Rhai - Embedded Scripting for Rust

Rhai is an embedded scripting language and evaluation engine for Rust that gives a safe and easy way to add scripting to any application.

Rhai's current features set:

• Easy-to-use language similar to JS+Rust with dynamic typing but no garbage collector.
• Tight integration with native Rust functions and types, including getters/setters, methods and indexers.
• Freely pass Rust variables/constants into a script via an external Scope.
• Easily call a script-defined function from Rust.
• Low compile-time overhead (~0.6 sec debug/~3 sec release for rhai_runner sample app).
• Fairly efficient evaluation (1 million iterations in 0.75 sec on my 5 year old laptop).
• Relatively little unsafe code (yes there are some for performance reasons, and most unsafe code is limited to one single source file, all with names starting with "unsafe_").
• Re-entrant scripting Engine can be made Send + Sync (via the sync feature).
• Sand-boxed - the scripting Engine, if declared immutable, cannot mutate the containing environment without explicit permission.
• Rugged (protection against stack-overflow and runaway scripts etc.).
• Track script evaluation progress and manually terminate a script run.
• no-std support.
• Function overloading.
• Operator overloading.
• Organize code base with dynamically-loadable Modules.
• Compiled script is optimized for repeated evaluations.
• Support for minimal builds by excluding unneeded language features.
• Very few additional dependencies (right now only num-traits to do checked arithmetic operations); for no-std builds, a number of additional dependencies are pulled in to provide for functionalities that used to be in std.

Note: Currently, the version is 0.14.1, so the language and API's may change before they stabilize.

Installation

Install the Rhai crate by adding this line to dependencies:

[dependencies]
rhai = "0.14.1"

Use the latest released crate version on crates.io:

[dependencies]
rhai = "*"

Crate versions are released on crates.io infrequently, so if you want to track the latest features, enhancements and bug fixes, pull directly from GitHub:

[dependencies]
rhai = { git = "https://github.com/jonathandturner/rhai" }

Beware that in order to use pre-releases (e.g. alpha and beta), the exact version must be specified in the Cargo.toml.

Optional features

Feature	Description
unchecked	Exclude arithmetic checking (such as over-flows and division by zero), stack depth limit and operations count limit. Beware that a bad script may panic the entire system!
no_function	Disable script-defined functions.
no_index	Disable arrays and indexing features.
no_object	Disable support for custom types and object maps.
no_float	Disable floating-point numbers and math.
no_optimize	Disable the script optimizer.
no_module	Disable modules.
only_i32	Set the system integer type to i32 and disable all other integer types. INT is set to i32.
only_i64	Set the system integer type to i64 and disable all other integer types. INT is set to i64.
no_std	Build for no-std. Notice that additional dependencies will be pulled in to replace std features.
sync	Restrict all values types to those that are Send + Sync. Under this feature, all Rhai types, including Engine, Scope and AST, are all Send + Sync.

By default, Rhai includes all the standard functionalities in a small, tight package. Most features are here to opt-out of certain functionalities that are not needed. Excluding unneeded functionalities can result in smaller, faster builds as well as more control over what a script can (or cannot) do.

Performance builds

Some features are for performance. For example, using only_i32 or only_i64 disables all other integer types (such as u16). If only a single integer type is needed in scripts - most of the time this is the case - it is best to avoid registering lots of functions related to other integer types that will never be used. As a result, performance will improve.

If only 32-bit integers are needed - again, most of the time this is the case - using only_i32 disables also i64. On 64-bit targets this may not gain much, but on some 32-bit targets this improves performance due to 64-bit arithmetic requiring more CPU cycles to complete.

Also, turning on no_float, and only_i32 makes the key Dynamic data type only 8 bytes small on 32-bit targets while normally it can be up to 16 bytes (e.g. on x86/x64 CPU's) in order to hold an i64 or f64. Making Dynamic small helps performance due to better cache efficiency.

Minimal builds

In order to compile a _minimal_build - i.e. a build optimized for size - perhaps for embedded targets, it is essential that the correct linker flags are used in cargo.toml:

[profile.release]
lto = "fat"         # turn on Link-Time Optimizations
codegen-units = 1   # trade compile time with maximum optimization
opt-level = "z"     # optimize for size

Opt out of as many features as possible, if they are not needed, to reduce code size because, remember, by default all code is compiled in as what a script requires cannot be predicted. If a language feature is not needed, omitting them via special features is a prudent strategy to optimize the build for size.

Omitting arrays (no_index) yields the most code-size savings, followed by floating-point support (no_float), checked arithmetic (unchecked) and finally object maps and custom types (no_object). Disable script-defined functions (no_function) only when the feature is not needed because code size savings is minimal.

Engine::new_raw creates a raw engine which does not register any utility functions. This makes the scripting language quite useless as even basic arithmetic operators are not supported. Selectively include the necessary functionalities by loading specific packages to minimize the footprint. Packages are sharable (even across threads via the sync feature), so they only have to be created once.

Related

Other cool projects to check out:

• ChaiScript - A strong inspiration for Rhai. An embedded scripting language for C++ that I helped created many moons ago, now being led by my cousin.
• Check out the list of scripting languages for Rust on awesome-rust

Examples

A number of examples can be found in the examples folder:

Example	Description
arrays_and_structs	demonstrates registering a new type to Rhai and the usage of arrays on it
custom_types_and_methods	shows how to register a type and methods for it
hello	simple example that evaluates an expression and prints the result
no_std	example to test out no-std builds
reuse_scope	evaluates two pieces of code in separate runs, but using a common Scope
rhai_runner	runs each filename passed to it as a Rhai script
simple_fn	shows how to register a Rust function to a Rhai Engine
repl	a simple REPL, interactively evaluate statements from stdin

Examples can be run with the following command:

cargo run --example name

The repl example is a particularly good one as it allows you to interactively try out Rhai's language features in a standard REPL (Read-Eval-Print Loop).

Example Scripts

There are also a number of examples scripts that showcase Rhai's features, all in the scripts folder:

Language feature scripts	Description
array.rhai	arrays in Rhai
assignment.rhai	variable declarations
comments.rhai	just comments
for1.rhai	for loops
function_decl1.rhai	a function without parameters
function_decl2.rhai	a function with two parameters
function_decl3.rhai	a function with many parameters
if1.rhai	if example
loop.rhai	endless loop in Rhai, this example emulates a do..while cycle
op1.rhai	just a simple addition
op2.rhai	simple addition and multiplication
op3.rhai	change evaluation order with parenthesis
string.rhai	string operations
while.rhai	while loop

Example scripts	Description
speed_test.rhai	a simple program to measure the speed of Rhai's interpreter (1 million iterations)
primes.rhai	use Sieve of Eratosthenes to find all primes smaller than a limit
fibonacci.rhai	calculate the n-th Fibonacci number using a really dumb algorithm
mat_mul.rhai	matrix multiplication test to measure the speed of Rhai's interpreter

To run the scripts, either make a tiny program or use of the rhai_runner example:

cargo run --example rhai_runner scripts/any_script.rhai

Hello world

To get going with Rhai, create an instance of the scripting engine via Engine::new and then call the eval method:

use rhai::{Engine, EvalAltResult};

fn main() -> Result<(), Box<EvalAltResult>>
{
    let engine = Engine::new();

    let result = engine.eval::<i64>("40 + 2")?;

    println!("Answer: {}", result);             // prints 42

    Ok(())
}

EvalAltResult is a Rust enum containing all errors encountered during the parsing or evaluation process.

Script evaluation

The type parameter is used to specify the type of the return value, which must match the actual type or an error is returned. Rhai is very strict here. Use Dynamic for uncertain return types. There are two ways to specify the return type - turbofish notation, or type inference.

let result = engine.eval::<i64>("40 + 2")?;     // return type is i64, specified using 'turbofish' notation

let result: i64 = engine.eval("40 + 2")?;       // return type is inferred to be i64

result.is::<i64>() == true;

let result: Dynamic = engine.eval("boo()")?;    // use 'Dynamic' if you're not sure what type it'll be!

let result = engine.eval::<String>("40 + 2")?;  // returns an error because the actual return type is i64, not String

Evaluate a script file directly:

let result = engine.eval_file::<i64>("hello_world.rhai".into())?;       // 'eval_file' takes a 'PathBuf'

Compiling scripts (to AST)

To repeatedly evaluate a script, compile it first into an AST (abstract syntax tree) form:

// Compile to an AST and store it for later evaluations
let ast = engine.compile("40 + 2")?;

for _ in 0..42 {
    let result: i64 = engine.eval_ast(&ast)?;

    println!("Answer #{}: {}", i, result);      // prints 42
}

Compiling a script file is also supported:

let ast = engine.compile_file("hello_world.rhai".into())?;

Calling Rhai functions from Rust

Rhai also allows working backwards from the other direction - i.e. calling a Rhai-scripted function from Rust via Engine::call_fn. Functions declared with private are hidden and cannot be called from Rust (see also modules).

