diff --git a/CHANGES.markdown b/CHANGES.markdown --- a/CHANGES.markdown +++ b/CHANGES.markdown @@ -1,5 +1,31 @@ -% Changelog for `weighted-regexp` +% Changelog for [`weighted-regexp`] +[`weighted-regexp`]: http://hackage.haskell.org/package/weighted-regexp + +# 0.3.0.1 + +Moved build dependencies for QuickCheck and Criterion test programs +under a conditional so they are only pulled in if one actually +compiles these programs using the flags `-fQuickCheck` or +`-fCriterion`. (Thank you Brent!) + +# 0.3.0.0 + +## Implemented workaround for [GHC ticket 4227] + +Currently, GHC can SPECIALIZE functions only where they are defined. +The types `Leftmost`, `Longest`, and `LeftLong` are now defined in +separate modules to bring them into the scope of the matching +functions. Specialization makes the matching functions almost three +times faster for the types mentioned above. + +This workaround allows to specialize the matching functions for types +defined in this package. Users, however, must use the matching +functions unspecialized for their own types. + +Along with this change, the constructors of the matching types are no +longer exported. + # 0.2.0.0 ## More general types for matching functions @@ -29,20 +55,22 @@ ## Strict numeric semiring The lazy definition of arithmetic operations for the `Numeric` -semiring has been dropped in favour of the more efficient -standard implementation. As a consequence, `matchingCount` no -longer works with infinite regular expressions. +semiring has been dropped in favour of the more efficient standard +implementation. As a consequence, `matchingCount` no longer works with +infinite regular expressions. ## SPECIALIZE pragmas prevent memory leak -The generalization of the matching functions leads to a memory -leak that can be avoided by specializing them for concrete -semirings. Corresponding pragmas have been added for `Bool` and -for `Numeric` types but not for the more complex semirings defined -in the extra matching modules. It is unclear what is the best way -to specialize them too because the pragma must be placed in the -module where the matching functions are defined but, there, not -all semirings are in scope. +The generalization of the matching functions leads to a memory leak +that can be avoided by specializing them for concrete +semirings. Corresponding pragmas have been added for `Bool` and for +`Numeric` types but not for the more complex semirings defined in the +extra matching modules. It is unclear what is the best way to +specialize them too because the pragma must be placed in the module +where the matching functions are defined but, there, not all semirings +are in scope. See [GHC ticket 4227]. + +[GHC ticket 4227]: http://hackage.haskell.org/trac/ghc/ticket/4227 ## Fixed mistake in Criterion benchmarks diff --git a/README.markdown b/README.markdown --- a/README.markdown +++ b/README.markdown @@ -52,9 +52,11 @@ -You can install the `weighted-regexp` library by typing the following +You can install the [`weighted-regexp`] library by typing the following into a terminal: +[`weighted-regexp`]: http://hackage.haskell.org/package/weighted-regexp + bash# cabal update bash# cabal install weighted-regexp @@ -169,8 +171,7 @@ fast in this example, in fact, faster than re2[^cpp]. [^cpp]: The following C++ program uses the [re2] library and needs -*4.5s* to match `.*a.{20}a.*` against a string of ~2M random a's ad -b's: +*4s* to search for `a.