ICU4X's core functionality is completely available from C++, with headers generated by Diplomat. The port is header-only; no additional C++ translation units need to be compiled to use ICU4X from C++.

The docs can be found here.

Typically C++ users can build ICU4X by building the icu_capi Rust crate, and linking the resultant static library to their C++ application. This crate contains all of the relevant Diplomat-generated extern "C" declarations, as well as an idiomatic C++ wrapper using these functions.

Using ICU4X in C++ is best demonstrated via these examples. For example, here's an example showing off decimal formatting in ICU4X, built with [this Makefile](Makefile).

We are still working on improving the user experience of using ICU4X from other languages. As such, this tutorial may be a bit sparse, but we are happy to answer questions on our discussions forum and help you out

Building ICU4X

After installing Rust, create a local Rust configuration for ICU4X:

touch Cargo.toml

[package]
name = "unused"
 
[lib]
path = "unused"
 
[dependencies]
icu_capi = { version = "2.0.0", default-features = false, features = [] }

Some of the keys are required by the parser, but won't be used by us.

icu_capi supports a list of optional features:

default enables a default set of features
std [default] set this when building for a target with a Rust standard library, otherwise see below
compiled_data [default] to include data and enable DataProvider-less constructors
simple_logger [default] enable basic stdout logging of error metadata. Further loggers can be added on request.
default_components [default] activate all stable ICU4X components. For smaller builds, this can be disabled, and components can be added with features like list.
buffer_provider for working with blob data providers (DataProvider::create_from_byte_slice())

You can now set features by updating the features key in Cargo.toml:

icu_capi = { version = "2.0.0", default-features = false, features = ["default", "buffer_provider"] }

You can now build a staticlib with the following command:

cargo rustc --release -p icu_capi --crate-type staticlib

Be sure to pass --release to get an optimized build
Set CARGO_PROFILE_RELEASE_LTO=true to enable link-time optimization
Set CARGO_PROFILE_RELEASE_OPT_LEVEL="s" to optimize for size
See [cargo profiles] for more options

You should now have a target/release/libicu_capi.a, ready to compile into your C++ binary.

Using ICU4X from C++

Here's an annotated, shorter version of the decimal example:

#include <icu4x/DecimalFormatter.hpp>
#include <icu4x/Logger.hpp>
 
#include <iostream>
#include <array>
 
int main() {
    // For basic logging
    Logger::init_simple_logger();
 
    // Create a locale object representing Bangla
    std::unique_ptr<Locale> locale = Locale::from_string("bn").ok().value();
 
    std::cout << "Running test for locale " << locale->to_string() << std::endl;
 
    // Create a formatter object with the appropriate settings
    std::unique_ptr<DecimalFormatter> formatter = DecimalFormatter::create_with_grouping_strategy(
        *locale.get(), DecimalGroupingStrategy::Auto).ok().value();
 
    // Create a decimal representing the number 1,000,007
    std::unique_ptr<Decimal> decimal = Decimal::from(1000007);
    
    // Format it to a string
    std::string out = formatter->format(*decimal.get());
    
    // Report formatted value
    std::cout << "Formatted value is " << out << std::endl;
    if (out != "১০,০০,০০৭") {
        std::cout << "Output does not match expected output" << std::endl;
        return 1;
    }
 
    return 0;
}

Compiling with the ICU4X header files

The header files are shipped inside the crate, which Cargo has put somewhere on your system. You can find its location with

HEADERS=$(cargo metadata --format-version 1 | jq '.packages[] | select(.name == "icu_capi").manifest_path' | xargs dirname)/bindings/cpp

Then you can build with

g++ -Ltarget/release -I"$HEADERS" main.cpp -licu_capi

C++ versions beyond C++17 are supported, as are other C++ compilers.

Embedded platforms (`no_std`)

Users wishing to use ICU4X on a no_std platform will need to provide an allocator and a panic hook in order to build a linkable library. The icu_capi crate can provide a looping panic handler, and a malloc-backed allocator, under the looping_panic_handler and libc_alloc features, respectively.

icu_capi = { version = "2.0.0", default-features = false, features = ["default_components", "buffer_provider", "looping_panic_handler", "libc_alloc"] }

This can be built the same way, with an explicitly specified --target (in this case, thumbv7em-none-eabi, but it can be any no_std target)

cargo rustc --release -p icu_capi --crate-type staticlib --target thumbv7em-none-eabi

Tips

Fallible methods return diplomat::result, a Result type that can most commonly be converted to a std::optional over its Ok/Err types by calling .ok() or .err(). Most methods either use ICU4XError (an enum of error codes) as their error type, or std::monostate. Further error details can be logged by enabling a logger via ICU4XLogger, further loggers may be added on request.

The C++ headers include C headers for the underlying APIs as well, under namespace capi, found in the .h files. While these can be used directly, we recommend against it unless you are writing C code. These headers are not intended to be ergonomic and primarily exist for the C++ headers to use internally.

Slices are represented using std::span if available, otherwise a simple wrapper called diplomat::span is used.

These bindings may be customized by running diplomat-tool directly (including replacing the types used with alternate types like mozilla::Span), please ask on our discussions forum for more help on this.

Building ICU4X

Using ICU4X from C++

Compiling with the ICU4X header files

Embedded platforms (no_std)

Tips

Embedded platforms (`no_std`)