// Define functions in a script.
let ast = engine.compile(true,
    r"
        // a function with two parameters: String and i64
        fn hello(x, y) {
            x.len() + y
        }

        // functions can be overloaded: this one takes only one parameter
        fn hello(x) {
            x * 2
        }

        // this one takes no parameters
        fn hello() {
            42
        }

        // this one is private and cannot be called by 'call_fn'
        private hidden() {
            throw "you shouldn't see me!";
        }
    ")?;

// A custom scope can also contain any variables/constants available to the functions
let mut scope = Scope::new();

// Evaluate a function defined in the script, passing arguments into the script as a tuple.
// Beware, arguments must be of the correct types because Rhai does not have built-in type conversions.
// If arguments of the wrong types are passed, the Engine will not find the function.

let result: i64 = engine.call_fn(&mut scope, &ast, "hello", ( String::from("abc"), 123_i64 ) )?;
//                                                          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
//                                                          put arguments in a tuple

let result: i64 = engine.call_fn(&mut scope, &ast, "hello", (123_i64,) )?;
//                                                          ^^^^^^^^^^ tuple of one

let result: i64 = engine.call_fn(&mut scope, &ast, "hello", () )?;
//                                                          ^^ unit = tuple of zero

// The following call will return a function-not-found error because
// 'hidden' is declared with 'private'.
let result: () = engine.call_fn(&mut scope, &ast, "hidden", ())?;

Creating Rust anonymous functions from Rhai script

It is possible to further encapsulate a script in Rust such that it becomes a normal Rust function. Such an anonymous function is basically a boxed closure, very useful as call-back functions. Creating them is accomplished via the Func trait which contains create_from_script (as well as its companion method create_from_ast):

use rhai::{Engine, Func};                       // use 'Func' for 'create_from_script'

let engine = Engine::new();                     // create a new 'Engine' just for this

let script = "fn calc(x, y) { x + y.len() < 42 }";

// Func takes two type parameters:
//   1) a tuple made up of the types of the script function's parameters
//   2) the return type of the script function
//
// 'func' will have type Box<dyn Fn(i64, String) -> Result<bool, Box<EvalAltResult>>> and is callable!
let func = Func::<(i64, String), bool>::create_from_script(
//                ^^^^^^^^^^^^^ function parameter types in tuple

                engine,                         // the 'Engine' is consumed into the closure
                script,                         // the script, notice number of parameters must match
                "calc"                          // the entry-point function name
)?;

func(123, "hello".to_string())? == false;       // call the anonymous function

schedule_callback(func);                        // pass it as a callback to another function

// Although there is nothing you can't do by manually writing out the closure yourself...
let engine = Engine::new();
let ast = engine.compile(script)?;
schedule_callback(Box::new(move |x: i64, y: String| -> Result<bool, Box<EvalAltResult>> {
    engine.call_fn(&mut Scope::new(), &ast, "calc", (x, y))
}));

Raw Engine

Engine::new creates a scripting Engine with common functionalities (e.g. printing to the console via print). In many controlled embedded environments, however, these are not needed.

Use Engine::new_raw to create a raw Engine, in which nothing is added, not even basic arithmetic and logic operators!

Packages

Rhai functional features are provided in different packages that can be loaded via a call to Engine::load_package. Packages reside under rhai::packages::* and the trait rhai::packages::Package must be loaded in order for packages to be used.

use rhai::Engine;
use rhai::packages::Package                     // load the 'Package' trait to use packages
use rhai::packages::CorePackage;                // the 'core' package contains basic functionalities (e.g. arithmetic)

let mut engine = Engine::new_raw();             // create a 'raw' Engine
let package = CorePackage::new();               // create a package - can be shared among multiple `Engine` instances

engine.load_package(package.get());             // load the package manually. 'get' returns a reference to the shared package

The follow packages are available:

Package	Description	In CorePackage	In StandardPackage
ArithmeticPackage	Arithmetic operators (e.g. +, -, *, /)	Yes	Yes
BasicIteratorPackage	Numeric ranges (e.g. range(1, 10))	Yes	Yes
LogicPackage	Logic and comparison operators (e.g. ==, >)	Yes	Yes
BasicStringPackage	Basic string functions	Yes	Yes
BasicTimePackage	Basic time functions (e.g. timestamps)	Yes	Yes
MoreStringPackage	Additional string functions	No	Yes
BasicMathPackage	Basic math functions (e.g. sin, sqrt)	No	Yes
BasicArrayPackage	Basic array functions	No	Yes
BasicMapPackage	Basic object map functions	No	Yes
CorePackage	Basic essentials
StandardPackage	Standard library

Packages typically contain Rust functions that are callable within a Rhai script. All functions registered in a package is loaded under the global namespace (i.e. they're available without module qualifiers). Once a package is created (e.g. via new), it can be shared (via get) among multiple instances of Engine, even across threads (if the sync feature is turned on). Therefore, a package only has to be created once.

Packages are actually implemented as modules, so they share a lot of behavior and characteristics. The main difference is that a package loads under the global namespace, while a module loads under its own namespace alias specified in an import statement (see also modules). A package is static (i.e. pre-loaded into an Engine), while a module is dynamic (i.e. loaded with the import statement).

Custom packages can also be created. See the macro def_package!.

Evaluate expressions only

Sometimes a use case does not require a full-blown scripting language, but only needs to evaluate expressions. In these cases, use the compile_expression and eval_expression methods or their _with_scope variants.

let result = engine.eval_expression::<i64>("2 + (10 + 10) * 2")?;

When evaluation expressions, no full-blown statement (e.g. if, while, for) - not even variable assignments - is supported and will be considered parse errors when encountered.

// The following are all syntax errors because the script is not an expression.
engine.eval_expression::<()>("x = 42")?;
let ast = engine.compile_expression("let x = 42")?;
let result = engine.eval_expression_with_scope::<i64>(&mut scope, "if x { 42 } else { 123 }")?;

Values and types

The following primitive types are supported natively:

Category	Equivalent Rust types	type_of()	to_string()
Integer number	u8, i8, u16, i16, u32, i32 (default for only_i32), u64, i64 (default)	"i32", "u64" etc.	"42", "123" etc.
Floating-point number (disabled with no_float)	f32, f64 (default)	"f32" or "f64"	"123.4567" etc.
Boolean value	bool	"bool"	"true" or "false"
Unicode character	char	"char"	"A", "x" etc.
Unicode string	String (not &str)	"string"	"hello" etc.
Array (disabled with no_index)	rhai::Array	"array"	"[ ?, ?, ? ]"
Object map (disabled with no_object)	rhai::Map	"map"	#{ "a": 1, "b": 2 }
Timestamp (implemented in the BasicTimePackage)	std::time::Instant	"timestamp"	not supported
Dynamic value (i.e. can be anything)	rhai::Dynamic	the actual type	actual value
System integer (current configuration)	rhai::INT (i32 or i64)	"i32" or "i64"	"42", "123" etc.
System floating-point (current configuration, disabled with no_float)	rhai::FLOAT (f32 or f64)	"f32" or "f64"	"123.456" etc.
Nothing/void/nil/null (or whatever you want to call it)	()	"()"	"" (empty string)

All types are treated strictly separate by Rhai, meaning that i32 and i64 and u32 are completely different - they even cannot be added together. This is very similar to Rust.

The default integer type is i64. If other integer types are not needed, it is possible to exclude them and make a smaller build with the only_i64 feature.

If only 32-bit integers are needed, enabling the only_i32 feature will remove support for all integer types other than i32, including i64. This is useful on some 32-bit targets where using 64-bit integers incur a performance penalty.

If no floating-point is needed or supported, use the no_float feature to remove it.

The to_string function converts a standard type into a string for display purposes.

The type_of function detects the actual type of a value. This is useful because all variables are Dynamic in nature.

// Use 'type_of()' to get the actual types of values
type_of('c') == "char";
type_of(42) == "i64";

let x = 123;
x.type_of() == "i64";                           // method-call style is also OK
type_of(x) == "i64";

x = 99.999;
type_of(x) == "f64";

x = "hello";
if type_of(x) == "string" {
    do_something_with_string(x);
}

Dynamic values

A Dynamic value can be any type. However, if the sync feature is used, then all types must be Send + Sync.

Because type_of() a Dynamic value returns the type of the actual value, it is usually used to perform type-specific actions based on the actual value's type.

let mystery = get_some_dynamic_value();

if type_of(mystery) == "i64" {
    print("Hey, I got an integer here!");
} else if type_of(mystery) == "f64" {
    print("Hey, I got a float here!");
} else if type_of(mystery) == "string" {
    print("Hey, I got a string here!");
} else if type_of(mystery) == "bool" {
    print("Hey, I got a boolean here!");
} else if type_of(mystery) == "array" {
    print("Hey, I got an array here!");
} else if type_of(mystery) == "map" {
    print("Hey, I got an object map here!");
} else if type_of(mystery) == "TestStruct" {
    print("Hey, I got the TestStruct custom type here!");
} else {
    print("I don't know what this is: " + type_of(mystery));
}

In Rust, sometimes a Dynamic forms part of a returned value - a good example is an array with Dynamic elements, or an object map with Dynamic property values. To get the real values, the actual value types must be known in advance. There is no easy way for Rust to decide, at run-time, what type the Dynamic value is (short of using the type_name function and match against the name).

A Dynamic value's actual type can be checked via the is method. The cast method then converts the value into a specific, known type. Alternatively, use the try_cast method which does not panic but returns None when the cast fails.

let list: Array = engine.eval("...")?;          // return type is 'Array'
let item = list[0];                             // an element in an 'Array' is 'Dynamic'

item.is::<i64>() == true;                       // 'is' returns whether a 'Dynamic' value is of a particular type

let value = item.cast::<i64>();                 // if the element is 'i64', this succeeds; otherwise it panics
let value: i64 = item.cast();                   // type can also be inferred

let value = item.try_cast::<i64>().unwrap();    // 'try_cast' does not panic when the cast fails, but returns 'None'

The type_name method gets the name of the actual type as a static string slice, which you may match against.

let list: Array = engine.eval("...")?;          // return type is 'Array'
let item = list[0];                             // an element in an 'Array' is 'Dynamic'

match item.type_name() {                        // 'type_name' returns the name of the actual Rust type
    "i64" => ...
    "alloc::string::String" => ...
    "bool" => ...
    "path::to::module::TestStruct" => ...
}

The following conversion traits are implemented for Dynamic:

• From<i64> (i32 if only_i32)
• From<f64> (if not no_float)
• From<bool>
• From<String>
• From<char>
• From<Vec<T>> (into an array)
• From<HashMap<String, T>> (into an object map).

