes.

For example, you can perform standard operations such as getting a substring or finding the length of the string on values of the string type.

string s = "abc".substring(1, 2);

int n = s.length();

The substring() and length() functions are string lang library functions called using the convenient method call syntax. However, the functions are called on variables/values of the string type rather than objects.

A lang library ballerina/lang.T is defined for each built-in type T. Ballerina automatically imports these lang libraries and pre-declares a prefix T, which you can use to access the constructs of the module. Therefore, in the case of the above code example, you can also find the length of the string value referring to the length() function as string:length() and using the function call syntax.

string s = "abc".substring(1, 2);

int n = string:length(s);

The lang library can also be imported explicitly as import ballerina/lang.'string.

import ballerina/lang.'string;

string s = "abc".substring(1, 2);

int n = string:length(s);


Arrays are ordered data structures consisting of values of the same type.

You can declare an array of integers as:

int[] v = [1, 2, 3];

You can index the individual elements of this array using the v[i] notation. Array indexing starts with zero. So the second element of v can be accessed as v[1].

int n = v[1];

Arrays are mutable. Ballerina supports ordering based on a lexicographical ordering of members.

The == and != comparison operators perform a deep comparison of two arrays based on the members and their order instead of the memory location.

Langlib functions for arrays are available to perform operations such as finding the length of an array.

int len = v.length();

foreach statement

A foreach statement iterates over an array by binding a variable to each array member at every loop iteration.

function sum(float[] v) returns float {
    float r = 0.0;
    foreach float x in v {
        r += x;
    return r;        

In this code example, the x variable of the float type is bound to each element of the v array. Similarly, it also works with the string type by iterating over each character in the string.

You can also construct the foreach statement using a range expression such as ..<.

function sum(float[] v) returns float {
    float r = 0.0;
    foreach int i in 0 ..< v.length() {
        r += v[i];
    return r;

The above statement sets the i variable of the int type to a sequence of linearly incrementing numbers that range from zero to the length of the array v. Therefore, the foreach statement iterates over the length of the array and increments i during each iteration.

while Statement

The while loop in Ballerina behaves exactly like in C.

type LinkedList record {
    string value;
    LinkedList? next;

function len(LinkedList ll) returns int {
    int n = 0;
    LinkedList? nextLL =;

    while nextLL != () {
        n += 1;
        nextLL =;

    return n;

In this example, the while loop is checking for a boolean condition to check for the end of the linked list ll.

The while statement also supports the usual break and continue statements.

Binary data

Binary data is represented by an array of bytes.

byte[] data = base64 `yPHaytRgJPg+QjjylUHakEwz1fWPx/wXCW41JSmqYW8=`;

An array of bytes can also be represented in the source code using base-64 or base-16 literal formats.

The byte type represents a number between 0 and 0xFF. You can define a variable of the byte type with the byte keyword.

byte b = 0xFF;

In Ballerina, byte is a subtype of the integer type. Therefore, it supports all the bitwise operators available on integers. Some of these operators produce a value that belongs to the integer type, and some operators like the & operator produce a value that belongs to the byte type.


A map is an associative structure of multiple string values as keys and their values. If you want to create a map of integers, then you would define it as:

map<int> m = {
    "x": 1,
    "y": 2

The syntax to define a map is very similar to JSON. Maps are mutable, and m["x"] will return the integer value stored against x as the key, or nil if the key is not present.

m["x"] = 5;

int? v = m["x"];

When used in a foreach loop, it will iterate over all the values of the map. You can also use lang library functions such as get(k) to get the value mapped to the string key k, or keys( ) to return an array containing all the key strings of the map.

You can perform a deep comparison by using the == and != comparison operators on a map. Two maps are equal if they have the same set of keys and the values for each key are equal.

Type definitions

Ballerina allows you to define a name for a type. You can create user-defined type identifiers for the built-in types supported by the language.

For example, if you have an array of maps of the string type, map<string>[], you can define a type definition for it as follows:

type MapArray map<string>[];

MapArray arr = [
    {"x": "foo"},
    {"y": "bar"}

This is similar to the typedef concept in C, where the identifier MapArray acts as an alias for the actual type.


A record is a collection of fields of a specific type. With record types, you have control over what your keys are.

record {int x; int y; } r = {
    x: 1,
    y: 2

The above code defines a record r with two integer fields x and y, and initializes their values to 1 and 2, respectively.

You can also define it using the type definition as follows:

type Coord record {
    int x;
    int y;

Subsequently, you can use the Coord record to declare a variable and set or access its field values.

Coord c = {x: 1, y: 2};

int a = c.x;

Records are mutable. c.x is an lvalue. Rules for record comparison are the same as those for maps.

Structural typing

In Ballerina, the data types are defined in terms of a set of values. A type is a label for a set of values. These values can be part of a finite set or an infinite set.

