rust_icu: low-level rust language bindings for the ICU library

Item	Description
Testing
Source	https://github.com/google/rust_icu
README	https://github.com/google/rust_icu/blob/main/README.md
Coverage	View report
Docs	https://docs.rs/crate/rust_icu

This is a library of low level native rust language bindings for the International Components for Unicode (ICU) library for C (a.k.a. ICU4C).

If you just want quick instructions on how to download and install, see the quickstart guide

See the ICU project home page for details about the ICU library. The library source can be viewed on Github.

The latest version of this file is available at https://github.com/google/rust_icu.

This is not an officially supported Google product.

Why wrap ICU (vs. doing anything else)?

• The rust language Internationalisation page confirms that ICU support in rust is spotty, so having a functional wrapper helps advance the state of the art.

• Projects such as Fuchsia already depend on ICU, and having rust bindings allows for an easy way to use Unicode algorithms without taking on more dependencies.

• Cooperation on the interface with projects such as the ICU4X could allow seamless transition to an all-rust implementation in the future.

Structure of the repository

The repository is organized as a cargo workspace of rust crates. Each crate corresponds to the respective header in the ICU4C library's C API. Please consult the coverage report for details about function coverage in the headers.

Limitations

The generated rust language binding methods of today limit the availability of language bindings to the available C API. The ICU library's C API (sometimes referred to as ICU4C in the documentation) is distinct from the ICU C++ API.

The bindings offered by this library have somewhat limited applicability, which means it may sometimes not work for you out of the box. If you come across such a case, feel free to file a bug for us to fix. Pull requests are welcome.

The limitations we know of today are as follows:

• There isn't a guaranted feature parity. Some algorithms that are implemented in C++ don't have a C equivalent, and vice-versa. This is usually not a problem if you are using the library from C++, since you are free to choose whichever API surface works for you. But it is an issue for rust bindings, since we can only use the C API at the moment.

• A C++ implementation of a new algorithm is not necessarily always reflected in the C API, leading to feature disparity between the C and C++ API surfaces. See for example this bug as an illustration.

• While using icu_config feature will likely allow you some freedom to auto-generate bindings for your own library version, we still need to keep a list of explicitly supported ICU versions to ensure that the wrappers are stable.

Compatibility

The compatibility guarantee is as follows:

1. Automated tests are executed for last three major ICU library versions in all feature combinations of interest.
2. Automated tests are executed for the ICU library version in use by the docs.rs system (so the documentation could be built).

Features

The rust_icu library is intended to be compiled with cargo, with one of several features enabled. Compilation with cargo allows us to do some library detection in a custom build.rs file in the rust_icu_sys library and adapt the build process to your build environment. However, since not every development environment will use the same settings, we opted to offer certain features (below) as configuration options.

While our intention is to keep the list of features below up to date with the actual list in Cargo.toml, the list may periodically go out of date.

To use any of the features, you will need to activate the feature in all the rust_icu_* crates that you intend to use. Failing to do this will result in confusing compilation end result.

Prerequisites

Required

• rust_icu source code

  Clone with git:

  git clone https://github.com/google/rust_icu.git
  

• rustup

  Install from https://rustup.rs. Used to set toolchain defaults. This will install cargo as well.

• Clang

  You must have Clang installed to access the right headers.

• The ICU library development environmnet

  You will need access to the ICU libraries for the rust_icu bindings to link against. Download and installation of ICU is out of scope of this document. Please read through the ICU introduction to learn how to build and install.

  Sometimes, the ICU library will be preinstalled on your system, or you can pull the library in from your package management program. However, this library won't necessarily be the one that you need to link into the program you are developing. In short, it is your responsibility to have a developer version of ICU handy somewhere on your system.

  We have a quickstart install that may get you well on the way in case your environment happens to be configured very similarly to ours and you want to build ICU from source.

Optional

• GNU Make, if you want to use the make-based build and test.

  Installing GNU Make is beyond the scope of this file. Please refer to your OS instructions for installation.

• docker, if you decide to use docker-based build and test.

  Installing docker is beyond the scope of this file, please see the docker installation instructions for details. As installing docker is intrusive to the host machine, your company may have internal documentation on how to install docker properly.

• icu-config utility, if icu_config feature is used.

  You need to install the ICU library on your system, such that the binary icu-config is somewhere in your $PATH. The build script will use it to discover the library settings and generate correct link scripts. If you use the feature but icu-config is not found,

• bindgen utility, if bindgen feature is used.

  bindgen user guide for instructions on how to install it.

• rustfmt utility, if bindgen feature is used.

  See https://github.com/rust-lang/rustfmt for instructions on how to install.

