10 releases

0.0.1-alpha.10 May 30, 2023
0.0.1-alpha.9 Apr 20, 2023
0.0.1-alpha.4 Mar 27, 2023

#33 in Internationalization (i18n)

Download history 1/week @ 2023-06-10 41/week @ 2023-06-17 27/week @ 2023-06-24 23/week @ 2023-07-01 13/week @ 2023-07-08 17/week @ 2023-07-15 26/week @ 2023-07-22 9/week @ 2023-07-29 37/week @ 2023-08-05 20/week @ 2023-08-12 11/week @ 2023-08-19 8/week @ 2023-08-26 33/week @ 2023-09-02 21/week @ 2023-09-09 12/week @ 2023-09-16 11/week @ 2023-09-23

77 downloads per month
Used in 3 crates






Apache-2 licensed


Glossa is a language localisation library.


It can be divided into two categories.

  • Const map: Load localisation data efficiently through constant data.
    • Description: Convert configuration files into constant (const fn) code at compile time and read constant data at runtime.
    • Pros: Efficient
    • Cons:
      • Requires codegen, which may result in some redundant code after expansion.
      • Currently only supports simple key-value (K-V) pairs.
  • fluent
    • Description: Manage Fluent resources at runtime.
    • Pros: Fluent syntax may be more suitable for localisation.
    • Cons: Requires more resources than const map.

Note: Fluent also supports loading localisation resources (localised files) at compile time, but data needs to be parsed at runtime.
The former is just the simple K-V pair that uses some const maps from phf to store data. Because it's simple, it's efficient.

The two types of functionalities are independent of each other. For the latter, please read Fluent.md.



Use a code generator to generate code.


glossa-codegen has the following features:

  • yaml
    • Enabled by default.
    • The default file extension is "yaml" or "yml"
  • ron
    • The default ext is "ron"
  • toml
    • The ext is "toml"
  • json
    • ext: "json"
  • highlight

In addition to highlight, this corresponds to different formats of configuration. You can enable all features or add them as needed.

File & Map Name

By default, the file format is determined based on the file name extension, and the map name (table name) is set based on the file name. Whether deserialisation is needed at compile-time is determined by the enabled feature.

yaml map name

Assuming there are two files under the directory assets/l10n/en, named test.yaml and test.yml, then we can consider them to have the same name.

To avoid conflicts, their map names are:

  • test
  • test.yml

If we have these files:

  • test.yaml
  • test.json
  • test.yml
  • test.ron
  • test.toml

Sorting them would result in:

  • test.json
  • test.ron
  • test.toml
  • test.yaml
  • test.yml

Only the map named test is in test.json, whereas the rest of the maps are named after their respective file names.

When using .get() with MapLoader, you need to pass in the map-name


Before writing build.rs, we need to prepare the localisation resource files.

de (Deutsch, Lateinisch, Deutschland)

  • assets/l10n/de/error.yaml
text-not-found: Kein lokalisierter Text gefunden

en (English, Latin, United States)

  • assets/l10n/en/error.yaml
text-not-found: No localized text found

en-GB (English, Latin, Great Britain)

  • assets/l10n/en-GB/error.yaml
text-not-found: No localised text found

es (español, latino, España)

  • assets/l10n/es/error.yaml
text-not-found: No se encontró texto localizado

pt (português, latim, Brasil)

  • assets/l10n/pt/error.yaml

Note: "pt" refers to "Portuguese (Brazil)", not "Portuguese (Portugal)" (português, Portugal).

text-not-found: Nenhum texto localizado encontrado

build script

First, add the dependency:

cargo add --build glossa-codegen

Then start creating build.rs.

This file is at the same level as Cargo.toml.

Project structure and file location

For a simple single-project structure:

build_rs in the single project structure.svg

A slightly more complex multi-project structure:

build_rs in the multi projects.svg

You can also specify the path to build.rs manually, instead of using the default.



use glossa_codegen::{consts::*, prelude::*};
use std::{io, path::PathBuf};

fn main() -> io::Result<()> {
    // Specify the version as the current package version to avoid repetitive compilation for the same version.
    let ver = get_pkg_version!();

    // This is a constant array: ["src", "assets", "localisation.rs"], which is converted into a path for storing automatically generated Rust code related to localisation.
    // path: "src/assets/localisation.rs".
    let rs_path = PathBuf::from_iter(default_l10n_rs_file_arr());

    // If it's the same version, then exit.
    if is_same_version(&rs_path, Some(ver))? {
        // When developing, we can comment out the `return` statement below so that every change will be recompiled and won't exit prematurely.
        return Ok(());

    // If the path is "src/assets/localisation.rs", then it will append `mod localisation;` and related `use` statements to "src/assets/mod.rs".