Value conversions

The to_float function converts a supported number to FLOAT (f32 or f64), and the to_int function converts a supported number to INT (i32 or i64). That's about it. For other conversions, register custom conversion functions.

let x = 42;
let y = x * 100.0;                              // <- error: cannot multiply i64 with f64
let y = x.to_float() * 100.0;                   // works
let z = y.to_int() + x;                         // works

let c = 'X';                                    // character
print("c is '" + c + "' and its code is " + c.to_int());    // prints "c is 'X' and its code is 88"

Traits

A number of traits, under the rhai:: module namespace, provide additional functionalities.

Trait	Description	Methods
RegisterFn	Trait for registering functions	register_fn
RegisterDynamicFn	Trait for registering functions returning Dynamic	register_dynamic_fn
RegisterResultFn	Trait for registering fallible functions returning Result<T, Box<EvalAltResult>>	register_result_fn
Func	Trait for creating anonymous functions from script	create_from_ast, create_from_script
ModuleResolver	Trait implemented by module resolution services	resolve

Working with functions

Rhai's scripting engine is very lightweight. It gets most of its abilities from functions. To call these functions, they need to be registered with the Engine.

use rhai::{Dynamic, Engine, EvalAltResult};
use rhai::RegisterFn;                           // use 'RegisterFn' trait for 'register_fn'
use rhai::{Dynamic, RegisterDynamicFn};         // use 'RegisterDynamicFn' trait for 'register_dynamic_fn'

// Normal function
fn add(x: i64, y: i64) -> i64 {
    x + y
}

// Function that returns a Dynamic value
fn get_an_any() -> Dynamic {
    Dynamic::from(42_i64)
}

fn main() -> Result<(), Box<EvalAltResult>>
{
    let engine = Engine::new();

    engine.register_fn("add", add);

    let result = engine.eval::<i64>("add(40, 2)")?;

    println!("Answer: {}", result);             // prints 42

    // Functions that return Dynamic values must use register_dynamic_fn()
    engine.register_dynamic_fn("get_an_any", get_an_any);

    let result = engine.eval::<i64>("get_an_any()")?;

    println!("Answer: {}", result);             // prints 42

    Ok(())
}

To return a Dynamic value from a Rust function, use the Dynamic::from method.

use rhai::Dynamic;

fn decide(yes_no: bool) -> Dynamic {
    if yes_no {
        Dynamic::from(42_i64)
    } else {
        Dynamic::from(String::from("hello!"))   // remember &str is not supported by Rhai
    }
}

Generic functions

Rust generic functions can be used in Rhai, but separate instances for each concrete type must be registered separately. This is essentially function overloading (Rhai does not natively support generics).

use std::fmt::Display;

use rhai::{Engine, RegisterFn};

fn show_it<T: Display>(x: &mut T) -> () {
    println!("put up a good show: {}!", x)
}

fn main()
{
    let engine = Engine::new();

    engine.register_fn("print", show_it as fn(x: &mut i64)->());
    engine.register_fn("print", show_it as fn(x: &mut bool)->());
    engine.register_fn("print", show_it as fn(x: &mut String)->());
}

This example shows how to register multiple functions (or, in this case, multiple overloaded versions of the same function) under the same name. This enables function overloading based on the number and types of parameters.

Fallible functions

If a function is fallible (i.e. it returns a Result<_, Error>), it can be registered with register_result_fn (using the RegisterResultFn trait).

The function must return Result<_, Box<EvalAltResult>>. Box<EvalAltResult> implements From<&str> and From<String> etc. and the error text gets converted into Box<EvalAltResult::ErrorRuntime>.

The error values are Box-ed in order to reduce memory footprint of the error path, which should be hit rarely.

use rhai::{Engine, EvalAltResult, Position};
use rhai::RegisterResultFn;                     // use 'RegisterResultFn' trait for 'register_result_fn'

// Function that may fail
fn safe_divide(x: i64, y: i64) -> Result<i64, Box<EvalAltResult>> {
    if y == 0 {
        // Return an error if y is zero
        Err("Division by zero!".into())         // short-cut to create Box<EvalAltResult::ErrorRuntime>
    } else {
        Ok(x / y)
    }
}

fn main()
{
    let engine = Engine::new();

    // Fallible functions that return Result values must use register_result_fn()
    engine.register_result_fn("divide", safe_divide);

    if let Err(error) = engine.eval::<i64>("divide(40, 0)") {
       println!("Error: {:?}", *error);         // prints ErrorRuntime("Division by zero detected!", (1, 1)")
    }
}

Overriding built-in functions

Any similarly-named function defined in a script overrides any built-in function.

// Override the built-in function 'to_int'
fn to_int(num) {
    print("Ha! Gotcha! " + num);
}

print(to_int(123));     // what happens?

Operator overloading

In Rhai, a lot of functionalities are actually implemented as functions, including basic operations such as arithmetic calculations. For example, in the expression "a + b", the + operator is not built-in, but calls a function named "+" instead!

let x = a + b;
let x = +(a, b);        // <- the above is equivalent to this function call

Similarly, comparison operators including ==, != etc. are all implemented as functions, with the stark exception of && and ||. Because they short-circuit, && and || are handled specially and not via a function; as a result, overriding them has no effect at all.

Operator functions cannot be defined as a script function (because operators syntax are not valid function names). However, operator functions can be registered to the Engine via the methods Engine::register_fn, Engine::register_result_fn etc. When a custom operator function is registered with the same name as an operator, it overloads (or overrides) the built-in version.

use rhai::{Engine, EvalAltResult, RegisterFn};

let mut engine = Engine::new();

fn strange_add(a: i64, b: i64) -> i64 { (a + b) * 42 }

engine.register_fn("+", strange_add);               // overload '+' operator for two integers!

let result: i64 = engine.eval("1 + 0");             // the overloading version is used

println!("result: {}", result);                     // prints 42

let result: f64 = engine.eval("1.0 + 0.0");         // '+' operator for two floats not overloaded

println!("result: {}", result);                     // prints 1.0

fn mixed_add(a: i64, b: f64) -> f64 { (a as f64) + b }

engine.register_fn("+", mixed_add);                 // register '+' operator for an integer and a float

let result: i64 = engine.eval("1 + 1.0");           // prints 2.0 (normally an error)

Use operator overloading for custom types (described below) only. Be very careful when overloading built-in operators because script writers expect standard operators to behave in a consistent and predictable manner, and will be annoyed if a calculation for '+' turns into a subtraction, for example.

Operator overloading also impacts script optimization when using OptimizationLevel::Full. See the relevant section for more details.

Custom types and methods

Here's an more complete example of working with Rust. First the example, then we'll break it into parts:

use rhai::{Engine, EvalAltResult};
use rhai::RegisterFn;

#[derive(Clone)]
struct TestStruct {
    field: i64
}

impl TestStruct {
    fn update(&mut self) {
        self.field += 41;
    }

    fn new() -> Self {
        TestStruct { field: 1 }
    }
}

fn main() -> Result<(), Box<EvalAltResult>>
{
    let engine = Engine::new();

    engine.register_type::<TestStruct>();

    engine.register_fn("update", TestStruct::update);
    engine.register_fn("new_ts", TestStruct::new);

    let result = engine.eval::<TestStruct>("let x = new_ts(); x.update(); x")?;

    println!("result: {}", result.field);           // prints 42

    Ok(())
}

All custom types must implement Clone. This allows the Engine to pass by value. You can turn off support for custom types via the no_object feature.

#[derive(Clone)]
struct TestStruct {
    field: i64
}

Next, we create a few methods that we'll later use in our scripts. Notice that we register our custom type with the Engine.

impl TestStruct {
    fn update(&mut self) {
        self.field += 41;
    }

    fn new() -> Self {
        TestStruct { field: 1 }
    }
}

let engine = Engine::new();

engine.register_type::<TestStruct>();

To use native types, methods and functions with the Engine, we need to register them. There are some convenience functions to help with these. Below, the update and new methods are registered with the Engine.

Note: Engine follows the convention that methods use a &mut first parameter so that invoking methods can update the value in memory.

engine.register_fn("update", TestStruct::update);   // registers 'update(&mut TestStruct)'
engine.register_fn("new_ts", TestStruct::new);      // registers 'new()'

Finally, we call our script. The script can see the function and method we registered earlier. We need to get the result back out from script land just as before, this time casting to our custom struct type.

let result = engine.eval::<TestStruct>("let x = new_ts(); x.update(); x")?;

println!("result: {}", result.field);               // prints 42

In fact, any function with a first argument (either by copy or via a &mut reference) can be used as a method call on that type because internally they are the same thing: methods on a type is implemented as a functions taking a &mut first argument.

fn foo(ts: &mut TestStruct) -> i64 {
    ts.field
}

engine.register_fn("foo", foo);

let result = engine.eval::<i64>("let x = new_ts(); x.foo()")?;

println!("result: {}", result);                     // prints 1

If the no_object feature is turned on, however, the method style of function calls (i.e. calling a function as an object-method) is no longer supported.

// Below is a syntax error under 'no_object' because 'len' cannot be called in method style.
let result = engine.eval::<i64>("let x = [1, 2, 3]; x.len()")?;

type_of() works fine with custom types and returns the name of the type. If register_type_with_name is used to register the custom type with a special "pretty-print" name, type_of() will return that name instead.

engine.register_type::<TestStruct>();
engine.register_fn("new_ts", TestStruct::new);
let x = new_ts();
print(x.type_of());                                 // prints "path::to::module::TestStruct"

engine.register_type_with_name::<TestStruct>("Hello");
engine.register_fn("new_ts", TestStruct::new);
let x = new_ts();
print(x.type_of());                                 // prints "Hello"

Getters and setters

Similarly, custom types can expose members by registering a get and/or set function.

