MessageFormatter for .NET

- better UI strings.

This is an implementation of the ICU Message Format in .NET. For official information about the format, go to: https://unicode-org.github.io/icu/userguide/format_parse/messages/

Quickstart

var mf = new MessageFormatter();

var str = @"You have {notifications, plural,
              zero {no notifications}
               one {one notification}
               =42 {a universal amount of notifications}
             other {# notifications}
            }. Have a nice day, {name}!";
var formatted = mf.FormatMessage(str, new Dictionary<string, object>{
  {"notifications", 4},
  {"name", "Jeff"}
});

//Result: You have 4 notifications. Have a nice day, Jeff!

Or, if you don't like dictionaries, and don't mind a bit of reflection..

var formatted = mf.FormatMessage(str, new {
  notifications = 0,
  name = "Jeff"
});

//Result: You have no notifications. Have a nice day, Jeff!

You can use a static method, too:

var formatted = MessageFormatter.Format(str, new {
  notifications = 1,
  name = "Jeff"
});

//Result: You have one notification. Have a nice day, Jeff!

Installation

Either clone this repo and build it, or install it with NuGet:

Install-Package MessageFormat

Features

• It's fast. Everything is hand-written; no parser-generators, not even regular expressions.
• It's portable. The library is targeting .NET Standard 2.0.
• It's (relatively) small. For a .NET library, ~25kb is not a lot.
• It's very white-space tolerant. You can structure your blocks so they are more readable - look at the example above.
• Nesting is supported. You can nest your blocks as you please, there's no special structure required to do this, just ensure your braces match.
• Adding your own formatters. I don't know why you would need to, but if you want, you can add your own formatters, and take advantage of the code in my base classes to help you parse patterns. Look at the source, this is how I implemented the built-in formatters.
• Exceptions make at least a little sense. When exceptions are thrown due to a bad pattern, the exception should include useful information.
• There are unit tests. Run them yourself if you want, they're using XUnit.
• Built-in cache. If you are formatting messages in a tight loop, with different data for each iteration, and if you are reusing the same instance of MessageFormatter, the formatter will cache the tokens of each pattern (nested, too), so it won't have to spend CPU time to parse out literals every time. I benchmarked it, and on my monster machine, it didn't make much of a difference (10000 iterations).
• Built-in pluralization formatters. Generated from the CLDR pluralization rule data.

Performance

If you look at MessageFormatterCachingTests, you will find a "with cache" and "without cache" test.

My machine runs on a Core i7 3960x, and with about 100,000 iterations with random data (generated beforehand), it takes about 2 seconds (1892ms) with the cache, and about 3 seconds (3236ms) without it. These results are with a debug build, when it is in release mode the time taken is reduced by about 40%! :)

Supported formats

MessageFormat.NET supports the most commonly used formats:

• Select Format: {gender, select, male{He likes} female{She likes} other{They like}} cheeseburgers
• Plural Format: There {msgCount, plural, zero {are no unread messages} one {is 1 unread message} other{are # unread messages}}. (where # is the actual number, with the offset (if any) subtracted).
• Simple variable replacement: Your name is {name}
• Numbers: Your age is {age, number}
• Dates: You were born {birthday, date}
• Time: The time is {now, time}

You can also specify a predefined style, for example {birthday, date, short}. The supported predefined styles are:

• For the number format: integer, currency, percent
• For the date format: short, full
• For the time format: short, medium

These are currently mapped to the built-in .NET format specifiers. This package does not ship with any locale data beyond the pluralizers that are generated based on CLDR data, so if you wish to provide your own localized formatting, read the section below.

Customize formatting

If you wish to control exactly how number, date and time are formatted, you can either:

• Derive CustomValueFormatter and override the format methods
• Instantiate a CustomValueFormatters and assign a lambda to the desired properties Then pass it in as the customValueFormatter parameter to new MessageFormatter.

Example: A custom formatter that allows the use of .NET's formatting tokens. This is for illustration purposes only and is not recommended for use in real apps.

// This is using the lambda-based approach.
var custom = new CustomValueFormatters
{
    // The formatter must set the `formatted` out parameter and return `true`
    // If the formatter returns `false`, the built-in formatting is used.
    Number = (CultureInfo _, object? value, string? style, out string? formatted) =>
    {
        formatted = string.Format($"{{0:{style}}}", value);
        return true;
    }
};

// Create a MessageFormatter with the custom value formatter.
var formatter = new MessageFormatter(locale: "en-US", customValueFormatter: custom);

// Format a message.
var message = formatter.FormatMessage("{value, number, $0.0}", new { value = 23 });
// "$23.0"

Adding your own pluralizer functions

Since MessageFormat 5.0, pluralizers based on the official CLDR data ship with the package, so this is no longer needed.

Same thing as with MessageFormat.js, you can add your own pluralizer function. The Pluralizers property is a IDictionary<string, Pluralizer>, so you can remove the built-in ones if you want.

var mf = new MessageFormatter();
mf.Pluralizers.Add("<locale>", n => {
  // ´n´ is the number being pluralized.
  if(n == 0)
    return "zero";
  if(n == 1)
    return "one";
  return "other";
});

There's no restrictions on what strings you may return, nor what strings you may use in your pluralization block.

var mf = new MessageFormatter(true, "en"); // true = use cache
mf.Pluralizers["en"] = n =>
{
    // ´n´ is the number being pluralized.
    if (n == 0)
        return "zero";
    if (n == 1)
        return "one";
    if (n > 1000)
        return "thatsalot";
    return "other";
};

mf.FormatMessage("You have {number, plural, thatsalot {a shitload of notifications} other {# notifications}}", new Dictionary<string, object>{
  {"number", 1001}
});

Escaping literals

Simple - the literals are {, } and # (in a plural block). If literals occur in the text portions, then they need to be quoted by enclosing them in pairs of single quotes ('). A pair of single quotes always represents one single quote ('' -> '), which still applies inside quoted text. (This '{isn''t}' obvious → This {isn't} obvious)

Anything else?

There's not a lot - Alex Sexton of MessageFormat.js did a great job documenting his library, and like I said, I wrote my implementation so it would be (somewhat) compatible with his.

Bugs / issues

If you have issues with the library, and the exception makes no sense, please open an issue and include your message, as well as the data you used.

Author

I'm Jeff Hansen, a software developer who likes to fiddle with string parsing when it is not too difficult. I also do a lot of ASP.NET Web API back-end development, and quite a bit of web front-end stuff.

You can find me on Twitter: @jeffijoe.