For example, the int type has a finite set consisting of values within the range of 64-bit signed numbers. The string type is an infinite set that contains sequences of any number or order of characters. In this way, Ballerina partitions the universe of all values into the basic types it supports such as nil, boolean, integer, floating point, and string. Values of these basic types are immutable, and each value belongs to exactly one basic type.

There is also a concept of semantic subtyping, which means a subset of the values of the basic types. For example, the byte type is a subset of integer as it allows only a subset of values that are a part of the integer value set.

Additionally, Ballerina allows you to define your own types, such as arrays, maps, and records, which are mutable.


Ballerina's type system allows you to define custom types based on the combination of two or more types. A union is, therefore, a superset of that combination.

A variable of type T? can hold a value of type T or nil as explained in the section on the Nil type.

Similarly, you can define a variable that holds an integer or a string.

type flexType string|int;

flexType a = 1;

flexType b = "Hello";

You can create a union of multiple types using the | symbol in the type definition. Therefore, in the above example, flexType can hold an integer value as well as a string value, and the Ballerina compiler won't complain.

You can also apply the concept of unions to structured data types like records. So it is possible to have a union of a record type containing fields of the string type and the basic string type.

type StructuredName record {
    string firstName;
    string lastName;

type Name StructuredName|string;

At runtime, you can define the business logic by checking the currently held value of the union-typed variable using the is operator.

function nameToString(Name nm) returns string {
    if nm is string {
        return nm;
    } else {
        return nm.firstName + " " + nm.lastName;

In this way, the is operator in a condition causes the declared type to be narrowed.

Error reporting

Ballerina does not support the notion of exceptions. Instead, errors are handled as part of the normal control flow. Errors are first-class citizens in Ballerina. Ballerina supports a few mechanisms that are centered around an error type to handle errors. The error type is a basic type. Ballerina programs can pass around error values similar to values of any other type.

Functions report errors by returning error values.

function parse(string s) returns int|error {
    int n = 0;
    int[] cps = s.toCodePointInts();
    foreach int cp in cps {
        int p = cp - 0x30;
        if p < 0 || p > 9 {
            return error("not a digit");
        n = n * 10 + p;
    return n;

In the above code example, the function parse() returns either an integer or an error. At the point of returning the error, the error constructor error("not a digit") is called.

In case a function is expected to explicitly return only error values, a return type of error? is used. Ignoring an error also has to be done explicitly.

An error value includes a string message and the stack trace from the point where error(...) is called. An error value is immutable.

Error handling

When an error is reported from a function, it is passed up to its caller. Ultimately it is handled at the main function, which can return it. Alternatively, you can use the is operator to check for the error type as part of the regular control flow to make a decision.

function intFromBytes(byte[] bytes) returns int|error {
    string|error ret = string:fromBytes(bytes);

    if ret is error {
        return ret;
    } else {
        return int:fromString(ret);

In the above code example, the function defines a local variable ret to hold the value returned by the string:fromBytes() function. The ret variable is of a union type of the string and error types. Therefore to ascertain the actual type, is operator is used.

check expression

Instead of using the is operator within the regular conditional flow to check for errors, Ballerina supports the check keyword as a shorthand method. Error handling and return statements become much more concise when using the check keyword.

For example, you can write the intFromBytes() function defined in the Error handling section as follows:

function intFromBytes(byte[] bytes) returns int|error {

    string str = check string:fromBytes(bytes);
    return int:fromString(str);

In this case, the expression following the check keyword (string:fromBytes(bytes)) is evaluated and the intFromBytes function returns immediately if the string:fromBytes() function returns an error.

Error subtyping

Ballerina also allows fine-grained error detection based on error subtypes. Error subtypes can be defined using the distinct keyword.

type XErr distinct error;
type YErr distinct error;

type Err XErr|YErr;

Now, you can use the defined subtype in an error constructor to create error values of that type.

Err err = error XErr("Whoops!");

You can use the is operator to distinguish between the distinct subtypes.

function desc(Err err) returns string {
    return err is XErr ? "X" : "Y";


In Ballerina, there is a distinction between normal and abnormal errors. Normal errors are the ones that you typically handle as part of the usual business logic and are under your control. But some errors are not under the control of the programmer. These are abnormal errors. Things such as bugs in library code or out of memory errors fall under this category.

Normal errors are handled by returning values of the error type. Abnormal errors are handled using the panic statement.

function divide(int m, int n) returns int {
    if n == 0 {
        panic error("division by 0");
    return m / n;

A panic statement results in immediate program termination and has an associated error value.

any type

Ballerina also has an any type, which means any value except an error value. A variable of type any can be cast to a specific type using the type cast expression (<T>x).

any x = 1;

int n = <int>x;

It can be converted to string.

string s = x.toString();

You can also use the is operator to test the type of the actual value held by the variable.

float f = x is int|float ? <float>x : 0.0;

Ignore return values and errors

Ballerina does not allow silently ignoring return values of functions unless they return nil.

