Evalexpr is an expression evaluator and tiny scripting language in Rust. It has a small and easy to use interface and can be easily integrated into any application. It is very lightweight and comes with no further dependencies. Evalexpr is available on crates.io, and its API Documentation is available on docs.rs.

Minimum Supported Rust Version: 1.65.0

Add evalexpr as dependency to your Cargo.toml:

[dependencies]
evalexpr = "<desired version>"

Then you can use evalexpr to evaluate expressions like this:

use evalexpr::*;

assert_eq!(eval("1 + 2 + 3"), Ok(Value::from(6)));
// `eval` returns a variant of the `Value` enum,
// while `eval_[type]` returns the respective type directly.
// Both can be used interchangeably.
assert_eq!(eval_int("1 + 2 + 3"), Ok(6));
assert_eq!(eval("1 /* inline comments are supported */ - 2 * 3 // as are end-of-line comments"), Ok(Value::from(-5)));
assert_eq!(eval("1.0 + 2 * 3"), Ok(Value::from(7.0)));
assert_eq!(eval("true && 4 > 2"), Ok(Value::from(true)));

You can chain expressions and assign to variables like this:

use evalexpr::*;

let mut context = HashMapContext::new();
// Assign 5 to a like this
assert_eq!(eval_empty_with_context_mut("a = 5", &mut context), Ok(EMPTY_VALUE));
// The HashMapContext is type safe, so this will fail now
assert_eq!(eval_empty_with_context_mut("a = 5.0", &mut context),
           Err(EvalexprError::expected_int(Value::from(5.0))));
// We can check which value the context stores for a like this
assert_eq!(context.get_value("a"), Some(&Value::from(5)));
// And use the value in another expression like this
assert_eq!(eval_int_with_context_mut("a = a + 2; a", &mut context), Ok(7));
// It is also possible to save a bit of typing by using an operator-assignment operator
assert_eq!(eval_int_with_context_mut("a += 2; a", &mut context), Ok(9));

And you can use variables and functions in expressions like this:

use evalexpr::*;

let context = context_map! {
    "five" => 5,
    "twelve" => 12,
    "f" => Function::new(|argument| {
        if let Ok(int) = argument.as_int() {
            Ok(Value::Int(int / 2))
        } else if let Ok(float) = argument.as_float() {
            Ok(Value::Float(float / 2.0))
        } else {
            Err(EvalexprError::expected_number(argument.clone()))
        }
    }),
    "avg" => Function::new(|argument| {
        let arguments = argument.as_tuple()?;

        if let (Value::Int(a), Value::Int(b)) = (&arguments[0], &arguments[1]) {
            Ok(Value::Int((a + b) / 2))
        } else {
            Ok(Value::Float((arguments[0].as_number()? + arguments[1].as_number()?) / 2.0))
        }
    })
}.unwrap(); // Do proper error handling here

assert_eq!(eval_with_context("five + 8 > f(twelve)", &context), Ok(Value::from(true)));
// `eval_with_context` returns a variant of the `Value` enum,
// while `eval_[type]_with_context` returns the respective type directly.
// Both can be used interchangeably.
assert_eq!(eval_boolean_with_context("five + 8 > f(twelve)", &context), Ok(true));
assert_eq!(eval_with_context("avg(2, 4) == 3", &context), Ok(Value::from(true)));

You can also precompile expressions like this:

use evalexpr::*;

let precompiled = build_operator_tree("a * b - c > 5").unwrap(); // Do proper error handling here

let mut context = context_map! {
    "a" => 6,
    "b" => 2,
    "c" => 3
}.unwrap(); // Do proper error handling here
assert_eq!(precompiled.eval_with_context(&context), Ok(Value::from(true)));

context.set_value("c".into(), 8.into()).unwrap(); // Do proper error handling here
assert_eq!(precompiled.eval_with_context(&context), Ok(Value::from(false)));
// `Node::eval_with_context` returns a variant of the `Value` enum,
// while `Node::eval_[type]_with_context` returns the respective type directly.
// Both can be used interchangeably.
assert_eq!(precompiled.eval_boolean_with_context(&context), Ok(false));

While primarily meant to be used as a library, evalexpr is also available as a command line tool. It can be installed and used as follows:

cargo install evalexpr
evalexpr 2 + 3 # outputs `5` to stdout.