Testing

There are a few options to run the test for rust_icu.

Cargo

Building and testing using cargo is the canonical way of building and testing rust code.

In the case of the rust_icu library you may find that your system's default ICU development package is ancient, in which case you will need to build your own ICU4C library (see below for that). That will make it necessary to pass in PKG_CONFIG_PATH and LD_LIBRARY_PATH environment variables to help the bulid code locate and use the library you built, instead of the system default.

The following tests should all build and pass. Note that because the libraries needed are in a custom location, we need to set LD_LIBRARY_PATH when running the tests, as well as PKG_CONFIG_PATH.

If you find that you are able to use your system's default ICU installation, you can safely omit the two libraries.

env PKG_CONFIG_PATH="$HOME/local/lib/pkgconfig" \
    LD_LIBRARY_PATH="$HOME/local/lib" \
        bash -c 'cargo test'

If you think that the above approach is too much of a hassle, consider trying out the Docker-based approach.

GNU Make

If you happen to like the GNU way of doing things, you may appreciate the GNU Make approach.

The easiest way is to use GNU Make and run:

make test

You may want to use this method if you are working on rust_icu, have your development environment all set up and would like a shorthand to run the tests.

Docker-based

See optional dependencies section above.

To run a hermetic build and test of the rust_icu source code, issue the following command:

make docker-test

This will run docker-based build and test of the source code on your local machine. This is a good way to test that your code works with a specific reference version of ICU.

Prior art

There is plenty of prior art that has been considered:

• https://github.com/servo/rust-icu
• https://github.com/open-i18n/rust-unic
• https://github.com/fullcontact/icu-sys
• https://github.com/rust-locale
• https://github.com/unicode-rs

The current state of things is that I'd like to do a few experiments on my own first, then see if the work can be folded into any of the above efforts.

See also:

• https://github.com/rust-lang/rfcs/issues/797
• https://unicode-rs.github.io
• https://github.com/i18n-concept/rust-discuss

Assumptions

There are a few competing approaches for ICU bindings. However, it seems, at least based on information available in rust's RFC repos, that the work on ICU support in rust is still ongoing.

These are the assumptions made in the making of this library:

• We need a complete, reusable and painless ICU low-level library for rust.

  This, for example, means that we must rely on an external ICU library, and not lug the library itself with the binding code. Such modularity allows the end user of the library to use an ICU library of their choice, and incorporate it in their respective systems.

• No ICU algorithms will be reimplemented as part of the work on this library.

  An ICU reimplementation will likely take thousands of engineer years to complete. For an API that is as subtle and complex as ICU, I think that it is probably a better return on investment to maintain a single central implementation.

  Also, the existence of this library doesn't prevent reimplementation. If someone else wants to try their hand at reimplementing ICU, that's fine too.

• This library should serve as a low-level basis for a rust implementation.

  A low level ICU API may not be an appropriate seam for the end users. A rust-ful API should be layered on top of these bindings. It will probably be a good idea to subdivide that functionality into crates, to match the expectations of rust developers.

  I'll gladly reuse the logical subdivision already made in some of the above mentioned projects.

• I'd like to explore ways to combine with existing implementations to build a complete ICU support for rust.

  Hopefully it will be possible to combine the good parts of all the rust bindings available today into a unified rust library. I am always available to discuss options.

  The only reason I started a separate effort instead of contributing to any of the projects listed in the "Prior Art" section is that I wanted to try what a generated library would look like in rust.

Additional instructions

Quickstart guide

Before you begin, please ensure the following prerequisites are met:

• You have docker installed and it runs on your system.
• You have GNU Make.
• You have git.
• You have plenty of disk space. The docker images for the build environment are a bit large, so a few GiB are needed to fit all of them.
• You have an Internet connection.

From there, the following sequence of commands will check out, build and test the rust_icu source code.

mkdir -p ~/tmp
cd tmp
git clone https://github.com/google/rust_icu
cd rust_icu
make docker-test

You can now make changes to the code and tests. You can re-run the compile and test cycle by running make docker-test.

ICU installation instructions

These instructions follow the "out-of-tree" build instructions from the ICU repository.

Assumptions

The instructions below are not self-contained. They assume that:

• you have your system set up such that you can follow the ICU build instructions effectively. This requires some upfront time investment.
• you can build ICU from source, and your project has access to ICU source.
• your setup is Linux, with some very specific settings that worked for me. You may be able to adapt them to work on yours.

