An efficient packed Unicode text type.


Version on this page:
LTS Haskell 21.13:2.0.2
Stackage Nightly 2023-09-24:2.0.2
Latest on Hackage:2.1

See all snapshots text appears in

BSD-2-Clause licensed and maintained by Bryan O'Sullivan
This version can be pinned in stack with:text-,8541

text: Fast, packed Unicode strings, using stream fusion

This package provides the Data.Text library, a library for the space- and time-efficient manipulation of Unicode text in Haskell.

Please refer to the package description on Hackage for more information.

Get involved!

Please report bugs via the github issue tracker.

Master git repository:

  • git clone git://github.com/haskell/text.git

There’s also a Mercurial mirror:

  • hg clone https://bitbucket.org/bos/text

(You can create and contribute changes using either Mercurial or git.)


The base code for this library was originally written by Tom Harper, based on the stream fusion framework developed by Roman Leshchinskiy, Duncan Coutts, and Don Stewart.

The core library was fleshed out, debugged, and tested by Bryan O’Sullivan bos@serpentine.com, and he is the current maintainer.


  • Spec compliance: toCaseFold now follows the Unicode 9.0 spec (updated from 8.0).

  • Bug fix: the lazy takeWhileEnd function violated the lazy text invariant (gh-184).

  • Bug fix: Fixed usage of size hints causing incorrect behavior (gh-197).

  • New function: unsnoc (gh-173).

  • Reduce memory overhead in encodeUTF8 (gh-194).

  • Improve UTF-8 decoder error-recovery (gh-182).

  • Minor documentation improvements (@since annotations, more examples, clarifications).

  • The toTitle function now correctly handles letters that immediately follow punctuation. Before, "there's" would turn into "There'S". Now, it becomes "There's".

  • The implementation of unstreaming is faster, resulting in operations such as map and intersperse speeding up by up to 30%, with smaller code generated.

  • The optimised length comparison function is now more likely to be used after some rewrite rule tweaking.

  • Bug fix: an off-by-one bug in takeEnd is fixed.

  • Bug fix: a logic error in takeWord16 is fixed.

  • The switch to integer-pure in was apparently mistaken. The build flag has been renamed accordingly. Your army of diligent maintainers apologizes for the churn.

  • Spec compliance: toCaseFold now follows the Unicode 8.0 spec (updated from 7.0)

  • An STG lint error has been fixed

  • The integer-simple package, upon which this package optionally depended, has been replaced with integer-pure. The build flag has been renamed accordingly.

  • Bug fix: For the Binary instance, If UTF-8 decoding fails during a get, the error is propagated via fail instead of an uncatchable crash.

  • New function: takeWhileEnd

  • New instances for the Text types:

    • if base >= 4.7: PrintfArg
    • if base >= 4.9: Semigroup

  • Bug fix: As it turns out, moving the literal rewrite rules to simplifier phase 2 does not prevent competition with the unpack rule, which is also active in this phase. Unfortunately this was hidden due to a silly test environment mistake. Moving literal rules back to phase 1 finally fixes GHC Trac #10528 correctly.

  • Bug fix: Run literal rewrite rules in simplifier phase 2. The behavior of the simplifier changed in GHC 7.10.2, causing these rules to fail to fire, leading to poor code generation and long compilation times. See GHC Trac #10528.

  • Expose unpackCString#, which you should never use.

  • Added Binary instances for both Text types. (If you have previously been using the text-binary package to get a Binary instance, it is now obsolete.)

  • Fixed a space leak in UTF-8 decoding

  • Feature parity: repeat, cycle, iterate are now implemented for lazy Text, and the Data instance is more complete

  • Build speed: an inliner space explosion has been fixed with toCaseFold

  • Bug fix: encoding Int to a Builder would infinite-loop if the integer-simple package was used

  • Deprecation: OnEncodeError and EncodeError are deprecated, as they are never used

  • Internals: some types that are used internally in fusion-related functions have moved around, been renamed, or been deleted (we don’t bump the major version if .Internal modules change)

  • Spec compliance: toCaseFold now follows the Unicode 7.0 spec (updated from 6.3)

  • Fixed an incompatibility with base < 4.5

  • Bumped lower bound on deepseq to 1.4 for compatibility with the upcoming GHC 7.10

  • The Data.Data instance now allows gunfold to work, via a virtual pack constructor

  • dropEnd, takeEnd: new functions

  • Comparing the length of a Text against a number can now short-circuit in more cases

  • streamDecodeUtf8: fixed gh-70, did not return all unconsumed bytes in single-byte chunks

  • encodeUtf8: Performance is improved by up to 4x.

  • encodeUtf8Builder, encodeUtf8BuilderEscaped: new functions, available only if bytestring >= is installed, that allow very fast and flexible encoding of a Text value to a bytestring Builder.

    As an example of the performance gain to be had, the encodeUtf8BuilderEscaped function helps to double the speed of JSON encoding in the latest version of aeson! (Note: if all you need is a plain ByteString, encodeUtf8 is still the faster way to go.)

  • All of the internal module hierarchy is now publicly exposed. If a module is in the .Internal hierarchy, or is documented as internal, use at your own risk - there are no API stability guarantees for internal modules!

  • decodeUtf8: Fixed a regression that caused us to incorrectly identify truncated UTF-8 as valid (gh-61)

  • Added support for Unicode 6.3.0 to case conversion functions

  • New function toTitle converts words in a string to title case

  • New functions peekCStringLen and withCStringLen simplify interoperability with C functionns

  • Added support for decoding UTF-8 in stream-friendly fashion

  • Fixed a bug in mapAccumL

  • Added trusted Haskell support

  • Removed support for GHC 6.10 (released in 2008) and older