    // A new file will be created here:
    //    - Linux(non android): "/dev/shm/localisation.tmp"
    //    - Other:"src/assets/localisation.tmp"
    // After the code generation is complete, rename(move) the file: "/dev/shm/localisation.tmp" -> "src/assets/localisation.rs".
    // Note: If written directly to `rs_path` and `cargo` is interrupted during building, it may result in incomplete generated code. Therefore, `tmp_path` is used as a temporary buffer file.
    let tmp_path = get_shmem_path(&rs_path)?;
    let writer = MapWriter::new(&tmp_path, &rs_path);

    // default_l10n_dir_arr() is also a constant array: ["assets", "l10n"].
    // If the current localisation resource path is at the parent level, then you can use `path = PathBuf::from_iter([".."].into_iter().chain(default_l10n_dir_arr()));`.
    let l10n_path = PathBuf::from_iter(default_l10n_dir_arr());

    let generator = Generator::new(l10n_path).with_version(ver);
    // Invoke the generator here to generate code.


We created a writer above.

Now let's modify the code and change writer to mut writer so that it can be modified.

// Whether to automatically generate documentation, defaults to true
*writer.get_gen_doc_mut() = false;
// Modify the visibility of the automatically generated function, defaults to `pub(crate)`
*writer.get_visibility_mut() = "pub(super)";

Additional Information

The above content is the most basic usage, but there are actually more advanced usages.

Starting from 0.0.1-alpha.4, it is possible to compile localised text into string with syntax highlighting at compile time.

Unlike caching/parsing regexes at runtime, the constant string does not require expensive runtime parsing.

Here is a screenshot of the help information for a CLI tool currently in development, which uses the advanced features of glossa-codegen.

Please do not mind what language is being used inside. As a project that supports l10n, it supports various languages. This is just a demonstration of its localisation and syntax highlighting effects.


L10n + constant syntax highlighting = 😍

Don't worry, let's take it slowly. We'll introduce these features after we finish the beginner's tutorial.

By the way, it may not be as perfect as you imagine.

If we choose a Monokai theme before compilation, it will generate highlighted text with the Monokai theme.

If we need themes like One Dark and ayu-dark, we can either generate them at runtime or generate highlighted text for each theme during compilation.

The latter is a trade-off between space (binary file size) and time.


After running cargo b, the code will be generated automatically. If your l10n rs file is src/assets/localisation.rs, you will also need to manually add pub(crate) mod assets; to lib.rs or main.rs (depending on the type of crate you have).

Get Text

Now that the code has been generated, let's write a function to test it!

But before that, we need to add some dependencies.

cargo add phf glossa

The test function is as follows:

    fn new_loader() {
        use crate::assets::localisation::locale_hashmap;
        use glossa::{fallback::FallbackChain, GetText, MapLoader};

        let loader = MapLoader::new(locale_hashmap());
        // Here, for simplicity, `get_or_default()` is used.
        // Actually, the usage of `.get()` is the same, but it returns Result<&str>, not Cow<str>.
        let msg = loader.get_or_default("error", "text-not-found");
        assert_eq!(msg, "No localized text found");

If your system language is "en", the test should pass.

Note that locale_hashmap() is not a const fn but a regular function. However, this does not mean that it is particularly expensive.

The time complexity of HashMap query operation is O(1).

Its value points to a sub-map, and all sub-maps and their sub-maps are consts.

In addition, if the ahash feature is enabled, the RandomState of ahash will be used by default instead of the std::collections.

You can also use OnceCell to create global static data, creating data only once.

pub(crate) fn locales() -> &'static MapLoader {
    static RES: OnceCell<MapLoader> = OnceCell::new();
    RES.get_or_init(|| MapLoader::new(locale_hashmap()))

Wait a minute, don't waste time on these things, our previous test failed.

All right, let's revisit what we did before.
We have previously created localisation resource files for German, Spanish and Portuguese.

Firstly, it will automatically detect the system language. If the localisation resource does not exist, it will automatically use a fallback chain. If the localisation resource exists and your system language is not English, then the above test will fail.

Let's continue to test:

let loader = locales();
let msg = loader.get("error", "text-not-found")?;

Assuming your language is German (de-Latn-DE)

assert_eq!(msg, "Kein lokalisierter Text gefunden");

Spanish (es-Latn-ES)

assert_eq!(msg, "No se encontró texto localizado");

Portuguese (pt-Latn-BR)

assert_eq!(msg, "Nenhum texto localizado encontrado");

Constant Syntax Highlighting Text

You need to enable the highlight feature:

cargo add --build glossa-codegen --features=highlight

In the build.rs file, you'll need to import the following modules:

use glossa_codegen::{
    highlight::{HighLight, HighLightFmt, HighLightRes},
use std::{
    io::{self, BufWriter},

Quick Start

Let's start with a simple example!