#[derive(Clone)]
struct TestStruct {
    field: i64
}

impl TestStruct {
    fn get_field(&mut self) -> i64 {
        self.field
    }

    fn set_field(&mut self, new_val: i64) {
        self.field = new_val;
    }

    fn new() -> Self {
        TestStruct { field: 1 }
    }
}

let engine = Engine::new();

engine.register_type::<TestStruct>();

engine.register_get_set("xyz", TestStruct::get_field, TestStruct::set_field);
engine.register_fn("new_ts", TestStruct::new);

let result = engine.eval::<i64>("let a = new_ts(); a.xyz = 42; a.xyz")?;

println!("Answer: {}", result);                     // prints 42

Indexers

Custom types can also expose an indexer by registering an indexer function. A custom type with an indexer function defined can use the bracket '[]' notation to get a property value (but not update it - indexers are read-only).

#[derive(Clone)]
struct TestStruct {
    fields: Vec<i64>
}

impl TestStruct {
    fn get_field(&mut self, index: i64) -> i64 {
        self.fields[index as usize]
    }

    fn new() -> Self {
        TestStruct { fields: vec![1, 2, 42, 4, 5] }
    }
}

let engine = Engine::new();

engine.register_type::<TestStruct>();

engine.register_fn("new_ts", TestStruct::new);
engine.register_indexer(TestStruct::get_field);

let result = engine.eval::<i64>("let a = new_ts(); a[2]")?;

println!("Answer: {}", result);                     // prints 42

Needless to say, register_type, register_type_with_name, register_get, register_set, register_get_set and register_indexer are not available when the no_object feature is turned on. register_indexer is also not available when the no_index feature is turned on.

Scope - Initializing and maintaining state

By default, Rhai treats each Engine invocation as a fresh one, persisting only the functions that have been defined but no global state. This gives each evaluation a clean starting slate. In order to continue using the same global state from one invocation to the next, such a state must be manually created and passed in.

All Scope variables are Dynamic, meaning they can store values of any type. If the sync feature is used, however, then only types that are Send + Sync are supported, and the entire Scope itself will also be Send + Sync. This is extremely useful in multi-threaded applications.

In this example, a global state object (a Scope) is created with a few initialized variables, then the same state is threaded through multiple invocations:

use rhai::{Engine, Scope, EvalAltResult};

fn main() -> Result<(), Box<EvalAltResult>>
{
    let engine = Engine::new();

    // First create the state
    let mut scope = Scope::new();

    // Then push (i.e. add) some initialized variables into the state.
    // Remember the system number types in Rhai are i64 (i32 if 'only_i32') ond f64.
    // Better stick to them or it gets hard working with the script.
    scope.push("y", 42_i64);
    scope.push("z", 999_i64);

    // 'set_value' adds a variable when one doesn't exist
    scope.set_value("s", "hello, world!".to_string());  // remember to use 'String', not '&str'

    // First invocation
    engine.eval_with_scope::<()>(&mut scope, r"
        let x = 4 + 5 - y + z + s.len();
        y = 1;
    ")?;

    // Second invocation using the same state
    let result = engine.eval_with_scope::<i64>(&mut scope, "x")?;

    println!("result: {}", result);                     // prints 979

    // Variable y is changed in the script - read it with 'get_value'
    assert_eq!(scope.get_value::<i64>("y").expect("variable y should exist"), 1);

    // We can modify scope variables directly with 'set_value'
    scope.set_value("y", 42_i64);
    assert_eq!(scope.get_value::<i64>("y").expect("variable y should exist"), 42);

    Ok(())
}

Engine configuration options

Method	Description
set_optimization_level	Set the amount of script optimizations performed. See script optimization.
set_max_expr_depths	Set the maximum nesting levels of an expression/statement. See maximum statement depth.
set_max_call_levels	Set the maximum number of function call levels (default 50) to avoid infinite recursion. See maximum call stack depth.
set_max_operations	Set the maximum number of operations that a script is allowed to consume. See maximum number of operations.
set_max_modules	Set the maximum number of modules that a script is allowed to load. See maximum number of modules.

---

Rhai Language Guide

Comments

Comments are C-style, including '/* ... */' pairs and '//' for comments to the end of the line.

let /* intruder comment */ name = "Bob";

// This is a very important comment

/* This comment spans
   multiple lines, so it
   only makes sense that
   it is even more important */

/* Fear not, Rhai satisfies all nesting needs with nested comments:
   /*/*/*/*/**/*/*/*/*/
*/

Statements

Statements are terminated by semicolons ';' - they are mandatory, except for the last statement where it can be omitted.

A statement can be used anywhere where an expression is expected. The last statement of a statement block (enclosed by '{' .. '}' pairs) is always the return value of the statement. If a statement has no return value (e.g. variable definitions, assignments) then the value will be ().

let a = 42;             // normal assignment statement
let a = foo(42);        // normal function call statement
foo < 42;               // normal expression as statement

let a = { 40 + 2 };     // 'a' is set to the value of the statement block, which is the value of the last statement
//              ^ notice that the last statement does not require a terminating semicolon (although it also works with it)
//                ^ notice that a semicolon is required here to terminate the assignment statement; it is syntax error without it

4 * 10 + 2              // this is also a statement, which is an expression, with no ending semicolon because
                        // it is the last statement of the whole block

Variables

Variables in Rhai follow normal C naming rules (i.e. must contain only ASCII letters, digits and underscores '_').

Variable names must start with an ASCII letter or an underscore '_', must contain at least one ASCII letter, and must start with an ASCII letter before a digit. Therefore, names like '_', '_42', '3a' etc. are not legal variable names, but '_c3po' and 'r2d2' are. Variable names are also case sensitive.

Variables are defined using the let keyword. A variable defined within a statement block is local to that block.

let x = 3;              // ok
let _x = 42;            // ok
let x_ = 42;            // also ok
let _x_ = 42;           // still ok

let _ = 123;            // <- syntax error: illegal variable name
let _9 = 9;             // <- syntax error: illegal variable name

let x = 42;             // variable is 'x', lower case
let X = 123;            // variable is 'X', upper case
x == 42;
X == 123;

{
    let x = 999;        // local variable 'x' shadows the 'x' in parent block
    x == 999;           // access to local 'x'
}
x == 42;                // the parent block's 'x' is not changed

Constants

Constants can be defined using the const keyword and are immutable. Constants follow the same naming rules as variables.

const x = 42;
print(x * 2);           // prints 84
x = 123;                // <- syntax error: cannot assign to constant

Constants must be assigned a value, not an expression.

const x = 40 + 2;       // <- syntax error: cannot assign expression to constant

Numbers

Integer numbers follow C-style format with support for decimal, binary ('0b'), octal ('0o') and hex ('0x') notations.

The default system integer type (also aliased to INT) is i64. It can be turned into i32 via the only_i32 feature.

Floating-point numbers are also supported if not disabled with no_float. The default system floating-point type is i64 (also aliased to FLOAT).

'_' separators can be added freely and are ignored within a number.

Format	Type
123_345, -42	i64 in decimal
0o07_76	i64 in octal
0xabcd_ef	i64 in hex
0b0101_1001	i64 in binary
123_456.789	f64

Numeric operators

Numeric operators generally follow C styles.

Operator	Description	Integers only
+	Plus
-	Minus
*	Multiply
/	Divide (integer division if acting on integer types)
%	Modulo (remainder)
~	Power
&	Binary And bit-mask	Yes
|	Binary Or bit-mask	Yes
^	Binary Xor bit-mask	Yes
<<	Left bit-shift	Yes
>>	Right bit-shift	Yes

let x = (1 + 2) * (6 - 4) / 2;  // arithmetic, with parentheses
let reminder = 42 % 10;         // modulo
let power = 42 ~ 2;             // power (i64 and f64 only)
let left_shifted = 42 << 3;     // left shift
let right_shifted = 42 >> 3;    // right shift
let bit_op = 42 | 99;           // bit masking

Unary operators

Operator	Description
+	Plus
-	Negative

let number = -5;
number = -5 - +5;

Numeric functions

The following standard functions (defined in the [BasicMathPackage] but excluded if using a raw Engine) operate on i8, i16, i32, i64, f32 and f64 only:

Function	Description
abs	absolute value
to_float	converts an integer type to f64

Floating-point functions

The following standard functions (defined in the BasicMathPackage but excluded if using a raw Engine) operate on f64 only:

Category	Functions
Trigonometry	sin, cos, tan, sinh, cosh, tanh in degrees
Arc-trigonometry	asin, acos, atan, asinh, acosh, atanh in degrees
Square root	sqrt
Exponential	exp (base e)
Logarithmic	ln (base e), log10 (base 10), log (any base)
Rounding	floor, ceiling, round, int, fraction
Conversion	to_int
Testing	is_nan, is_finite, is_infinite

Strings and Chars

String and char literals follow C-style formatting, with support for Unicode ('\uxxxx' or '\Uxxxxxxxx') and hex ('\xxx') escape sequences.

Hex sequences map to ASCII characters, while '\u' maps to 16-bit common Unicode code points and '\U' maps the full, 32-bit extended Unicode code points.

Standard escape sequences:

Escape sequence	Meaning
\\	back-slash \
\t	tab
\r	carriage-return CR
\n	line-feed LF
\"	double-quote " in strings
\'	single-quote ' in characters
\xxx	Unicode in 2-digit hex
\uxxxx	Unicode in 4-digit hex
\Uxxxxxxxx	Unicode in 8-digit hex

Internally Rhai strings are stored as UTF-8 just like Rust (they are Rust String's!), but there are major differences. In Rhai a string is the same as an array of Unicode characters and can be directly indexed (unlike Rust). This is similar to most other languages where strings are internally represented not as UTF-8 but as arrays of multi-byte Unicode characters. Individual characters within a Rhai string can also be replaced just as if the string is an array of Unicode characters. In Rhai, there is also no separate concepts of String and &str as in Rust.

