Erlang logo
User's Guide
Reference Manual
Release Notes

Reference Manual
Version 3.1

Expand All
Contract All

Table of Contents





Functions for converting Unicode characters.


This module contains functions for converting between different character representations. It converts between ISO Latin-1 characters and Unicode characters, but it can also convert between different Unicode encodings (like UTF-8, UTF-16, and UTF-32).

The default Unicode encoding in Erlang is in binaries UTF-8, which is also the format in which built-in functions and libraries in OTP expect to find binary Unicode data. In lists, Unicode data is encoded as integers, each integer representing one character and encoded simply as the Unicode code point for the character.

Other Unicode encodings than integers representing code points or UTF-8 in binaries are referred to as "external encodings". The ISO Latin-1 encoding is in binaries and lists referred to as latin1-encoding.

It is recommended to only use external encodings for communication with external entities where this is required. When working inside the Erlang/OTP environment, it is recommended to keep binaries in UTF-8 when representing Unicode characters. ISO Latin-1 encoding is supported both for backward compatibility and for communication with external entities not supporting Unicode character sets.


encoding() =
    latin1 |
    unicode |
    utf8 |
    utf16 |
    {utf16, endian()} |
    utf32 |
    {utf32, endian()}

endian() = big | little

unicode_binary() = binary()

A binary() with characters encoded in the UTF-8 coding standard.

chardata() = charlist() | unicode_binary()

charlist() =
    maybe_improper_list(char() | unicode_binary() | charlist(),
                        unicode_binary() | [])

external_unicode_binary() = binary()

A binary() with characters coded in a user-specified Unicode encoding other than UTF-8 (that is, UTF-16 or UTF-32).

external_chardata() =
    external_charlist() | external_unicode_binary()

external_charlist() =
    maybe_improper_list(char() |
                        external_unicode_binary() |
                        external_unicode_binary() | [])

latin1_binary() = binary()

A binary() with characters coded in ISO Latin-1.

latin1_char() = byte()

An integer() representing a valid ISO Latin-1 character (0-255).

latin1_chardata() = latin1_charlist() | latin1_binary()

Same as iodata().

latin1_charlist() =
    maybe_improper_list(latin1_char() |
                        latin1_binary() |
                        latin1_binary() | [])

Same as iolist().


bom_to_encoding(Bin) -> {Encoding, Length}


Bin = binary()
A binary() such that byte_size(Bin) >= 4.
Encoding =
    latin1 | utf8 | {utf16, endian()} | {utf32, endian()}
Length = integer() >= 0
endian() = big | little

Checks for a UTF Byte Order Mark (BOM) in the beginning of a binary. If the supplied binary Bin begins with a valid BOM for either UTF-8, UTF-16, or UTF-32, the function returns the encoding identified along with the BOM length in bytes.

If no BOM is found, the function returns {latin1,0}.

characters_to_binary(Data) -> Result


Result =
    binary() |
    {error, binary(), RestData} |
    {incomplete, binary(), binary()}

Same as characters_to_binary(Data, unicode, unicode).

characters_to_binary(Data, InEncoding) -> Result


InEncoding = encoding()
Result =
    binary() |
    {error, binary(), RestData} |
    {incomplete, binary(), binary()}

Same as characters_to_binary(Data, InEncoding, unicode).

characters_to_binary(Data, InEncoding, OutEncoding) -> Result


InEncoding = OutEncoding = encoding()
Result =
    binary() |
    {error, binary(), RestData} |
    {incomplete, binary(), binary()}

Behaves as characters_to_list/2, but produces a binary instead of a Unicode list.

InEncoding defines how input is to be interpreted if binaries are present in Data

OutEncoding defines in what format output is to be generated.



An alias for utf8, as this is the preferred encoding for Unicode characters in binaries.


An alias for {utf16,big}.


An alias for {utf32,big}.

The atoms big and little denote big- or little-endian encoding.

Errors and exceptions occur as in characters_to_list/2, but the second element in tuple error or incomplete is a binary() and not a list().

characters_to_list(Data) -> Result


Result =
    list() |
    {error, list(), RestData} |
    {incomplete, list(), binary()}

Same as characters_to_list(Data, unicode).

characters_to_list(Data, InEncoding) -> Result


InEncoding = encoding()
Result =
    list() |
    {error, list(), RestData} |
    {incomplete, list(), binary()}

Converts a possibly deep list of integers and binaries into a list of integers representing Unicode characters. The binaries in the input can have characters encoded as one of the following:

  • ISO Latin-1 (0-255, one character per byte). Here, case parameter InEncoding is to be specified as latin1.

