This is regex-tdfa which is a pure Haskell regular expression library (for POSIX extended regular expressions) originally written by Christopher Kuklewicz.

The name “tdfa” stands for Tagged-DFA.

Getting started

Importing and using

Declare a dependency on the regex-tdfa library in your .cabal file:

build-depends: regex-tdfa ^>= 1.3.1

In Haskell modules where you need to use regexes import the respective regex-tdfa module:

import Text.Regex.TDFA


λ> emailRegex = "[a-zA-Z0-9+._-]+@[a-zA-Z-]+\\.[a-z]+"
λ> "my email is [email protected]" =~ emailRegex :: Bool
>>> True

-- non-monadic
<to-match-against> =~ <regex>

-- monadic, uses 'fail' on lack of match
<to-match-against> =~~ <regex>

(=~) and (=~~) are polymorphic in their return type. This is so that regex-tdfa can pick the most efficient way to give you your result based on what you need. For instance, if all you want is to check whether the regex matched or not, there’s no need to allocate a result string. If you only want the first match, rather than all the matches, then the matching engine can stop after finding a single hit.

This does mean, though, that you may sometimes have to explicitly specify the type you want, especially if you’re trying things out at the REPL.

Common use cases

Get the first match

-- returns empty string if no match
a =~ b :: String  -- or ByteString, or Text...

λ> "alexis-de-tocqueville" =~ "[a-z]+" :: String
>>> "alexis"

λ> "alexis-de-tocqueville" =~ "[0-9]+" :: String
>>> ""

Check if it matched at all

a =~ b :: Bool

λ> "alexis-de-tocqueville" =~ "[a-z]+" :: Bool
>>> True

Get first match + text before/after

-- if no match, will just return whole
-- string in the first element of the tuple
a =~ b :: (String, String, String)

λ> "alexis-de-tocqueville" =~ "de" :: (String, String, String)
>>> ("alexis-", "de", "-tocqueville")

λ> "alexis-de-tocqueville" =~ "kant" :: (String, String, String)
>>> ("alexis-de-tocqueville", "", "")

Get first match + submatches

-- same as above, but also returns a list of /just/ submatches
-- submatch list is empty if regex doesn't match at all
a =~ b :: (String, String, String, [String])

λ> "div[attr=1234]" =~ "div\\[([a-z]+)=([^]]+)\\]"
     :: (String, String, String, [String])
>>> ("", "div[attr=1234]", "", ["attr","1234"])

Get all non-overlapping matches

-- can also return Data.Array instead of List
getAllTextMatches (a =~ b) :: [String]

λ> getAllTextMatches ("john anne yifan" =~ "[a-z]+") :: [String]
>>> ["john","anne","yifan"]

λ> getAllTextMatches ("0a0b0" =~ "0[a-z]0") :: [String]
>>> ["0a0"]

Note that "0b0" is not included in the result since it overlaps with "0a0".

Special characters

regex-tdfa only supports a small set of special characters and is much less featureful than some other regex engines you might be used to, such as PCRE.

  • \` — Match start of entire text (similar to ^ in other regex engines)
  • \' — Match end of entire text (similar to $ in other regex engines)
  • \< — Match beginning of word
  • \> — Match end of word
  • \b — Match beginning or end of word
  • \B — Match neither beginning nor end of word

Less common stuff

Get match indices

-- can also return Data.Array instead of List
getAllMatches (a =~ b) :: [(Int, Int)]  -- (index, length)

λ> getAllMatches ("john anne yifan" =~ "[a-z]+") :: [(Int, Int)]
>>> [(0,4), (5,4), (10,5)]

Get submatch indices

-- match of __entire__ regex is first element, not first capture
-- can also return Data.Array instead of List
getAllSubmatches (a =~ b) :: [(Int, Int)]  -- (index, length)

λ> getAllSubmatches ("div[attr=1234]" =~ "div\\[([a-z]+)=([^]]+)\\]")
     :: [(Int, Int)]
>>> [(0,14), (4,4), (9,4)]


regex-tdfa does not provide find-and-replace.

The relevant links

This documentation is also available in Text.Regex.TDFA haddock.

This was also documented at the Haskell wiki. The original Darcs repository was at When not updated, this was forked and maintained by Roman Cheplyaka as regex-tdfa-rc.

Then the repository moved to, which was primarily maintained by Artyom (neongreen).

Finally, maintainership was passed on again and the repository moved to its current location at

Avoiding backslashes

If you find yourself writing a lot of regexes, take a look at raw-strings-qq. It’ll let you write regexes without needing to escape all your backslashes.

{-# LANGUAGE QuasiQuotes #-}

import Text.RawString.QQ
import Text.Regex.TDFA

λ> "2 * (3 + 1) / 4" =~ [r|\([^)]+\)|] :: String
>>> "(3 + 1)"

Known bugs and infelicities

  • Regexes with large character classes combined with {m,n} are very slow and memory-hungry (#3).

    An example of such a regex is ^[\x0020-\xD7FF]{1,255}$.

  • POSIX submatch semantics are broken in some rare cases (#2).