Strings can be built up from other strings and types via the + operator (provided by the MoreStringPackage but excluded if using a raw Engine). This is particularly useful when printing output.

type_of() a string returns "string".

let name = "Bob";
let middle_initial = 'C';
let last = "Davis";

let full_name = name + " " + middle_initial + ". " + last;
full_name == "Bob C. Davis";

// String building with different types
let age = 42;
let record = full_name + ": age " + age;
record == "Bob C. Davis: age 42";

// Unlike Rust, Rhai strings can be indexed to get a character
// (disabled with 'no_index')
let c = record[4];
c == 'C';

ts.s = record;                          // custom type properties can take strings

let c = ts.s[4];
c == 'C';

let c = "foo"[0];                       // indexing also works on string literals...
c == 'f';

let c = ("foo" + "bar")[5];             // ... and expressions returning strings
c == 'r';

// Escape sequences in strings
record += " \u2764\n";                  // escape sequence of '❤' in Unicode
record == "Bob C. Davis: age 42 ❤\n";   // '\n' = new-line

// Unlike Rust, Rhai strings can be directly modified character-by-character
// (disabled with 'no_index')
record[4] = '\x58'; // 0x58 = 'X'
record == "Bob X. Davis: age 42 ❤\n";

// Use 'in' to test if a substring (or character) exists in a string
"Davis" in record == true;
'X' in record == true;
'C' in record == false;

Built-in functions

The following standard methods (defined in the MoreStringPackage but excluded if using a raw Engine) operate on strings:

Function	Parameter(s)	Description
len	none	returns the number of characters (not number of bytes) in the string
pad	character to pad, target length	pads the string with an character to at least a specified length
append	character/string to append	Adds a character or a string to the end of another string
clear	none	empties the string
truncate	target length	cuts off the string at exactly a specified number of characters
contains	character/sub-string to search for	checks if a certain character or sub-string occurs in the string
index_of	character/sub-string to search for, start index (optional)	returns the index that a certain character or sub-string occurs in the string, or -1 if not found
sub_string	start index, length (optional)	extracts a sub-string (to the end of the string if length is not specified)
crop	start index, length (optional)	retains only a portion of the string (to the end of the string if length is not specified)
replace	target character/sub-string, replacement character/string	replaces a sub-string with another
trim	none	trims the string of whitespace at the beginning and end

Examples

let full_name == " Bob C. Davis ";
full_name.len() == 14;

full_name.trim();
full_name.len() == 12;
full_name == "Bob C. Davis";

full_name.pad(15, '$');
full_name.len() == 15;
full_name == "Bob C. Davis$$$";

let n = full_name.index_of('$');
n == 12;

full_name.index_of("$$", n + 1) == 13;

full_name.sub_string(n, 3) == "$$$";

full_name.truncate(6);
full_name.len() == 6;
full_name == "Bob C.";

full_name.replace("Bob", "John");
full_name.len() == 7;
full_name == "John C.";

full_name.contains('C') == true;
full_name.contains("John") == true;

full_name.crop(5);
full_name == "C.";

full_name.crop(0, 1);
full_name == "C";

full_name.clear();
full_name.len() == 0;

Arrays

Arrays are first-class citizens in Rhai. Like C, arrays are accessed with zero-based, non-negative integer indices. Array literals are built within square brackets '[' ... ']' and separated by commas ','. All elements stored in an array are Dynamic, and the array can freely grow or shrink with elements added or removed.

The Rust type of a Rhai array is rhai::Array. type_of() an array returns "array".

Arrays are disabled via the no_index feature.

Built-in functions

The following methods (defined in the BasicArrayPackage but excluded if using a raw Engine) operate on arrays:

Function	Parameter(s)	Description
push	element to insert	inserts an element at the end
append	array to append	concatenates the second array to the end of the first
+ operator	first array, second array	concatenates the first array with the second
insert	element to insert, position (beginning if <= 0, end if >= length)	insert an element at a certain index
pop	none	removes the last element and returns it (() if empty)
shift	none	removes the first element and returns it (() if empty)
remove	index	removes an element at a particular index and returns it, or returns () if the index is not valid
len	none	returns the number of elements
pad	element to pad, target length	pads the array with an element to at least a specified length
clear	none	empties the array
truncate	target length	cuts off the array at exactly a specified length (discarding all subsequent elements)

Examples

let y = [2, 3];         // array literal with 2 elements

y.insert(0, 1);         // insert element at the beginning
y.insert(999, 4);       // insert element at the end

y.len() == 4;

y[0] == 1;
y[1] == 2;
y[2] == 3;
y[3] == 4;

(1 in y) == true;       // use 'in' to test if an item exists in the array
(42 in y) == false;     // 'in' uses the '==' operator (which users can override)
                        // to check if the target item exists in the array

y[1] = 42;              // array elements can be reassigned

(42 in y) == true;

y.remove(2) == 3;       // remove element

y.len() == 3;

y[2] == 4;              // elements after the removed element are shifted

ts.list = y;            // arrays can be assigned completely (by value copy)
let foo = ts.list[1];
foo == 42;

let foo = [1, 2, 3][0];
foo == 1;

fn abc() {
    [42, 43, 44]        // a function returning an array
}

let foo = abc()[0];
foo == 42;

let foo = y[0];
foo == 1;

y.push(4);              // 4 elements
y.push(5);              // 5 elements

y.len() == 5;

let first = y.shift();  // remove the first element, 4 elements remaining
first == 1;

let last = y.pop();     // remove the last element, 3 elements remaining
last == 5;

y.len() == 3;

for item in y {         // arrays can be iterated with a 'for' statement
    print(item);
}

y.pad(10, "hello");     // pad the array up to 10 elements

y.len() == 10;

y.truncate(5);          // truncate the array to 5 elements

y.len() == 5;

y.clear();              // empty the array

y.len() == 0;

push and pad are only defined for standard built-in types. For custom types, type-specific versions must be registered:

engine.register_fn("push", |list: &mut Array, item: MyType| list.push(Box::new(item)) );

Object maps

Object maps are dictionaries. Properties are all Dynamic and can be freely added and retrieved. Object map literals are built within braces '#{' ... '}' (name : value syntax similar to Rust) and separated by commas ','. The property name can be a simple variable name following the same naming rules as variables, or an arbitrary string literal.

Property values can be accessed via the dot notation (object . property) or index notation (object [ property ]). The dot notation allows only property names that follow the same naming rules as variables. The index notation allows setting/getting properties of arbitrary names (even the empty string).

Important: Trying to read a non-existent property returns () instead of causing an error.

The Rust type of a Rhai object map is rhai::Map. type_of() an object map returns "map".

Object maps are disabled via the no_object feature.

Built-in functions

The following methods (defined in the BasicMapPackage but excluded if using a raw Engine) operate on object maps:

Function	Parameter(s)	Description
has	property name	does the object map contain a property of a particular name?
len	none	returns the number of properties
clear	none	empties the object map
remove	property name	removes a certain property and returns it (() if the property does not exist)
mixin	second object map	mixes in all the properties of the second object map to the first (values of properties with the same names replace the existing values)
+ operator	first object map, second object map	merges the first object map with the second
keys	none	returns an array of all the property names (in random order), not available under no_index
values	none	returns an array of all the property values (in random order), not available under no_index

Examples

let y = #{              // object map literal with 3 properties
    a: 1,
    bar: "hello",
    "baz!$@": 123.456,  // like JS, you can use any string as property names...
    "": false,          // even the empty string!

    a: 42               // <- syntax error: duplicated property name
};

y.a = 42;               // access via dot notation
y.baz!$@ = 42;          // <- syntax error: only proper variable names allowed in dot notation
y."baz!$@" = 42;        // <- syntax error: strings not allowed in dot notation

y.a == 42;

y["baz!$@"] == 123.456; // access via index notation

"baz!$@" in y == true;  // use 'in' to test if a property exists in the object map
("z" in y) == false;

ts.obj = y;             // object maps can be assigned completely (by value copy)
let foo = ts.list.a;
foo == 42;

let foo = #{ a:1, b:2, c:3 }["a"];
foo == 1;

fn abc() {
    #{ a:1, b:2, c:3 }  // a function returning an object map
}

let foo = abc().b;
foo == 2;

let foo = y["a"];
foo == 42;

y.has("a") == true;
y.has("xyz") == false;

y.xyz == ();            // a non-existing property returns '()'
y["xyz"] == ();

y.len() == 3;

y.remove("a") == 1;     // remove property

y.len() == 2;
y.has("a") == false;

for name in keys(y) {   // get an array of all the property names via the 'keys' function
    print(name);
}

for val in values(y) {  // get an array of all the property values via the 'values' function
    print(val);
}

y.clear();              // empty the object map

y.len() == 0;

Parsing from JSON

The syntax for an object map is extremely similar to JSON, with the exception of null values which can technically be mapped to (). A valid JSON string does not start with a hash character # while a Rhai object map does - that's the major difference!

JSON numbers are all floating-point while Rhai supports integers (INT) and floating-point (FLOAT) if the no_float feature is not turned on. Most common generators of JSON data distinguish between integer and floating-point values by always serializing a floating-point number with a decimal point (i.e. 123.0 instead of 123 which is assumed to be an integer). This style can be used successfully with Rhai object maps.

Use the parse_json method to parse a piece of JSON into an object map:

// JSON string - notice that JSON property names are always quoted
//               notice also that comments are acceptable within the JSON string
let json = r#"{
                "a": 1,                 // <- this is an integer number
                "b": true,
                "c": 123.0,             // <- this is a floating-point number
                "$d e f!": "hello",     // <- any text can be a property name
                "^^^!!!": [1,42,"999"], // <- value can be array or another hash
                "z": null               // <- JSON 'null' value
              }
"#;

// Parse the JSON expression as an object map
// Set the second boolean parameter to true in order to map 'null' to '()'
let map = engine.parse_json(json, true)?;

map.len() == 6;                         // 'map' contains all properties in the JSON string

// Put the object map into a 'Scope'
let mut scope = Scope::new();
scope.push("map", map);

let result = engine.eval_with_scope::<INT>(r#"map["^^^!!!"].len()"#)?;

result == 3;                            // the object map is successfully used in the script

timestamp's

Timestamps are provided by the BasicTimePackage (excluded if using a raw Engine) via the timestamp function.