Before creating the generator, please refer to the previous section for the necessary preparations.


let mut generator = Generator::new(path).with_version(ver);

// Use the default syntax highlighting resources.
// The default theme is Monokai Extended, and the default syntax set only contains a few syntaxes.
let res = HighLightRes::default();

let os_str = |s| Cow::from(OsStr::new(s));
// The default format is markdown, and the default map name suffix is `_md`.
let fname_and_fmt = |s| (os_str(s), HighLightFmt::default());

// Specify the file names that need to be highlighted.
let map = HashMap::from_iter([fname_and_fmt("opt.toml"), fname_and_fmt("parser.yaml")]);

*generator.get_highlight_mut() = Some(HighLight::new(res, map));




You can use syntect to load custom theme-sets and syntax-sets.
These resources are essentially Sublime themes and syntaxes. You can use HighLightRes::new() to specify the theme_set, or first get a mutable reference to the theme-set using *res.get_theme_set_mut(), and then modify it.

let mut res = HighLightRes::default();
// *res.get_theme_set_mut() = custom theme set
// *res.get_syntax_set_mut() = custom syntax set // requires 'static lifetime, you can create it using OnceCell

// Custom theme name, e.g. ayu-dark
*res.get_name_mut() = Cow::from("ayu-dark");

// Whether to enable theme background
*res.get_background_mut() = false;


First, let me explain what is meant by the term 'map name'.

about map name

Map refers to a mapping relationship, which is commonly referred to as a 'table'.

As it converts localised text into maps, understanding the map name corresponding to the highlighted text is crucial.


let mut fmt = HighLightFmt::default();

// Here, the syntax name is specified as "md". By default, the syntax set only supports Markdown, TOML, JSON, and YAML etc.
// If you want to support more syntaxes, you need to customise the syntax-set of HighLightRes.
// md corresponds to the filename extension of the Markdown format.
// You can think of the syntax name as corresponding to different file extensions.
*fmt.get_syntax_mut() = Cow::from("md");

// Modifies the suffix of the default map name.
// Assuming a file is named opt.toml, the raw text corresponds to the map name `opt`.
// Since the suffix is `_markdown`, the map name of the generated highlighted text is `opt_markdown`.
// If it is None, there will be no raw text map, only a highlighted text map.
*fmt.get_suffix_mut() = Some(Cow::from("_markdown"));



Syntax highlighting is optional. If syntax highlighting is required, then the theme is mandatory.

Previously, we specified a common theme name in HighLightRes.

If you need to generate highlighted text for more themes, you need to modify extra.

// This tuple contains (map name suffix, (theme name, whether to enable the theme background))
let ayu_light = ("_ayu_light", ("ayu-light", true));
let monokai_bright = ("_Monokai-Bright", ("Monokai Extended Bright", false));

let extra_map = HashMap::from_iter([monokai_bright, ayu_light]);

*fmt.get_extra_mut() = Some(extra_map);

About common theme and extra themes

The common theme name is included in the HighLightRes structure, while extra theme names are included in HighLightFmt.

The reason for emphasizing "name" is that theme names can be stored separately, but theme resources cannot.

If you don't need to control "different files corresponding to different themes" so finely, you don't need to pay attention to this section.

You can choose a common theme + extra themes, just modify *fmt.get_extra_mut() like Extra.

If you don't need a common theme, you must set the theme name of HighLightRes to empty.

*res.get_name_mut() = Cow::from("");

The rest of the work is to modify extra for different formats.

For example:

  • *md_fmt.get_extra_mut() = Some(ext_map1)
  • *rs_fmt.get_extra_mut() = Some(ext_map2)
  • *html_fmt.get_extra_mut() = Some(ext_map3)

Highlight File Map


The following statement was used to create a Highlight File Map.

let fname_and_fmt = |s| (os_str(s), HighLightFmt::default());
let map = HashMap::from_iter([fname_and_fmt("opt.toml"), fname_and_fmt("parser.yaml")]);

If we change it to this, all specified files will use the HighLightFmt specified above:

let fname_and_fmt = |s| (os_str(s), fmt.clone());

Don't forget that we created extra themes for fmt above.

In many cases, we need to make more adjustments instead of using extra themes for all files.

// Assuming you have set a custom syntax-set that includes the syntax you want for LaTex.
let mut tex_fmt = HighLightFmt::default();
*tex_fmt.get_suffix_mut() = Some(Cow::from("_tex"));
*tex_fmt.get_syntax_mut() = Cow::from("latex");
// tex_fmt specifies an extra dracula theme.
*tex_fmt.get_extra_mut() =
    Some(HashMap::from_iter([("_tex_dracula", ("dracula", false))]));

let dracula_latex = |s| (os_str(s), tex_fmt);

let highlight_map = HashMap::from_iter([
    (os_str("parser.ron"), HighLightFmt::default()),

Actually, using LaTex as an example may not be appropriate because it only provides syntax highlighting and cannot render LaTex into SVG.




HighLightFmt_struct.svg extra_theme_map.svg


~61K SLoC