This crate offers a set of binary and unary operators for building expressions. Operators have a precedence to determine their order of evaluation, where operators of higher precedence are evaluated first. The precedence should resemble that of most common programming languages, especially Rust. Variables and values have a precedence of 200, and function literals have 190.

Supported binary operators:

Supported unary operators:

Operators that take numbers as arguments can either take integers or floating point numbers. If one of the arguments is a floating point number, all others are converted to floating point numbers as well, and the resulting value is a floating point number as well. Otherwise, the result is an integer. An exception to this is the exponentiation operator that always returns a floating point number. Example:

use evalexpr::*;

assert_eq!(eval("1 / 2"), Ok(Value::from(0)));
assert_eq!(eval("1.0 / 2"), Ok(Value::from(0.5)));
assert_eq!(eval("2^2"), Ok(Value::from(4.0)));

The aggregation operator aggregates a set of values into a tuple. A tuple can contain arbitrary values, it is not restricted to a single type. The operator is n-ary, so it supports creating tuples longer than length two. Example:

use evalexpr::*;

assert_eq!(eval("1, \"b\", 3"),
           Ok(Value::from(vec![Value::from(1), Value::from("b"), Value::from(3)])));

To create nested tuples, use parentheses:

use evalexpr::*;

assert_eq!(eval("1, 2, (true, \"b\")"), Ok(Value::from(vec![
    Value::from(1),
    Value::from(2),
    Value::from(vec![
        Value::from(true),
        Value::from("b")
    ])
])));

This crate features the assignment operator, that allows expressions to store their result in a variable in the expression context. If an expression uses the assignment operator, it must be evaluated with a mutable context.

Note that assignments are type safe when using the HashMapContext. That means that if an identifier is assigned a value of a type once, it cannot be assigned a value of another type.

use evalexpr::*;

let mut context = HashMapContext::new();
assert_eq!(eval_with_context("a = 5", &context), Err(EvalexprError::ContextNotMutable));
assert_eq!(eval_empty_with_context_mut("a = 5", &mut context), Ok(EMPTY_VALUE));
assert_eq!(eval_empty_with_context_mut("a = 5.0", &mut context),
           Err(EvalexprError::expected_int(5.0.into())));
assert_eq!(eval_int_with_context("a", &context), Ok(5));
assert_eq!(context.get_value("a"), Some(5.into()).as_ref());

For each binary operator, there exists an equivalent operator-assignment operator. Here are some examples:

use evalexpr::*;

assert_eq!(eval_int("a = 2; a *= 2; a += 2; a"), Ok(6));
assert_eq!(eval_float("a = 2.2; a /= 2.0 / 4 + 1; a"), Ok(2.2 / (2.0 / 4.0 + 1.0)));
assert_eq!(eval_string("a = \"abc\"; a += \"def\"; a"), Ok("abcdef".to_string()));
assert_eq!(eval_boolean("a = true; a &&= false; a"), Ok(false));

The expression chaining operator works as one would expect from programming languages that use the semicolon to end statements, like Rust, C or Java. It has the special feature that it returns the value of the last expression in the expression chain. If the last expression is terminated by a semicolon as well, then Value::Empty is returned. Expression chaining is useful together with assignment to create small scripts.

use evalexpr::*;

let mut context = HashMapContext::new();
assert_eq!(eval("1;2;3;4;"), Ok(Value::Empty));
assert_eq!(eval("1;2;3;4"), Ok(4.into()));

// Initialization of variables via script
assert_eq!(eval_empty_with_context_mut("hp = 1; max_hp = 5; heal_amount = 3;", &mut context),
           Ok(EMPTY_VALUE));
// Precompile healing script
let healing_script = build_operator_tree("hp = min(hp + heal_amount, max_hp); hp").unwrap(); // Do proper error handling here
// Execute precompiled healing script
assert_eq!(healing_script.eval_int_with_context_mut(&mut context), Ok(4));
assert_eq!(healing_script.eval_int_with_context_mut(&mut context), Ok(5));