The Rust type of a timestamp is std::time::Instant. type_of() a timestamp returns "timestamp".

Built-in functions

The following methods (defined in the BasicTimePackage but excluded if using a raw Engine) operate on timestamps:

Function	Parameter(s)	Description
elapsed	none	returns the number of seconds since the timestamp
- operator	later timestamp, earlier timestamp	returns the number of seconds between the two timestamps

Examples

let now = timestamp();

// Do some lengthy operation...

if now.elapsed() > 30.0 {
    print("takes too long (over 30 seconds)!")
}

Comparison operators

Comparing most values of the same data type work out-of-the-box for standard types supported by the system.

However, if using a raw Engine, comparisons can only be made between restricted system types - INT (i64 or i32 depending on only_i32 and only_i64), f64 (if not no_float), string, array, bool, char.

42 == 42;               // true
42 > 42;                // false
"hello" > "foo";        // true
"42" == 42;             // false

Comparing two values of different data types, or of unknown data types, always results in false.

42 == 42.0;             // false - i64 is different from f64
42 > "42";              // false - i64 is different from string
42 <= "42";             // false again

let ts = new_ts();      // custom type
ts == 42;               // false - types are not the same

Boolean operators

Operator	Description
!	Boolean Not
&&	Boolean And (short-circuits)
||	Boolean Or (short-circuits)
&	Boolean And (doesn't short-circuit)
|	Boolean Or (doesn't short-circuit)

Double boolean operators && and || short-circuit, meaning that the second operand will not be evaluated if the first one already proves the condition wrong.

Single boolean operators & and | always evaluate both operands.

this() || that();       // that() is not evaluated if this() is true
this() && that();       // that() is not evaluated if this() is false

this() | that();        // both this() and that() are evaluated
this() & that();        // both this() and that() are evaluated

Compound assignment operators

let number = 5;
number += 4;            // number = number + 4
number -= 3;            // number = number - 3
number *= 2;            // number = number * 2
number /= 1;            // number = number / 1
number %= 3;            // number = number % 3
number <<= 2;           // number = number << 2
number >>= 1;           // number = number >> 1

The += operator can also be used to build strings:

let my_str = "abc";
my_str += "ABC";
my_str += 12345;

my_str == "abcABC12345"

if statements

if foo(x) {
    print("It's true!");
} else if bar == baz {
    print("It's true again!");
} else if ... {
        :
} else if ... {
        :
} else {
    print("It's finally false!");
}

All branches of an if statement must be enclosed within braces '{' .. '}', even when there is only one statement. Like Rust, there is no ambiguity regarding which if clause a statement belongs to.

if (decision) print("I've decided!");
//            ^ syntax error, expecting '{' in statement block

Like Rust, if statements can also be used as expressions, replacing the ? : conditional operators in other C-like languages.

// The following is equivalent to C: int x = 1 + (decision ? 42 : 123) / 2;
let x = 1 + if decision { 42 } else { 123 } / 2;
x == 22;

let x = if decision { 42 }; // no else branch defaults to '()'
x == ();

while loops

let x = 10;

while x > 0 {
    x = x - 1;
    if x < 6 { continue; }  // skip to the next iteration
    print(x);
    if x == 5 { break; }    // break out of while loop
}

Infinite loop

let x = 10;

loop {
    x = x - 1;
    if x > 5 { continue; }  // skip to the next iteration
    print(x);
    if x == 0 { break; }    // break out of loop
}

for loops

Iterating through a range or an array is provided by the for ... in loop.

let array = [1, 3, 5, 7, 9, 42];

// Iterate through array
for x in array {
    if x > 10 { continue; } // skip to the next iteration
    print(x);
    if x == 42 { break; }   // break out of for loop
}

// The 'range' function allows iterating from first to last-1
for x in range(0, 50) {
    if x > 10 { continue; } // skip to the next iteration
    print(x);
    if x == 42 { break; }   // break out of for loop
}

// The 'range' function also takes a step
for x in range(0, 50, 3) {  // step by 3
    if x > 10 { continue; } // skip to the next iteration
    print(x);
    if x == 42 { break; }   // break out of for loop
}

// Iterate through object map
let map = #{a:1, b:3, c:5, d:7, e:9};

// Property names are returned in random order
for x in keys(map) {
    if x > 10 { continue; } // skip to the next iteration
    print(x);
    if x == 42 { break; }   // break out of for loop
}

// Property values are returned in random order
for val in values(map) {
    print(val);
}

return-ing values

return;                     // equivalent to return ();

return 123 + 456;           // returns 579

Errors and throw-ing exceptions

All of Engine's evaluation/consuming methods return Result<T, Box<rhai::EvalAltResult>> with EvalAltResult holding error information. To deliberately return an error during an evaluation, use the throw keyword.

if some_bad_condition_has_happened {
    throw error;            // 'throw' takes a string as the exception text
}

throw;                      // defaults to empty exception text: ""

Exceptions thrown via throw in the script can be captured by matching Err(EvalAltResult::ErrorRuntime( reason , position )) with the exception text captured by the first parameter.

let result = engine.eval::<i64>(r#"
    let x = 42;

    if x > 0 {
        throw x + " is too large!";
    }
"#);

println!(result);           // prints "Runtime error: 42 is too large! (line 5, position 15)"

Functions

Rhai supports defining functions in script (unless disabled with no_function):

fn add(x, y) {
    return x + y;
}

print(add(2, 3));

Implicit return

Just like in Rust, an implicit return can be used. In fact, the last statement of a block is always the block's return value regardless of whether it is terminated with a semicolon ';'. This is different from Rust.

fn add(x, y) {              // implicit return:
    x + y;                  // value of the last statement (no need for ending semicolon)
                            // is used as the return value
}

fn add2(x) {
    return x + 2;           // explicit return
}

print(add(2, 3));           // prints 5
print(add2(42));            // prints 44

No access to external scope

Functions are not closures. They do not capture the calling environment and can only access their own parameters. They cannot access variables external to the function itself.

let x = 42;

fn foo() { x }              // <- syntax error: variable 'x' doesn't exist

Passing arguments by value

Functions defined in script always take Dynamic parameters (i.e. the parameter can be of any type). It is important to remember that all arguments are passed by value, so all functions are pure (i.e. they never modifytheir arguments). Any update to an argument will not be reflected back to the caller. This can introduce subtle bugs, if not careful.

fn change(s) {              // 's' is passed by value
    s = 42;                 // only a COPY of 's' is changed
}

let x = 500;
x.change();                 // de-sugars to change(x)
x == 500;                   // 'x' is NOT changed!

Global definitions only

Functions can only be defined at the global level, never inside a block or another function.

// Global level is OK
fn add(x, y) {
    x + y
}

// The following will not compile
fn do_addition(x) {
    fn add_y(n) {           // <- syntax error: functions cannot be defined inside another function
        n + y
    }

    add_y(x)
}

Unlike C/C++, functions can be defined anywhere within the global level. A function does not need to be defined prior to being used in a script; a statement in the script can freely call a function defined afterwards. This is similar to Rust and many other modern languages.

Function overloading

Functions can be overloaded and are resolved purely upon the function's name and the number of parameters (but not parameter types, since all parameters are the same type - Dynamic). New definitions overwrite previous definitions of the same name and number of parameters.

fn foo(x,y,z) { print("Three!!! " + x + "," + y + "," + z) }
fn foo(x) { print("One! " + x) }
fn foo(x,y) { print("Two! " + x + "," + y) }
fn foo() { print("None.") }
fn foo(x) { print("HA! NEW ONE! " + x) }    // overwrites previous definition

foo(1,2,3);                 // prints "Three!!! 1,2,3"
foo(42);                    // prints "HA! NEW ONE! 42"
foo(1,2);                   // prints "Two!! 1,2"
foo();                      // prints "None."

Members and methods

Properties and methods in a Rust custom type registered with the Engine can be called just like in Rust.

let a = new_ts();           // constructor function
a.field = 500;              // property access
a.update();                 // method call, 'a' can be changed

update(a);                  // this works, but 'a' is unchanged because only
                            // a COPY of 'a' is passed to 'update' by VALUE

Custom types, properties and methods can be disabled via the no_object feature.

print and debug

The print and debug functions default to printing to stdout, with debug using standard debug formatting.

print("hello");             // prints hello to stdout
print(1 + 2 + 3);           // prints 6 to stdout
print("hello" + 42);        // prints hello42 to stdout
debug("world!");            // prints "world!" to stdout using debug formatting

Overriding print and debug with callback functions

When embedding Rhai into an application, it is usually necessary to trap print and debug output (for logging into a tracking log, for example) with the Engine::on_print and Engine::on_debug methods:

// Any function or closure that takes an '&str' argument can be used to override
// 'print' and 'debug'
engine.on_print(|x| println!("hello: {}", x));
engine.on_debug(|x| println!("DEBUG: {}", x));

// Example: quick-'n-dirty logging
let logbook = Arc::new(RwLock::new(Vec::<String>::new()));

// Redirect print/debug output to 'log'
let log = logbook.clone();
engine.on_print(move |s| log.write().unwrap().push(format!("entry: {}", s)));

let log = logbook.clone();
engine.on_debug(move |s| log.write().unwrap().push(format!("DEBUG: {}", s)));

// Evaluate script
engine.eval::<()>(script)?;

// 'logbook' captures all the 'print' and 'debug' output
for entry in logbook.read().unwrap().iter() {
    println!("{}", entry);
}

Modules

Rhai allows organizing code (functions, both Rust-based or script-based, and variables) into modules. Modules can be disabled via the no_module feature.

Exporting variables and functions from modules

A module is a single script (or pre-compiled AST) containing global variables and functions. The export statement, which can only be at global level, exposes selected variables as members of a module. Variables not exported are private and invisible to the outside. On the other hand, all functions are automatically exported, unless it is explicitly opt-out with the private prefix. Functions declared private are invisible to the outside.