  • One of the UTF-encodings, which is specified as parameter InEncoding.

Only when InEncoding is one of the UTF encodings, integers in the list are allowed to be > 255.

If InEncoding is latin1, parameter Data corresponds to the iodata() type, but for unicode, parameter Data can contain integers > 255 (Unicode characters beyond the ISO Latin-1 range), which makes it invalid as iodata().

The purpose of the function is mainly to convert combinations of Unicode characters into a pure Unicode string in list representation for further processing. For writing the data to an external entity, the reverse function characters_to_binary/3 comes in handy.

Option unicode is an alias for utf8, as this is the preferred encoding for Unicode characters in binaries. utf16 is an alias for {utf16,big} and utf32 is an alias for {utf32,big}. The atoms big and little denote big- or little-endian encoding.

If the data cannot be converted, either because of illegal Unicode/ISO Latin-1 characters in the list, or because of invalid UTF encoding in any binaries, an error tuple is returned. The error tuple contains the tag error, a list representing the characters that could be converted before the error occurred and a representation of the characters including and after the offending integer/bytes. The last part is mostly for debugging, as it still constitutes a possibly deep or mixed list, or both, not necessarily of the same depth as the original data. The error occurs when traversing the list and whatever is left to decode is returned "as is".

However, if the input Data is a pure binary, the third part of the error tuple is guaranteed to be a binary as well.

Errors occur for the following reasons:

  • Integers out of range.

    If InEncoding is latin1, an error occurs whenever an integer > 255 is found in the lists.

    If InEncoding is of a Unicode type, an error occurs whenever either of the following is found:

    • An integer > 16#10FFFF (the maximum Unicode character)

    • An integer in the range 16#D800 to 16#DFFF (invalid range reserved for UTF-16 surrogate pairs)

  • Incorrect UTF encoding.

    If InEncoding is one of the UTF types, the bytes in any binaries must be valid in that encoding.

    Errors can occur for various reasons, including the following:

    • "Pure" decoding errors (like the upper bits of the bytes being wrong).

    • The bytes are decoded to a too large number.

    • The bytes are decoded to a code point in the invalid Unicode range.

    • Encoding is "overlong", meaning that a number should have been encoded in fewer bytes.

    The case of a truncated UTF is handled specially, see the paragraph about incomplete binaries below.

    If InEncoding is latin1, binaries are always valid as long as they contain whole bytes, as each byte falls into the valid ISO Latin-1 range.

A special type of error is when no actual invalid integers or bytes are found, but a trailing binary() consists of too few bytes to decode the last character. This error can occur if bytes are read from a file in chunks or if binaries in other ways are split on non-UTF character boundaries. An incomplete tuple is then returned instead of the error tuple. It consists of the same parts as the error tuple, but the tag is incomplete instead of error and the last element is always guaranteed to be a binary consisting of the first part of a (so far) valid UTF character.

If one UTF character is split over two consecutive binaries in the Data, the conversion succeeds. This means that a character can be decoded from a range of binaries as long as the whole range is specified as input without errors occurring.


decode_data(Data) ->
   case unicode:characters_to_list(Data,unicode) of
      {incomplete,Encoded, Rest} ->
            More = get_some_more_data(),
            Encoded ++ decode_data([Rest, More]);
      {error,Encoded,Rest} ->
      List ->

However, bit strings that are not whole bytes are not allowed, so a UTF character must be split along 8-bit boundaries to ever be decoded.

A badarg exception is thrown for the following cases:

  • Any parameters are of the wrong type.
  • The list structure is invalid (a number as tail).
  • The binaries do not contain whole bytes (bit strings).

encoding_to_bom(InEncoding) -> Bin


Bin = binary()
A binary() such that byte_size(Bin) >= 4.
InEncoding = encoding()

Creates a UTF Byte Order Mark (BOM) as a binary from the supplied InEncoding. The BOM is, if supported at all, expected to be placed first in UTF encoded files or messages.

The function returns <<>> for latin1 encoding, as there is no BOM for ISO Latin-1.

Notice that the BOM for UTF-8 is seldom used, and it is really not a byte order mark. There are obviously no byte order issues with UTF-8, so the BOM is only there to differentiate UTF-8 encoding from other UTF formats.