About this package

This was inspired by the algorithm (and Master’s thesis) behind the regular expression library known as TRE or libtre. This was created by Ville Laurikari and tackled the difficult issue of efficient sub-match capture for POSIX regular expressions.

By building on this thesis and adding a few more optimizations, regex-tdfa matching input text of length N should have O(N) runtime, and should have a maximum memory bounded by the pattern size that does not scale with N. It should do this while returning well defined (and correct) values for the parenthesized sub-matches.

Regardless of performance, nearly every single OS and Libra for POSIX regular expressions has bugs in sub-matches. This was detailed on the Regex POSIX Haskell wiki page, and can be demonstrated with the regex-posix-unittest suite of checks. Test regex-tdfa-unittest should show regex-tdfa passing these same checks. I owe my understanding of the correct behvior and many of these unit tests to Glenn Fowler at AT&T (“An Interpretation of the POSIX regex Standard”).

Other related packages

You can find several other related packages by searching for “tdfa” on hackage.

Document notes

This was written 2016-04-30.


For the package version policy (PVP), see .

2022-02-19, Andreas Abel

  • No longer rely on the MonadFail instance for ST (future base library change, see #29).
  • Silence warning incomplete-uni-patterns (GHC >= 9.2).
  • Import from Data.List explicitly or qualified (warning compat-unqualified-imports).
  • Import from Control.Monad to allow mtl-2.3 in its rc3 incarnation. Revision 3

2022-01-31, Andreas Abel

  • Speculatively allow unreleased mtl-2.3 (works with release candidate mtl-2.3-rc4). Revision 2

2021-12-26, Andreas Abel

  • Allow text-2.0. Revision 1

2021-08-12, Andreas Abel

  • Compatibility with base-4.16 (GHC 9.2).

2021-06-03, Andreas Abel

  • Removed extension NoMonoPatBinds from .cabal-file for GHC 9.2 compatibility.
  • Removed some outdated documentation. Revision 2

2021-02-20, Andreas Abel

  • Compatibility with base-4.15 (GHC 9.0) and bytestring-0.11. Revision 1

2020-03-26, phadej

  • Compatibility with base-4.14 (GHC 8.10).

2019-11-26, hvr

1.3.0 Revision 1

2019-11-26, hvr

  • Tighten bounds on base, mtl, parsec and fail.


2019-09-29, Artyom (deprecated, not following PVP)

  • Compatibility with GHC 8.8 and regex-base-0.9.4 (h/t @asr).
  • Turned regex-tdfa-unittest into a regex-tdfa testsuite.

2019-05-09, Artyom

  • Significantly improved documentation (h/t William Yao).

2018-06-22, Artyom

  • Compatibility with containers-0.6.


2018-03-10, Artyom

  • Added Semigroup instances for some types (h/t Herbert Valerio Riedel).


2016-04-28, Artyom

  • New maintainer.
  • Now we don’t reexport the problematic Show instance for functions.


2015-08-29, Chris Kuklewicz

  • Updated dependency versions.


2014-02-02, Chris Kuklewicz

  • “Almost ghc-7.8” with the array 0.4 changes for Data.Array.Unsafe


Make ghc-7.0.2 on platorm 2011. happy


fix url below


Fix bug preventing []] [-] [^]] [^-] (thanks to Maxime Henrion)


try needUniqTags in POr in CorePattern.hs, try toAdvice b for PStar child




BROKEN after 100 characters the compressOrbit dies!




add gnu escapes


NewDFA code working


make NewDFA directory and String_NC


try NewDFATest_SBS with uncons


use uncons on SBS


try repaired NewDFATest_SBS

  • np13: try to improve readability with the mm combinator? Yes!
  • np12: expand o in the case where t lookup get Nothing? Yes – this is the fix!?
  • np11: break multi to not look at o and just return True? Yes !!!!
  • np10: Peel off CharMap/IntMap and DFA/DT with pattern matching? No
  • np9: INLINE endOf? No
  • np8: np6 and NOINLINE endOff? No
  • np7: just return True? Fast
  • np6: comment out ans check? No
  • np5: comment out all Multi0 code? No
  • np4: comment out all Single0 and Single code? No
  • np3: !off the multi? No
  • np2: comment out all Testing code? No


try to alter matchTest to not have the Bool args? No


arg, the prof is fast and the normal slow!


add NewDFATest.hs


go to many vs single?


try for pre-comparison of orbit-logs!


try alternate lazy/strict strategy in NewDFA. Fix offset laziness.


radical removal of flag array and adding of SetVal to handle groups


performance? up to v15


get string with NewDFA testing, unit tests and 1000 random regex pass


start changing to the new real DFA


more cleanup


try to debug 0.99.11: fixed updateWinner


improve above fix and make stuff work better – HAS BUG, along with old TDFA!


fixed ((.?)*)* patterns by changing PStar nullView when mayFirstBeNull


testing changing bestTrans/chooseWith/choose to include enterOrbit/newFlags/(_,True) info


testing changing Maximize to Minimize for Tags, decide (a*)* is canonical problem


Use (PGroup Nothing) in Pattern to decompose PBound


change to nested nonEmpty calls for PBound


remove PNonEmpty constructor


tests pnonempty' = \ p -> POr [ PEmpty, p ] instead of PNonEmpty