Compilation

mkdir -p $HOME/local
mkdir -p $HOME/tmp
cd $HOME/tmp
git clone https://github.com/unicode-org/icu.git
mkdir icu4c-build
cd icu4c-build
../icu/icu4c/source/runConfigureICU Linux \
  --prefix=$HOME/local \
  --enable-static
make
make install
make doc

If the compilation finishes with success, the directory $HOME/local/bin will have the file icu-config which is necessary to discover the library configuration.

You can also do a

make check

to run the unit tests.

If you add $HOME/local/bin to $PATH, or move icu-config to a directory that is listed in your $PATH you should be all set to compile rust_icu.

ICU rebuilding instructions

If you change the configuration of the ICU library with an intention to rebuild the library from source you should probably add an intervening make clean command.

Since the ICU build is not hermetic, this ensures there are no remnants of the old compilation process sitting around in the build directory. You need to do this for example if you upgrade the major version of the ICU library. If you forget to do so, you may see unexpected errors while compiling ICU, or while linking or running your programs.

Compiling for a set version of ICU

Assumptions

• You have selected the feature set [renaming,icu_version_in_env]o

OR:

• You have manually verified that the compatibility matrix has a "Yes" for the ICU version and feature set you want to use.

The following is a tested example.

env PKG_CONFIG_PATH="$HOME/local/lib/pkgconfig" \
    LD_LIBRARY_PATH="$HOME/local/lib" \
    RUST_ICU_MAJOR_VERSION_NUMBER=65 \
        bash -c 'cargo test'

The following would be an as of yet untested example of compiling rust_icu against a preexisting ICU version 66.

env PKG_CONFIG_PATH="$HOME/local/lib/pkgconfig" \
    LD_LIBRARY_PATH="$HOME/local/lib" \
    RUST_ICU_MAJOR_VERSION_NUMBER=66 \
        bash -c 'cargo test'

Adding support for a new version of ICU.

In general, as long as icu-config approach is supported, it should be possible to generate the library wrappers for newer versions of the ICU library, assuming that the underlying C APIs do not diverge too much.

An approach that yielded easy support for ICU 65.1 consisted of the following steps. Below, $RUST_ICU_SOURCE_DIR is the directory where you extracted the ICU source code.

• Download the new ICU version from source to $RUST_ICU_SOURCE_DIR.
• Build the ICU library following for example the compilation steps above with the new version.
• Get the file lib.rs from the output directory $RUST_ICU_SOURCE_DIR/target/debug/build/rust_icu_sys-..., rename it to lib_66.rs (if working with ICU version 66, otherwise append the version you are using).
• Save the file to the directory $RUST_ICU_SOURCE_DIR/rust_icu_sys/bindgen, this is the directory that contains the pre-generated sources.

These files lib_XX.rs may need to be generated again if build.rs is changed to include more features.

Adding more bindings

When adding more ICU wrappers, make sure to do the following:

• Check rust_icu_sys/build.rs and rust_icu_sys/bindgen/run_bindgen.sh to add appropriate lines into BINDGEN_SOURCE_MODULES, then BINDGEN_ALLOWLIST_FUNCTIONS and BINDGEN_ALLOWLIST_TYPES.

Testing with a specific feature set turned on

Here's an example of running a docker test on ICU 67, with features icu_version_in_env and renaming turned on instead of the default. Note that the parameters are mostly passed into the container that runs docker-test via environment variables.

make DOCKER_TEST_ENV=rust_icu_testenv-67 \
  RUST_ICU_MAJOR_VERSION_NUMBER=67 \
  DOCKER_TEST_CARGO_TEST_ARGS='--no-default-features --features icu_version_in_env,renaming' \
  docker-test

Some clarification:

• The environment variable RUST_ICU_MAJOR_VERSION_NUMBER is used for the feature icu_version_in_env to instruct cargo to use the file rust_icu_sys/bindgen/lib_67.rs as a prebuilt bindgen source file instead of trying to generate one on the fly.
• The environment variable DOCKER_TEST_CARGO_TEST_ARGS is used to pass the command line arguments to the cargo test which is used in the docker container. The environment is passed in verbatim to cargo test without quoting, so separate words in the environment end up being separate args to cargo test.
• The environment variable DOCKER_TEST_ENV is the base name of the Docker container used to run the test in. The container rust_icu_testenv-67 is a container image that contains preinstalled environment with a compiled version of ICU 67.

Refreshing static bindgen files

Requires docker.

Run make static-bindgen periodically, to refresh the statically generated bindgen files (named lib_XX.rs, where XX is an ICU version, e.g. 67) in the directory rust_icu_sys/bindgen which are used when bindgen features are turned off.

Invoking this make target will modify the local checkout with the newer versions of the files lib_XX.rs. Make a pull request and check them in.

For more information on why this is needed, see the bindgen README.md.