text-icu- Bindings to the ICU library

Safe HaskellSafe-Infered




Character set normalization functions for Unicode, implemented as bindings to the International Components for Unicode (ICU) libraries.


Unicode normalization API

The normalize function transforms Unicode text into an equivalent composed or decomposed form, allowing for easier sorting and searching of text. normalize supports the standard normalization forms described in http://www.unicode.org/unicode/reports/tr15/, Unicode Standard Annex #15: Unicode Normalization Forms.

Characters with accents or other adornments can be encoded in several different ways in Unicode. For example, take the character A-acute. In Unicode, this can be encoded as a single character (the "composed" form):


or as two separate characters (the "decomposed" form):


To a user of your program, however, both of these sequences should be treated as the same "user-level" character "A with acute accent". When you are searching or comparing text, you must ensure that these two sequences are treated equivalently. In addition, you must handle characters with more than one accent. Sometimes the order of a character's combining accents is significant, while in other cases accent sequences in different orders are really equivalent.

Similarly, the string "ffi" can be encoded as three separate letters:

      0066    LATIN SMALL LETTER F
      0066    LATIN SMALL LETTER F
      0069    LATIN SMALL LETTER I

or as the single character


The "ffi" ligature is not a distinct semantic character, and strictly speaking it shouldn't be in Unicode at all, but it was included for compatibility with existing character sets that already provided it. The Unicode standard identifies such characters by giving them "compatibility" decompositions into the corresponding semantic characters. When sorting and searching, you will often want to use these mappings.

normalize helps solve these problems by transforming text into the canonical composed and decomposed forms as shown in the first example above. In addition, you can have it perform compatibility decompositions so that you can treat compatibility characters the same as their equivalents. Finally, normalize rearranges accents into the proper canonical order, so that you do not have to worry about accent rearrangement on your own.

Form FCD, "Fast C or D", is also designed for collation. It allows to work on strings that are not necessarily normalized with an algorithm (like in collation) that works under "canonical closure", i.e., it treats precomposed characters and their decomposed equivalents the same.

It is not a normalization form because it does not provide for uniqueness of representation. Multiple strings may be canonically equivalent (their NFDs are identical) and may all conform to FCD without being identical themselves.

The form is defined such that the "raw decomposition", the recursive canonical decomposition of each character, results in a string that is canonically ordered. This means that precomposed characters are allowed for as long as their decompositions do not need canonical reordering.

Its advantage for a process like collation is that all NFD and most NFC texts - and many unnormalized texts - already conform to FCD and do not need to be normalized (NFD) for such a process. The FCD quickCheck will return Yes for most strings in practice.

normalize FCD may be implemented with NFD.

For more details on FCD see the collation design document: http://source.icu-project.org/repos/icu/icuhtml/trunk/design/collation/ICU_collation_design.htm

ICU collation performs either NFD or FCD normalization automatically if normalization is turned on for the collator object. Beyond collation and string search, normalized strings may be useful for string equivalence comparisons, transliteration/transcription, unique representations, etc.

The W3C generally recommends to exchange texts in NFC. Note also that most legacy character encodings use only precomposed forms and often do not encode any combining marks by themselves. For conversion to such character encodings the Unicode text needs to be normalized to NFC. For more usage examples, see the Unicode Standard Annex.

data NormalizationMode Source

Normalization modes.



No decomposition/composition.


Canonical decomposition.


Compatibility decomposition.


Canonical decomposition followed by canonical composition.


Compatibility decomposition followed by canonical composition.


"Fast C or D" form.

Normalization functions

normalize :: NormalizationMode -> Text -> TextSource

Normalize a string according the specified normalization mode.

Normalization checks

quickCheck :: NormalizationMode -> Text -> Maybe BoolSource

Perform an efficient check on a string, to quickly determine if the string is in a particular normalization form.

A Nothing result indicates that a definite answer could not be determined quickly, and a more thorough check is required, e.g. with isNormalized. The user may have to convert the string to its normalized form and compare the results.

A result of Just True or Just False indicates that the string definitely is, or is not, in the given normalization form.

isNormalized :: NormalizationMode -> Text -> BoolSource

Indicate whether a string is in a given normalization form.

Unlike quickCheck, this function returns a definitive result. For NFD, NFKD, and FCD normalization forms, both functions work in exactly the same ways. For NFC and NFKC forms, where quickCheck may return Nothing, this function will perform further tests to arrive at a definitive result.

Normalization-sensitive comparison

data CompareOption Source

Options to compare.



The caller knows that both strings fulfill the FCD conditions. If not set, compare will quickCheck for FCD and normalize if necessary.


Compare strings case-insensitively using case folding, instead of case-sensitively. If set, then the following case folding options are used.


When case folding, exclude the special I character. For use with Turkic (Turkish/Azerbaijani) text data.

compare :: [CompareOption] -> Text -> Text -> OrderingSource

Compare two strings for canonical equivalence. Further options include case-insensitive comparison and code point order (as opposed to code unit order).

Canonical equivalence between two strings is defined as their normalized forms (NFD or NFC) being identical. This function compares strings incrementally instead of normalizing (and optionally case-folding) both strings entirely, improving performance significantly.

Bulk normalization is only necessary if the strings do not fulfill the FCD conditions. Only in this case, and only if the strings are relatively long, is memory allocated temporarily. For FCD strings and short non-FCD strings there is no memory allocation.