Switch branches/tags
Nothing to show
Find file
Fetching contributors…
Cannot retrieve contributors at this time
366 lines (254 sloc) 10.3 KB

; -- Mode: Markdown; -- ; vim: filetype=markdown tw=76 expandtab shiftwidth=4 tabstop=4

i18n: ICU for Erlang

License: Apache License, Version 2.0

Author: Uvarov Michael (

Module for working with strings and dates. A string is a binary UTF-16 string.

All actions with Unicode were described in the Unicode Standards. ICU Documentation API Reference. Erlang NIF erl_nif API.


i18n is based on ICU 4.2 or newer, which you need to have installed:

brew install icu4c with Homebrew on OSX;

port install icu with MacPorts on OSX;

apt-get install libicu42 libicu-dev on Ubuntu 10.10;

apt-get install libicu44 libicu-dev on Ubuntu 11.10.

Also you need to have gcc for compilation :).

Enviroment Variables

You can configure the compilation process with environment variables. This application uses rebar for building, it also uses this patch for checking env vars.

Try it

Run in the terminal:



There are many operation which are locale-dependable. Each locale has its own rules for working with dates, with collation or even with the case transform. All these rules must be stored in the global memory store.

But when we talk about the Erlang memory management, we talk about own copy of data for each Erlang process. There is no the global store (maybe ETS, but it is not pure Erlang).

Also, there is no the common string format for Erlang. There are lists and binaries. A string as a list are is actually a list of Unicode code points. A binary string is a binary representation of the string (often it is in UTF-8 encoding).

Lists are very helpful, when we work with code points. But they are slow. Binaries are useful, when we store or transmit information.

One of the most powerful library for working with Unicode is ICU. We will use icu4c. It is version of ICU for C and C++ languages. Reasons why we use it are: it is fast, it has fast global data store, it is well tested and it works into the multithread environment very well.


Almost all locale-dependable modules of ICU uses resources. It can be an iterator, a collator, a parser, a transliterator and so on. The creation of the resource is take some time, but its using is fast. So, if you have locale-dependable process or gen_server, create resources and store them (one resource of same type for one longtime process).

We use ICU in the multithread enviroment, so all resources are cloned for each OS's thread by this library. When garbage collector deallocates the resource, all its copies will also deallocated. cloner.c does this work.

Any resource is immutable: you cannot change its internal state from Erlang code. If you load new version of the library, old resources will be here. But you cannot send a resource to another node.

The NIF module is one, because same resources are used in different modules (for example, an iterator resource can be used in string and collation modules).

Strings are represented as binaries in UTF-16 form (two bytes per code point). It can be UTF-16be (big endian) or UTF-16le (little endian). If you use nodes with different endianess, you must convert strings from one form to another.

There are two macroses in i18n/include/i18n.hrl:

-define(ISTR(X), i18n_string:from(X)).
-define(ITS(X),  i18n_string:to_utf8(X)).

When you get a string from the user or from the database, it is often in UTF-8 encoding. You can transform it UTF-16, process this string and transform it back to UTF-8. You can do it in two ways:

% Bad example

Len   = i18n_string:len(?ISTR(Utf8Str)),
Up    = ?ITS(i18n_string:to_upper(?ISTR(Utf8Str))),
UpLen = i18n_string:len(?ISTR(Up)).
% Good example

Str   = ?ISTR(Utf8Str),
Len   = i18n_string:len(Str),
Up    = i18n_string:to_upper(Str),
UpLen = i18n_string:len(Up).

In this example we will get the length of the string and the length of the uppercase version of the string. In first example we convert from one form to another very often, this example will be longer.

Main advice is to convert string when you get it, store string in this form as long as you need it for processing and then encode strings back to UTF-8.

Processing of long strings or first resource allocations can stop other erlang processes, which are scheduled in the same thread. Dirty schedulers can fix this problem, so I am waiting R15.

