{-# LANGUAGE NoImplicitPrelude #-} {-# LANGUAGE MultiParamTypeClasses #-} {-# LANGUAGE FlexibleContexts #-} {-# LANGUAGE FlexibleInstances #-} {-# OPTIONS_GHC -fno-warn-orphans #-} {-# OPTIONS_GHC -fno-warn-duplicate-exports #-} {-# LANGUAGE CPP #-} #if __GLASGOW_HASKELL__ >= 800 {-# OPTIONS_GHC -fno-warn-redundant-constraints #-} {-# OPTIONS_GHC -fno-warn-unused-imports #-} #endif module Text.RE.PCRE.Text.Lazy ( -- * Tutorial -- $tutorial -- * The 'Matches' and 'Match' Operators (*=~) , (?=~) -- * The 'SearchReplace' Operators , (*=~/) , (?=~/) -- * The 'Matches' Type , Matches , matchesSource , allMatches , anyMatches , countMatches , matches -- * The 'Match' Type , Match , matchSource , matched , matchedText -- * The Macros and Parsers -- $macros , module Text.RE.TestBench.Parsers -- * The 'RE' Type , RE , reSource -- * Options -- $options , SimpleREOptions(..) -- * Compiling and Escaping REs , SearchReplace(..) , compileRegex , compileRegexWith , compileSearchReplace , compileSearchReplaceWith , escape , escapeWith , escapeREString -- * The Classic regex-base Match Operators , (=~) , (=~~) -- * The re Quasi Quoters -- $re , re , reMultilineSensitive , reMultilineInsensitive , reBlockSensitive , reBlockInsensitive , reMS , reMI , reBS , reBI , re_ -- * The Ed Quasi Quoters -- $ed , edMultilineSensitive , edMultilineInsensitive , edBlockSensitive , edBlockInsensitive , ed , edMS , edMI , edBS , edBI , ed_ -- * The cp Quasi Quoters , cp -- * IsRegex -- $isregex , module Text.RE.Tools.IsRegex ) where import Control.Monad.Fail import qualified Data.Text.Lazy as TL import Data.Typeable import Prelude.Compat import Text.RE.REOptions import Text.RE.Replace import Text.RE.TestBench.Parsers import Text.RE.Tools.IsRegex import Text.RE.ZeInternals import Text.RE.ZeInternals.PCRE import Text.RE.ZeInternals.SearchReplace.PCRE.Text.Lazy import Text.Regex.Base import qualified Text.Regex.PCRE as PCRE import Text.Regex.PCRE.Text.Lazy() -- | find all the matches in the argument text; e.g., to count the number -- of naturals in s: -- -- @countMatches $ s *=~ [re|[0-9]+|]@ -- (*=~) :: TL.Text -> RE -> Matches TL.Text (*=~) bs rex = addCaptureNamesToMatches (reCaptureNames rex) $ match (reRegex rex) bs -- | find the first match in the argument text; e.g., to test if there -- is a natural number in the input text: -- -- @matched $ s ?=~ [re|[0-9]+|]@ -- (?=~) :: TL.Text -> RE -> Match TL.Text (?=~) bs rex = addCaptureNamesToMatch (reCaptureNames rex) $ match (reRegex rex) bs -- | search and replace all matches in the argument text; e.g., this section -- will convert every YYYY-MM-DD format date in its argument text into a -- DD\/MM\/YYYY date: -- -- @(*=~\/ [ed|${y}([0-9]{4})-0*${m}([0-9]{2})-0*${d}([0-9]{2})\/\/\/${d}\/${m}\/${y}|])@ -- (*=~/) :: TL.Text -> SearchReplace RE TL.Text -> TL.Text (*=~/) = flip searchReplaceAll -- | search and replace the first occurrence only (if any) in the input text -- e.g., to prefix the first string of four hex digits in the input text, -- if any, with @0x@: -- -- @(?=~\/ [ed|[0-9A-Fa-f]{4}\/\/\/0x$0|])@ -- (?=~/) :: TL.Text -> SearchReplace RE TL.Text -> TL.Text (?=~/) = flip searchReplaceFirst -- | the `regex-base` polymorphic match operator (=~) :: ( Typeable a , RegexContext PCRE.Regex TL.Text a ) => TL.Text -> RE -> a (=~) bs rex = addCaptureNames (reCaptureNames rex) $ match (reRegex rex) bs -- | the `regex-base` monadic, polymorphic match operator (=~~) :: ( Monad m, MonadFail m , Functor m , Typeable a , RegexContext PCRE.Regex TL.Text a ) => TL.Text -> RE -> m a (=~~) bs rex = addCaptureNames (reCaptureNames rex) <$> matchM (reRegex rex) bs instance IsRegex RE TL.Text where matchOnce = flip (?=~) matchMany = flip (*=~) makeRegexWith = \o -> compileRegexWith o . unpackR makeSearchReplaceWith = \o r t -> compileSearchReplaceWith o (unpackR r) (unpackR t) regexSource = packR . reSource -- $tutorial -- We have a regex tutorial at <http://tutorial.regex.uk>. -- $macros -- There are a number of RE macros and corresponding Haskell parsers -- for parsing the matched text into appropriate Haskell types. See -- the [Macros Tables](http://regex.uk/macros) for details. -- $options -- You can specify different compilation options by appending a -- to the name of an [re| ... |] or [ed| ... \/\/\/ ... |] quasi quoter -- to select the corresponding compilation option. For example, the -- section, -- -- @(?=~/ [edBlockInsensitive|foo$\/\/\/bar|])@ -- -- will replace a @foo@ suffix of the argument text, of any -- capitalisation, with a (lower case) @bar@. If you need to specify the -- options dynamically, use the @[re_| ... |]@ and @[ed_| ... \/\/\/ ... |]@ -- quasi quoters, which generate functions that take an 'IsOption' option -- (e.g., a 'SimpleReOptions' value) and yields a 'RE' or 'SearchReplace' -- as apropriate. For example if you have a 'SimpleReOptions' value in -- @sro@ then -- -- @(?=~/ [ed_|foo$\/\/\/bar|] sro)@ -- -- will compile the @foo$@ RE according to the value of @sro@. For more -- on specifying RE options see "Text.RE.REOptions". -- $re -- The @[re|.*|]@ quasi quoters, with variants for specifing different -- options to the RE compiler (see "Text.RE.REOptions"), and the -- specialised back-end types and functions. -- $ed -- The @[ed|.*\/\/\/foo|]@ quasi quoters, with variants for specifing different -- options to the RE compiler (see "Text.RE.REOptions"). -- $ed -- The -- | the @[ed| ... \/\/\/ ... |]@ quasi quoters; for example, -- -- @[ed|${y}([0-9]{4})-0*${m}([0-9]{2})-0*${d}([0-9]{2})\/\/\/${d}\/${m}\/${y}|])@ -- -- represents a @SearchReplace@ that will convert a YYYY-MM-DD format date -- into a DD\/MM\/YYYY format date. -- -- The only difference betweem these quasi quoters is the RE options that are set, -- using the same conventions as the @[re| ... |]@ quasi quoters. -- $isregex -- The 'IsRegex' class is used to abstact over the different regex back ends and -- the text types they work with -- see "Text.RE.Tools.IsRegex" for details.