Everything exported from a module is constant (read-only).

// This is a module script.

fn inc(x) { x + 1 }         // script-defined function - default public

private fn foo() {}         // private function - invisible to outside

let private = 123;          // variable not exported - default invisible to outside
let x = 42;                 // this will be exported below

export x;                   // the variable 'x' is exported under its own name

export x as answer;         // the variable 'x' is exported under the alias 'answer'
                            // another script can load this module and access 'x' as 'module::answer'

Importing modules

A module can be imported via the import statement, and its members are accessed via '::' similar to C++.

import "crypto" as crypto;  // import the script file 'crypto.rhai' as a module

crypto::encrypt(secret);    // use functions defined under the module via '::'

crypto::hash::sha256(key);  // sub-modules are also supported

print(crypto::status);      // module variables are constants

crypto::status = "off";     // <- runtime error - cannot modify a constant

import statements are scoped, meaning that they are only accessible inside the scope that they're imported. They can appear anywhere a normal statement can be, but in the vast majority of cases import statements are group at the beginning of a script. It is not advised to deviate from this common practice unless there is a Very Good Reason™. Especially, do not place an import statement within a loop; doing so will repeatedly re-load the same module during every iteration of the loop!

let mod = "crypto";

if secured {                // new block scope
    import mod as crypto;   // import module (the path needs not be a constant string)

    crypto::encrypt(key);   // use a function in the module
}                           // the module disappears at the end of the block scope

crypto::encrypt(others);    // <- this causes a run-time error because the 'crypto' module
                            //    is no longer available!

for x in range(0, 1000) {
    import "crypto" as c;   // <- importing a module inside a loop is a Very Bad Idea™

    c.encrypt(something);
}

Creating custom modules with Rust

To load a custom module (written in Rust) into an Engine, first create a Module type, add variables/functions into it, then finally push it into a custom Scope. This has the equivalent effect of putting an import statement at the beginning of any script run.

use rhai::{Engine, Scope, Module, i64};

let mut engine = Engine::new();
let mut scope = Scope::new();

let mut module = Module::new();             // new module
module.set_var("answer", 41_i64);           // variable 'answer' under module
module.set_fn_1("inc", |x: i64| Ok(x+1));   // use the 'set_fn_XXX' API to add functions

// Push the module into the custom scope under the name 'question'
// This is equivalent to 'import "..." as question;'
scope.push_module("question", module);

// Use module-qualified variables
engine.eval_expression_with_scope::<i64>(&scope, "question::answer + 1")? == 42;

// Call module-qualified functions
engine.eval_expression_with_scope::<i64>(&scope, "question::inc(question::answer)")? == 42;

Creating a module from an AST

It is easy to convert a pre-compiled AST into a module: just use Module::eval_ast_as_new. Don't forget the export statement, otherwise there will be no variables exposed by the module other than non-private functions (unless that's intentional).

use rhai::{Engine, Module};

let engine = Engine::new();

// Compile a script into an 'AST'
let ast = engine.compile(r#"
    // Functions become module functions
    fn calc(x) {
        x + 1
    }
    fn add_len(x, y) {
        x + y.len()
    }

    // Imported modules can become sub-modules
    import "another module" as extra;

    // Variables defined at global level can become module variables
    const x = 123;
    let foo = 41;
    let hello;

    // Variable values become constant module variable values
    foo = calc(foo);
    hello = "hello, " + foo + " worlds!";

    // Finally, export the variables and modules
    export
        x as abc,           // aliased variable name
        foo,
        hello,
        extra as foobar;    // export sub-module
"#)?;

// Convert the 'AST' into a module, using the 'Engine' to evaluate it first
let module = Module::eval_ast_as_new(Scope::new(), &ast, &engine)?;

// 'module' now can be loaded into a custom 'Scope' for future use.  It contains:
//   - sub-module: 'foobar' (renamed from 'extra')
//   - functions: 'calc', 'add_len'
//   - variables: 'abc' (renamed from 'x'), 'foo', 'hello'

Module resolvers

When encountering an import statement, Rhai attempts to resolve the module based on the path string. Module Resolvers are service types that implement the ModuleResolver trait. There are a number of standard resolvers built into Rhai, the default being the FileModuleResolver which simply loads a script file based on the path (with .rhai extension attached) and execute it to form a module.

Built-in module resolvers are grouped under the rhai::module_resolvers module namespace.

Module Resolver	Description
FileModuleResolver	The default module resolution service, not available under the no_std feature. Loads a script file (based off the current directory) with .rhai extension. The base directory can be changed via the FileModuleResolver::new_with_path() constructor function. FileModuleResolver::create_module() loads a script file and returns a module.
StaticModuleResolver	Loads modules that are statically added. This can be used when the no_std feature is turned on.

An Engine's module resolver is set via a call to Engine::set_module_resolver:

// Use the 'StaticModuleResolver'
let resolver = rhai::module_resolvers::StaticModuleResolver::new();
engine.set_module_resolver(Some(resolver));

// Effectively disable 'import' statements by setting module resolver to 'None'
engine.set_module_resolver(None);

Ruggedization - protect against DoS attacks

For scripting systems open to user-land scripts, it is always best to limit the amount of resources used by a script so that it does not consume more resources that it is allowed to.

The most important resources to watch out for are:

• Memory: A malignant script may continuously grow an array or object map until all memory is consumed. It may also create a large array or [objecct map] literal that exhausts all memory during parsing.
• CPU: A malignant script may run an infinite tight loop that consumes all CPU cycles.
• Time: A malignant script may run indefinitely, thereby blocking the calling system which is waiting for a result.
• Stack: A malignant script may attempt an infinite recursive call that exhausts the call stack. Alternatively, it may create a degenerated deep expression with so many levels that the parser exhausts the call stack when parsing the expression; or even deeply-nested statement blocks, if nested deep enough.
• Overflows: A malignant script may deliberately cause numeric over-flows and/or under-flows, divide by zero, and/or create bad floating-point representations, in order to crash the system.
• Files: A malignant script may continuously import an external module within an infinite loop, thereby putting heavy load on the file-system (or even the network if the file is not local). Even when modules are not created from files, they still typically consume a lot of resources to load.
• Data: A malignant script may attempt to read from and/or write to data that it does not own. If this happens, it is a severe security breach and may put the entire system at risk.

Maximum number of operations

Rhai by default does not limit how much time or CPU a script consumes. This can be changed via the Engine::set_max_operations method, with zero being unlimited (the default).

let mut engine = Engine::new();

engine.set_max_operations(500);             // allow only up to 500 operations for this script

engine.set_max_operations(0);               // allow unlimited operations

The concept of one single operation in Rhai is volatile - it roughly equals one expression node, loading one variable/constant, one operator call, one iteration of a loop, or one function call etc. with sub-expressions, statements and function calls executed inside these contexts accumulated on top. A good rule-of-thumb is that one simple non-trivial expression consumes on average 5-10 operations.

One operation can take an unspecified amount of time and real CPU cycles, depending on the particulars. For example, loading a constant consumes very few CPU cycles, while calling an external Rust function, though also counted as only one operation, may consume much more computing resources. If it helps to visualize, think of an operation as roughly equals to one instruction of a hypothetical CPU.

The operation count is intended to be a very course-grained measurement of the amount of CPU that a script is consuming, and allows the system to impose a hard upper limit.

A script exceeding the maximum operations count will terminate with an error result. This check can be disabled via the unchecked feature for higher performance (but higher risks as well).

Tracking progress

To track script evaluation progress and to force-terminate a script prematurely (for any reason), provide a closure to the Engine::on_progress method:

let mut engine = Engine::new();

engine.on_progress(|count| {                // 'count' is the number of operations performed
    if count % 1000 == 0 {
        println!("{}", count);              // print out a progress log every 1,000 operations
    }
    true                                    // return 'true' to continue the script
                                            // returning 'false' will terminate the script
});

The closure passed to Engine::on_progress will be called once every operation. Return false to terminate the script immediately.

Maximum number of modules

Rhai by default does not limit how many modules are loaded via the import statement. This can be changed via the Engine::set_max_modules method, with zero being unlimited (the default).

let mut engine = Engine::new();

engine.set_max_modules(5);                  // allow loading only up to 5 modules

engine.set_max_modules(0);                  // allow unlimited modules

Maximum call stack depth

Rhai by default limits function calls to a maximum depth of 128 levels (8 levels in debug build). This limit may be changed via the Engine::set_max_call_levels method.

When setting this limit, care must be also taken to the evaluation depth of each statement within the function. It is entirely possible for a malignant script to embed an recursive call deep inside a nested expression or statement block (see maximum statement depth).

The limit can be disabled via the unchecked feature for higher performance (but higher risks as well).

let mut engine = Engine::new();

engine.set_max_call_levels(10);             // allow only up to 10 levels of function calls

engine.set_max_call_levels(0);              // allow no function calls at all (max depth = zero)

A script exceeding the maximum call stack depth will terminate with an error result.

Maximum statement depth

Rhai by default limits statements and expressions nesting to a maximum depth of 128 (which should be plenty) when they are at global level, but only a depth of 32 when they are within function bodies. For debug builds, these limits are set further downwards to 32 and 16 respectively.

That is because it is possible to overflow the Engine's stack when it tries to recursively parse an extremely deeply-nested code stream.

// The following, if long enough, can easily cause stack overflow during parsing.
let a = (1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(...)+1)))))))))));

This limit may be changed via the Engine::set_max_expr_depths method. There are two limits to set, one for the maximum depth at global level, and the other for function bodies.

let mut engine = Engine::new();

engine.set_max_expr_depths(50, 5);          // allow nesting up to 50 layers of expressions/statements
                                            // at global level, but only 5 inside functions

Beware that there may be multiple layers for a simple language construct, even though it may correspond to only one AST node. That is because the Rhai parser internally runs a recursive chain of function calls and it is important that a malignant script does not panic the parser in the first place.