Functions can throw badarg or exceptions:

{i18n_error, Code :: atom(), tuple()}

For example:

(i18n@delta)1> i18n_regex:open(i18n:from("[[]")).
** exception error: {i18n_error,{'U_REGEX_MISSING_CLOSE_BRACKET',{line,1},

(i18n@delta)2>  i18n_message:open(i18n:from("{rr")).
** exception error: {i18n_error,{'U_UNMATCHED_BRACES',{line,0},{offset,3}}}
     in function  i18n_message:open/2


String case modifications

S=i18n_string:from_utf8(<<"the quick brown Fox jumps over the lazy Dog.">>).

i18n_string:to_utf8(i18n_string:to_title(I, S)).                            


CS = i18n_collation:open([secondary]). 
CT = i18n_collation:open(). 
S = i18n:from("abcd ABCD"). 
P = i18n:from("a"). 
SCSP = i18n_search:open(CS, P). 
SCTP = i18n_search:open(CT, P). 
i18n_search:index(SCSP, S).
i18n_search:index(SCTP, S).           
i18n_search:match_all(SCSP, S).           

Rule-based collation

R1 = i18n_collation:open().
R2 = i18n_collation:open_rules(i18n:from("& g <<< ca")). 
F = fun(R, L) -> 
    lists:map(fun i18n:to/1, 
            lists:map(fun i18n:from/1, L))) end.
L = ["ca", "h", "f", ""].
{F(R1, L), F(R2, L)}.


io:format("~ts", [<<"cà">>]).

Length of the string

With word iterator, punctuation and space characters are counted as "words" (see partitions for more information).

GI = i18n_iterator:open(grapheme),
WI = i18n_iterator:open(word),
WOI = i18n_iterator:open(word_only).

1> i18n_string:len(GI, i18n:from("Длина длинной строки.")).

(i18n@delta)14> i18n_string:len(WOI, i18n:from("Count of the words.")). 

(i18n@delta)15> i18n_string:len(WO, i18n:from("Count of the words.")).

Extracting words

1> lists:map(fun i18n:to/1,
        i18n:from("This string contains 5 words."))).

Partitions of the string

1> lists:map(fun i18n:to/1,
        i18n:from("This string contains 5 words."))).
[<<"This">>,<<" ">>,<<"string">>,<<" ">>,<<"contains">>,
 <<" ">>,<<"5">>,<<" ">>,<<"words">>,<<".">>]

2> Out = lists:map(fun i18n:to/1,
        i18n:from("Erlang är ett generellt programspråk som från början (år
1987) utvecklades på forskningsavdelningen hos telebolaget Ericsson vid
utvärderingen av olika programspråk för implementation av styrsystemen i

3> io:format("~w", [Out]). 

Message format

Simple message format:

M = i18n_message:open(i18n:from("Hello, {name}. Now {now, time, full}.")),
R = i18n_message:format(M, [
        {'name', i18n:from("Username")},
        {'now',  i18n_date:now()}
io:format("~n~ts~n", [i18n_string:to_utf8(R)]).


Hello, Username. Now 17:32:11 GMT+04:00.

Date format:

MLong = i18n_message:open(i18n:from("{0,date,long}")).
MShort = i18n_message:open(i18n:from("{0,date,short}")).
Epoch = i18n_date:new(1970,1,1).
NewDate = i18n_date:add(Epoch, [{day, 3}]).
i18n:to(i18n_message:format(MLong, [NewDate])).
i18n:to(i18n_message:format(MShort, [NewDate])).


(i18n@delta)1> MLong = i18n_message:open(i18n:from("{0,date,long}")).
(i18n@delta)2> MShort = i18n_message:open(i18n:from("{0,date,short}")).
(i18n@delta)3> Epoch = i18n_date:new(1970,1,1).
(i18n@delta)4> NewDate = i18n_date:add(Epoch, [{day, 3}]).
(i18n@delta)5> i18n:to(i18n_message:format(MLong, [NewDate])).
<<"1970 1 4">>
(i18n@delta)6> i18n:to(i18n_message:format(MShort, [NewDate])).

If you want to use ICU messages with gettext, then see l10n.

Using Unicode strings in source code

Because a list can be both a list of bytes (used in source files) and a list of code points (used by default), I suggest use the next form of writing Unicode strings in your code:

?ISTR(<<"Строка Unicode">>).