An expression evaluator that just evaluates expressions would be useful already, but this crate can do more. It allows using variables, assignments, statement chaining and user-defined functions within an expression. When assigning to variables, the assignment is stored in a context. When the variable is read later on, it is read from the context. Contexts can be preserved between multiple calls to eval by creating them yourself. Here is a simple example to show the difference between preserving and not preserving context between evaluations:

use evalexpr::*;

assert_eq!(eval("a = 5;"), Ok(Value::from(())));
// The context is not preserved between eval calls
assert_eq!(eval("a"), Err(EvalexprError::VariableIdentifierNotFound("a".to_string())));

let mut context = HashMapContext::new();
assert_eq!(eval_with_context_mut("a = 5;", &mut context), Ok(Value::from(())));
// Assignments require mutable contexts
assert_eq!(eval_with_context("a = 6", &context), Err(EvalexprError::ContextNotMutable));
// The HashMapContext is type safe
assert_eq!(eval_with_context_mut("a = 5.5", &mut context),
           Err(EvalexprError::ExpectedInt { actual: Value::from(5.5) }));
// Reading a variable does not require a mutable context
assert_eq!(eval_with_context("a", &context), Ok(Value::from(5)));

Note that the assignment is forgotten between the two calls to eval in the first example. In the second part, the assignment is correctly preserved. Note as well that to assign to a variable, the context needs to be passed as a mutable reference. When passed as an immutable reference, an error is returned.

Also, the HashMapContext is type safe. This means that assigning to a again with a different type yields an error. Type unsafe contexts may be implemented if requested. For reading a, it is enough to pass an immutable reference.

Contexts can also be manipulated in code. Take a look at the following example:

use evalexpr::*;

let mut context = HashMapContext::new();
// We can set variables in code like this...
context.set_value("a".into(), 5.into());
// ...and read from them in expressions
assert_eq!(eval_int_with_context("a", &context), Ok(5));
// We can write or overwrite variables in expressions...
assert_eq!(eval_with_context_mut("a = 10; b = 1.0;", &mut context), Ok(().into()));
// ...and read the value in code like this
assert_eq!(context.get_value("a"), Some(&Value::from(10)));
assert_eq!(context.get_value("b"), Some(&Value::from(1.0)));

Contexts are also required for user-defined functions. Those can be passed one by one with the set_function method, but it might be more convenient to use the context_map! macro instead:

use evalexpr::*;

let context = context_map!{
    "f" => Function::new(|args| Ok(Value::from(args.as_int()? + 5))),
}.unwrap_or_else(|error| panic!("Error creating context: {}", error));
assert_eq!(eval_int_with_context("f 5", &context), Ok(10));

For more information about user-defined functions, refer to the respective section.

This crate offers a set of builtin functions (see below for a full list). They can be disabled if needed as follows:

use evalexpr::*;
let mut context = HashMapContext::new();
assert_eq!(eval_with_context("max(1,3)",&context),Ok(Value::from(3)));
context.set_builtin_functions_disabled(true).unwrap(); // Do proper error handling here
assert_eq!(eval_with_context("max(1,3)",&context),Err(EvalexprError::FunctionIdentifierNotFound(String::from("max"))));

Not all contexts support enabling or disabling builtin functions. Specifically the EmptyContext has builtin functions disabled by default, and they cannot be enabled. Symmetrically, the EmptyContextWithBuiltinFunctions has builtin functions enabled by default, and they cannot be disabled.

The min and max functions can deal with a mixture of integer and floating point arguments. If the maximum or minimum is an integer, then an integer is returned. Otherwise, a float is returned.

The regex functions require the feature flag regex_support.

Operators take values as arguments and produce values as results. Values can be booleans, integer or floating point numbers, strings, tuples or the empty type. Values are denoted as displayed in the following table.

Integers are internally represented as i64, and floating point numbers are represented as f64. Tuples are represented as Vec<Value> and empty values are not stored, but represented by Rust's unit type () where necessary.

There exist type aliases for some of the types. They include IntType, FloatType, TupleType and EmptyType.

Values can be constructed either directly or using the From trait. They can be decomposed using the Value::as_[type] methods. The type of a value can be checked using the Value::is_[type] methods.

Examples for constructing a value:

Examples for deconstructing a value:

Values have a precedence of 200.

This crate allows to compile parameterizable formulas by using variables. A variable is a literal in the formula, that does not contain whitespace or can be parsed as value. For working with variables, a context is required. It stores the mappings from variables to their values.