// Allowed only if return value is ()

To ignore the return value, you can assign it to _, which acts as an implicitly declared variable of any type that cannot be referenced.

_ = getX();

When a function's return type includes an error type, it has to be handled explicitly. Otherwise, the checkpanic keyword can be used but it panics on error rather than returning.

checkpanic tryX();


Ballerina's type system follows a structured type system based on the set of values that belongs to a type. For example, the integer type allows a set of values. Similarly, the any type also allows a set of values, including the subset of values that belong to the integer type.

Therefore, you can have an array of the integer type that is assigned to an array of the any type.

int[] iv = [1, 2, 3];

any[] av = iv;

This is perfectly valid in Ballerina because the set of values allowed by an array of the integer type is a subset of the values allowed by an array of the any type. So this way, you can have some flexibility in allowing fixed inherent type substitution based on the set of values.

Ballerina's static type checking ensures that the result of a read operation on the array av will be consistent with static type. However, writing may result in an error because the inherent type of the iv array is array of integers. As a result, the code below will lead to a runtime error since it is trying to mutate the array to a type other than the inherent type (int[]).

av[0] = "str"; // panics

This is covariance, which means that a write to a mutable structure may result in a runtime error. Apart from arrays, maps and records are the other data structures with an inherent type that constrains mutation.


Ballerina provides another basic type which is the object type. Object types bundle together code and data.

Objects are initialized based on a class defined within a module.

function demoMyClass() {
    m:MyClass x = new m:MyClass(1234);;
    int n = x.n;

In the above code example, the function demoMyClass() uses the new expression to create an object x using the MyClass class that is defined in module m. You can use the . notation to call the object's methods or access its fields.

Define classes

Ballerina supports the concept of a class which is a structured type containing data and behavior.

class MyClass {
    private int n;

    public function init(int n = 0) {
        self.n = n;

    public function inc() {
        self.n += 1;

    public function get() returns int {
        return self.n;

The init method is the constructor for this class, and the keyword self is used to access the object. It also uses the standard access specifiers, private and public, which translate to accessibility within the class definition and outside of it.

init return type

The init() method of a class can have a return type. The return type must be a subtype of the union of error and nil.

class File {

    string path;

    string contents;

    function init(string p) returns error? {
        self.path = p;
        self.contents = check io:fileReadString(p);


File f = check new File("text.txt");

If the init() call returns normally, the newly constructed object is returned. Otherwise, the new expression returns the error value.

The return type of the init() method is nil if a return type is not explicitly specified. It also means that the init method will never return an error.


In Ballerina, the identity of an object is determined by the memory location in which the object resides at runtime. To check the identity, you can use the === and !== equality expressions.

MyClass obj1 = new MyClass();
MyClass obj2 = new MyClass();

// true
boolean b1 = (obj1 === obj1);

// false
boolean b2 = (obj1 === obj2);

In the above code example, both obj1 and obj2 are MyClass objects. obj1 === obj1 returns true as both operands of the operation refer to the same memory location. Similarly, obj1 === obj2 returns false since they refer to different memory locations.

The == operator is used to check for contents of a structure type like arrays.

// true
boolean b3 = ([1, 2, 3] == [1, 2, 3]);

// false
boolean b4 = ([1, 2, 3] === [1, 2, 3]);

In the above code example, the first operation results in true as both the arrays have exactly the same content. However, the second operation is false because it is using the === operator to check for memory locations, and the two inline arrays point to different memory locations, even though their contents are the same.

In the case of floating point numbers, IEEE defines -0.0 and +0.0 as the same. Therefore applying == to compare -0.0 and +0.0 would return true, whereas -0.0 === +0.0 would always be false as they are not identical.

const and final

Ballerina supports constants. Constants are defined using the const keyword and are immutable singleton types having a value known at compile time.

const MAX_VALUE = 1000;

Apart from const, there is also support for final. Variables defined with the final keyword cannot be reassigned to after being initialized.

final string msg = loadMessage();

This is also applicable for class fields.