Functions are placed under stricter limits because of the multiplicative effect of recursion. A script can effectively call itself while deep inside an expression chain within the function body, thereby overflowing the stack even when the level of recursion is within limit.

Make sure that C x ( 5 + F ) + S layered calls do not cause a stack overflow, where:

• C = maximum call stack depth,
• F = maximum statement depth for functions,
• S = maximum statement depth at global level.

A script exceeding the maximum nesting depths will terminate with a parsing error. The malignant AST will not be able to get past parsing in the first place.

The limits can be disabled via the unchecked feature for higher performance (but higher risks as well).

Checked arithmetic

By default, all arithmetic calculations in Rhai are checked, meaning that the script terminates with an error whenever it detects a numeric over-flow/under-flow condition or an invalid floating-point operation, instead of crashing the entire system. This checking can be turned off via the unchecked feature for higher performance (but higher risks as well).

Blocking access to external data

Rhai is sand-boxed so a script can never read from outside its own environment. Furthermore, an Engine created non-mut cannot mutate any state outside of itself; so it is highly recommended that Engine's are created immutable as much as possible.

let mut engine = Engine::new();             // create mutable 'Engine'

engine.register_get("add", add);            // configure 'engine'

let engine = engine;                        // shadow the variable so that 'engine' is now immutable

Script optimization

Rhai includes an optimizer that tries to optimize a script after parsing. This can reduce resource utilization and increase execution speed. Script optimization can be turned off via the no_optimize feature.

For example, in the following:

{
    let x = 999;            // NOT eliminated: Rhai doesn't check yet whether a variable is used later on
    123;                    // eliminated: no effect
    "hello";                // eliminated: no effect
    [1, 2, x, x*2, 5];      // eliminated: no effect
    foo(42);                // NOT eliminated: the function 'foo' may have side effects
    666                     // NOT eliminated: this is the return value of the block,
                            // and the block is the last one so this is the return value of the whole script
}

Rhai attempts to eliminate dead code (i.e. code that does nothing, for example an expression by itself as a statement, which is allowed in Rhai). The above script optimizes to:

{
    let x = 999;
    foo(42);
    666
}

Constants propagation is used to remove dead code:

const ABC = true;
if ABC || some_work() { print("done!"); }   // 'ABC' is constant so it is replaced by 'true'...
if true || some_work() { print("done!"); }  // since '||' short-circuits, 'some_work' is never called
if true { print("done!"); }                 // <- the line above is equivalent to this
print("done!");                             // <- the line above is further simplified to this
                                            //    because the condition is always true

These are quite effective for template-based machine-generated scripts where certain constant values are spliced into the script text in order to turn on/off certain sections. For fixed script texts, the constant values can be provided in a user-defined Scope object to the Engine for use in compilation and evaluation.

Beware, however, that most operators are actually function calls, and those functions can be overridden, so they are not optimized away:

const DECISION = 1;

if DECISION == 1 {          // NOT optimized away because you can define
    :                       // your own '==' function to override the built-in default!
    :
} else if DECISION == 2 {   // same here, NOT optimized away
    :
} else if DECISION == 3 {   // same here, NOT optimized away
    :
} else {
    :
}

because no operator functions will be run (in order not to trigger side effects) during the optimization process (unless the optimization level is set to OptimizationLevel::Full). So, instead, do this:

const DECISION_1 = true;
const DECISION_2 = false;
const DECISION_3 = false;

if DECISION_1 {
    :                       // this branch is kept and promoted to the parent level
} else if DECISION_2 {
    :                       // this branch is eliminated
} else if DECISION_3 {
    :                       // this branch is eliminated
} else {
    :                       // this branch is eliminated
}

In general, boolean constants are most effective for the optimizer to automatically prune large if-else branches because they do not depend on operators.

Alternatively, turn the optimizer to OptimizationLevel::Full

Here be dragons!

Optimization levels

There are actually three levels of optimizations: None, Simple and Full.

• None is obvious - no optimization on the AST is performed.

• Simple (default) performs relatively safe optimizations without causing side effects (i.e. it only relies on static analysis and will not actually perform any function calls).

• Full is much more aggressive, including running functions on constant arguments to determine their result. One benefit to this is that many more optimization opportunities arise, especially with regards to comparison operators.

An Engine's optimization level is set via a call to Engine::set_optimization_level:

// Turn on aggressive optimizations
engine.set_optimization_level(rhai::OptimizationLevel::Full);

If it is ever needed to re-optimize an AST, use the optimize_ast method:

// Compile script to AST
let ast = engine.compile("40 + 2")?;

// Create a new 'Scope' - put constants in it to aid optimization if using 'OptimizationLevel::Full'
let scope = Scope::new();

// Re-optimize the AST
let ast = engine.optimize_ast(&scope, &ast, OptimizationLevel::Full);

When the optimization level is OptimizationLevel::Full, the Engine assumes all functions to be pure and will eagerly evaluated all function calls with constant arguments, using the result to replace the call. This also applies to all operators (which are implemented as functions). For instance, the same example above:

// When compiling the following with OptimizationLevel::Full...

const DECISION = 1;
                            // this condition is now eliminated because 'DECISION == 1'
if DECISION == 1 {          // is a function call to the '==' function, and it returns 'true'
    print("hello!");        // this block is promoted to the parent level
} else {
    print("boo!");          // this block is eliminated because it is never reached
}

print("hello!");            // <- the above is equivalent to this
                            //    ('print' and 'debug' are handled specially)

Because of the eager evaluation of functions, many constant expressions will be evaluated and replaced by the result. This does not happen with OptimizationLevel::Simple which doesn't assume all functions to be pure.

// When compiling the following with OptimizationLevel::Full...

let x = (1+2)*3-4/5%6;      // <- will be replaced by 'let x = 9'
let y = (1>2) || (3<=4);    // <- will be replaced by 'let y = true'

Function side effect considerations

All of Rhai's built-in functions (and operators which are implemented as functions) are pure (i.e. they do not mutate state nor cause side any effects, with the exception of print and debug which are handled specially) so using OptimizationLevel::Full is usually quite safe unless you register your own types and functions.

If custom functions are registered, they may be called (or maybe not, if the calls happen to lie within a pruned code block). If custom functions are registered to replace built-in operators, they will also be called when the operators are used (in an if statement, for example) and cause side-effects.

Function volatility considerations

Even if a custom function does not mutate state nor cause side effects, it may still be volatile, i.e. it depends on the external environment and is not pure. A perfect example is a function that gets the current time - obviously each run will return a different value! The optimizer, when using OptimizationLevel::Full, assumes that all functions are pure, so when it finds constant arguments it will eagerly execute the function call. This causes the script to behave differently from the intended semantics because essentially the result of the function call will always be the same value.

Therefore, avoid using OptimizationLevel::Full if you intend to register non-pure custom types and/or functions.

Subtle semantic changes

Some optimizations can alter subtle semantics of the script. For example:

if true {                   // condition always true
    123.456;                // eliminated
    hello;                  // eliminated, EVEN THOUGH the variable doesn't exist!
    foo(42)                 // promoted up-level
}

foo(42)                     // <- the above optimizes to this

Nevertheless, if the original script were evaluated instead, it would have been an error - the variable hello doesn't exist, so the script would have been terminated at that point with an error return.

In fact, any errors inside a statement that has been eliminated will silently disappear:

print("start!");
if my_decision { /* do nothing... */ }  // eliminated due to no effect
print("end!");

// The above optimizes to:

print("start!");
print("end!");

In the script above, if my_decision holds anything other than a boolean value, the script should have been terminated due to a type error. However, after optimization, the entire if statement is removed (because an access to my_decision produces no side effects), thus the script silently runs to completion without errors.

Turning off optimizations

It is usually a bad idea to depend on a script failing or such kind of subtleties, but if it turns out to be necessary (why? I would never guess), turn it off by setting the optimization level to OptimizationLevel::None.

let engine = rhai::Engine::new();

// Turn off the optimizer
engine.set_optimization_level(rhai::OptimizationLevel::None);

eval - or "How to Shoot Yourself in the Foot even Easier"

Saving the best for last: in addition to script optimizations, there is the ever-dreaded... eval function!

let x = 10;

fn foo(x) { x += 12; x }

let script = "let y = x;";  // build a script
script +=    "y += foo(y);";
script +=    "x + y";

let result = eval(script);  // <- look, JS, we can also do this!

print("Answer: " + result); // prints 42

print("x = " + x);          // prints 10: functions call arguments are passed by value
print("y = " + y);          // prints 32: variables defined in 'eval' persist!

eval("{ let z = y }");      // to keep a variable local, use a statement block

print("z = " + z);          // <- error: variable 'z' not found

"print(42)".eval();         // <- nope... method-call style doesn't work

Script segments passed to eval execute inside the current Scope, so they can access and modify everything, including all variables that are visible at that position in code! It is almost as if the script segments were physically pasted in at the position of the eval call. But because of this, new functions cannot be defined within an eval call, since functions can only be defined at the global level, not inside a function call!

let script = "x += 32";
let x = 10;
eval(script);               // variable 'x' in the current scope is visible!
print(x);                   // prints 42

// The above is equivalent to:
let script = "x += 32";
let x = 10;
x += 32;
print(x);

For those who subscribe to the (very sensible) motto of "eval is evil", disable eval by overriding it, probably with something that throws.

fn eval(script) { throw "eval is evil! I refuse to run " + script }

let x = eval("40 + 2");     // 'eval' here throws "eval is evil! I refuse to run 40 + 2"

Or override it from Rust:

fn alt_eval(script: String) -> Result<(), Box<EvalAltResult>> {
    Err(format!("eval is evil! I refuse to run {}", script).into())
}

engine.register_result_fn("eval", alt_eval);