Variables do not have fixed types in the expression itself, but are typed by the context. Once a variable is assigned a value of a specific type, it cannot be assigned a value of another type. This might change in the future and can be changed by using a type-unsafe context (not provided by this crate as of now).

Here are some examples and counter-examples on expressions that are interpreted as variables:

Variables have a precedence of 200.

This crate allows to define arbitrary functions to be used in parsed expressions. A function is defined as a Function instance, wrapping an fn(&Value) -> EvalexprResult<Value>. The definition needs to be included in the Context that is used for evaluation. As of now, functions cannot be defined within the expression, but that might change in the future.

The function gets passed what ever value is directly behind it, be it a tuple or a single values. If there is no value behind a function, it is interpreted as a variable instead. More specifically, a function needs to be followed by either an opening brace (, another literal, or a value. While not including special support for multi-valued functions, they can be realized by requiring a single tuple argument.

Be aware that functions need to verify the types of values that are passed to them. The error module contains some shortcuts for verification, and error types for passing a wrong value type. Also, most numeric functions need to distinguish between being called with integers or floating point numbers, and act accordingly.

Here are some examples and counter-examples on expressions that are interpreted as function calls:

Functions have a precedence of 190.

To use this crate with serde, the serde_support feature flag has to be set. This can be done like this in the Cargo.toml:

[dependencies]
evalexpr = {version = "<desired version>", features = ["serde_support"]}

This crate implements serde::de::Deserialize for its type Node that represents a parsed expression tree. The implementation expects a serde string as input. Example parsing with ron format:

Evalexpr supports C-style inline comments and end-of-line comments. Inline comments are started with a /* and terminated with a */. End-of-line comments are started with a // and terminated with a newline character. For example:

use evalexpr::*;

assert_eq!(
    eval(
        "
        // input
        a = 1;  // assignment
        // output
        2 * a /* first double a */ + 2 // then add 2"
    ),
    Ok(Value::Int(4))
);

extern crate ron;
use evalexpr::*;

let mut context = context_map!{
    "five" => 5
}.unwrap(); // Do proper error handling here

// In ron format, strings are surrounded by "
let serialized_free = "\"five * five\"";
match ron::de::from_str::<Node>(serialized_free) {
    Ok(free) => assert_eq!(free.eval_with_context(&context), Ok(Value::from(25))),
    Err(error) => {
        () // Handle error
    }
}

With serde, expressions can be integrated into arbitrarily complex data.

The crate also implements Serialize and Deserialize for the HashMapContext, but note that only the variables get (de)serialized, not the functions.

This crate is primarily distributed under the terms of the MIT license. See LICENSE for details.

This crate makes extensive use of the Result pattern and is intended to never panic. The exception are panics caused by failed allocations. But unfortunately, Rust does not provide any features to prove this behavior. The developer of this crate has not found a good solution to ensure no-panic behavior in any way. Please report a panic in this crate immediately as issue on github.

Even if the crate itself is panic free, it allows the user to define custom functions that are executed by the crate. The user needs to ensure that the functions they provide to the crate never panic.

This crate was not built with untrusted input in mind, but due to its simplicity and freedom of panics it is likely secure, keeping the following in mind:

• Limit the length of the untrusted input.
• If a mutable context is maintained between evaluations of untrusted input, the untrusted input might fill it gradually until the application runs out of memory.
• If no context is provided, a temporary mutable context is implicitly provided. This is freed after evaluation of every single string, so gradual filling cannot happen.
• If no context or a mutable context is provided, and the regex_support feature is activated, the regex_replace builtin function can be used to build an exponentially sized string.

If you have any ideas for features or see any problems in the code, architecture, interface, algorithmics or documentation, please open an issue on github. If there is already an issue describing what you want to say, please add a thumbs up or whatever emoji you think fits to the issue, so I know which ones I should prioritize.

Notes for contributors:

• This crate uses the sync-readme cargo subcommand to keep the documentation in src/lib.rs and README.md in sync. The subcommand only syncs from the documentation in src/lib.rs to README.md. So please alter the documentation in the src/lib.rs rather than altering anything in between <!-- cargo-sync-readme start --> and <!-- cargo-sync-readme end --> in the README.md.