Ballerina supports the notion of enumerations which is a shorthand way for expressing unions of string constants.

enum Color {

Using the enum keyword, you can define this enumeration of three string values "RED", "GREEN", and "BLUE", which would otherwise be defined separately as a union of string constants:

const RED = "RED";
const GREEN = "GREEN";
const BLUE = "BLUE";

type Color RED|GREEN|BLUE;

You can also have an enumeration where members define associated string values explicitly.

enum Language {
    ENG = "English",
    TL = "Tamil",
    SI = "Sinhala"

match statement

The match statement is like the switch case statement in C, but it is more flexible. You can use it to match values of different types.

const KEY = "xyzzy";

function mtest(any v) returns string {

    match v {
        17 => {
            return "number";
        true => {
            return " boolean";
        "str" => {
            return "string";
        KEY => {
            return "constant";
        0|1 => {
            return "or";
        _ => {
            return "any";

The match statement defines multiple clause statement blocks with the => operator. The left-hand side of the => operator in each clause is the pattern for the value match. It can contain literals or identifiers referring to constants. It can also contain multiple patterns specified using |. Additionally, _ can be used to match any value that is not an error.

In the above example, the match statement uses the value held by v and matches it against six match clauses having distinct patterns. During execution, the patterns will be evaluated in order, using the equality expression (==) to test the pattern match, and the matched clause's statement block will be executed.

Type inference

Type inference allows the program to infer the actual data type of variables.

This feature is a convenient way to use a generic type for a local variable and helps avoid repeated type declarations.

var x = "str";

In this code example, the keyword var is used to declare the variable x. It means that the variable's actual type will be inferred from the type of the expression used to initialize it. In this case, it is the string type.

This is especially useful in a foreach loop to declare the local variable for iterating over a collection, like arrays.

function printLines(string[] sv) {
    foreach var s in sv {


In this code example, the type of variable s is inferred to be of type string, from the type of sv, which is an array of type string.

Type inference also works with classes.

var x = new MyClass();

Here the variable x is declared with var and the type is inferred as MyClass.

The opposite syntax is also valid, wherein the variable x is defined with the MyClass type and is initialized with new.

MyClass x = new;

Type inference is applicable for variables in local scopes. Therefore var should be used sparingly for variables used within a very limited scope, like in a foreach loop. Overusing it makes the code harder to understand.

Functional programming

Ballerina defines functions also as values, and they work as closures. Therefore function is also a type, which can be defined as a basic type as follows:

var isOdd = function(int n) returns boolean {
    return n % 2 != 0;

type IntFilter function (int n) returns boolean;

function isEven(int n) returns boolean {
    return n % 2 == 0;

IntFilter f = isEven;

In the above example, the variable isOdd defines an anonymous function. IntFilter is a function type that accepts a function with an int argument and returns a boolean value. And later on, the isEven function, with a matching signature, is assigned to a variable of this type.

These function values can be passed around as arguments. For example, the lang library for arrays defines a function filter(f) which accepts a function type f to perform specific filtering operations on arrays.

int[] nums = [1, 2, 3];

int[] evenNums = nums.filter(f);

The filter() function takes the function value as a parameter, which is isEven, assigned to f. Therefore, passing it as the argument results in filter() returning an array of even numbers.

Instead of passing a function variable, you can also pass anonymous functions as an argument.

int[] oddNums = nums.filter(n => n % 2 != 0);

As shown for the array oddNums, the type of parameter n is inferred from the array with which the filter() function is used.

Asynchronous function calls

Ballerina also supports asynchronous function calls. You can call a function asynchronously as follows:

future<int> f1 = start foo();

future<int> f2 = start foo();

The use of the start keyword before the function call makes it asynchronous. Each asynchronous call runs on a separate logical thread, also known as strand, that is cooperatively multitasked by default.

The result of this invocation is returned as a future. It is a separate basic type that has an attached type T as future<T>.

You can then wait for the result with the wait keyword.

int x1 = check wait f1;
int x2 = check wait f2;

Waiting on a future<T> returns T|error. Waiting on the same future more than once returns an error value.

Instead of waiting for each future separately, you can also do the following.

record {|int|error f1; int|error f2;|} f = wait {f1, f2};

int x1 = check f.f1;
int x2 = check f.f2;

In case you want to terminate the future, you can do f.cancel( ).


Ballerina supports a structured way to document code. Documentation lines start with # and contain structured documentation in Markdown format.

# Adds two integers.
# + x - an integer
# + y - another integer
# + return - the sum of `x` and `y`
public function add(int x, int y)
                     returns int {
    return x + y;

The above code example adds documentation to describe the parameters and the return type of the add() function. This is Ballerina-flavoured Markdown (BFM) which makes it convenient to generate pretty documentation using one of the platform tools.


Annotations are defined using the @ notation followed by a tag. This is a way of attaching metadata to the code.

@display {
    iconPath: "transform.png"
public function transform(string s) returns string {
    return s.toUpperAscii();

future<int> fut = @strand {thread: "any"} start foo();

In the above code example, the @display annotation is attached to the function transform(), and the @strand annotation applies to the keyword start, calling the function foo() asynchronously. These annotations use mapping constructor expressions, which is one of the ways to define them.

The annotations shown above use unprefixed tags and refer to standard platform-defined annotations. You can also have prefixed tags for user-defined annotations declared in modules.