{20}a` in a string of ~2M random a's ad b's: @@ -208,20 +209,20 @@ matching acceptance #matchings leftmost longest leftmost longest --------------- ----------- ----------- ---------- ---------- ----------------- - unique full [4.0 us] [4.6 us] - ambiguous full [11.7 us] [13.1 us] - partial [21.2 us] [74.2 us] [68.0 us] [73.8 us] + unique full [3.8 us] [4.8 us] + ambiguous full [11.7 us] [13.4 us] + partial [20.4 us] [27.2 us] [26.2 us] [27.5 us] Click on the numbers for a more detailed distribution of run times. -[4.0 us]: http://sebfisch.github.com/haskell-regexp/criterion/full-accept-phone-densities-800x600.png -[4.6 us]: http://sebfisch.github.com/haskell-regexp/criterion/full-count-phone-densities-800x600.png +[3.8 us]: http://sebfisch.github.com/haskell-regexp/criterion/full-accept-phone-densities-800x600.png +[4.8 us]: http://sebfisch.github.com/haskell-regexp/criterion/full-count-phone-densities-800x600.png [11.7 us]: http://sebfisch.github.com/haskell-regexp/criterion/full-accept-html-densities-800x600.png -[13.1 us]: http://sebfisch.github.com/haskell-regexp/criterion/full-count-html-densities-800x600.png -[21.2 us]: http://sebfisch.github.com/haskell-regexp/criterion/partial-accept-rna-densities-800x600.png -[74.2 us]: http://sebfisch.github.com/haskell-regexp/criterion/partial-leftmost-rna-densities-800x600.png -[68.0 us]: http://sebfisch.github.com/haskell-regexp/criterion/partial-longest-rna-densities-800x600.png -[73.8 us]: http://sebfisch.github.com/haskell-regexp/criterion/partial-leftlong-rna-densities-800x600.png +[13.4 us]: http://sebfisch.github.com/haskell-regexp/criterion/full-count-html-densities-800x600.png +[20.4 us]: http://sebfisch.github.com/haskell-regexp/criterion/partial-accept-rna-densities-800x600.png +[27.2 us]: http://sebfisch.github.com/haskell-regexp/criterion/partial-leftmost-rna-densities-800x600.png +[26.2 us]: http://sebfisch.github.com/haskell-regexp/criterion/partial-longest-rna-densities-800x600.png +[27.5 us]: http://sebfisch.github.com/haskell-regexp/criterion/partial-leftlong-rna-densities-800x600.png # Collaboration diff --git a/src/Text/RegExp/Matching/Longest.hs b/src/Text/RegExp/Matching/Longest.hs --- a/src/Text/RegExp/Matching/Longest.hs +++ b/src/Text/RegExp/Matching/Longest.hs @@ -1,5 +1,3 @@ -{-# LANGUAGE MultiParamTypeClasses, FlexibleInstances #-} - -- | -- Module : Text.RegExp.Matching.Longest -- Copyright : Thomas Wilke, Frank Huch, and Sebastian Fischer diff --git a/src/Text/RegExp/Matching/Longest/Type.hs b/src/Text/RegExp/Matching/Longest/Type.hs --- a/src/Text/RegExp/Matching/Longest/Type.hs +++ b/src/Text/RegExp/Matching/Longest/Type.hs @@ -1,3 +1,5 @@ +{-# LANGUAGE MultiParamTypeClasses, FlexibleInstances #-} + module Text.RegExp.Matching.Longest.Type where import Data.Semiring diff --git a/weighted-regexp.cabal b/weighted-regexp.cabal --- a/weighted-regexp.cabal +++ b/weighted-regexp.cabal @@ -1,5 +1,5 @@ Name: weighted-regexp -Version: 0.3.0.0 +Version: 0.3.0.1 Cabal-Version: >= 1.6 Synopsis: Weighted Regular Expression Matcher Description: @@ -49,7 +49,10 @@ Executable quickcheck-re Main-Is: quickcheck.lhs - Build-Depends: base >= 3 && < 5, QuickCheck < 2 + If flag(QuickCheck) + Build-Depends: base >= 3 && < 5, QuickCheck < 2 + Else + Buildable: False HS-Source-Dirs: src Other-Modules: Text.RegExp, Text.RegExp.Matching.Leftmost, @@ -72,8 +75,6 @@ OverloadedStrings, ScopedTypeVariables GHC-Options: -fhpc -fno-warn-missing-methods -fno-warn-orphans - If !flag(QuickCheck) - Buildable: False Flag Criterion Description: Build executable to run Criterion benchmarks @@ -81,7 +82,10 @@ Executable criterion-re Main-Is: criterion.lhs - Build-Depends: base >= 3 && < 5, criterion >= 0.5 && < 0.6 + If flag(Criterion) + Build-Depends: base >= 3 && < 5, criterion >= 0.5 && < 0.6 + Else + Buildable: False HS-Source-Dirs: src Other-Modules: Text.RegExp, Text.RegExp.Matching.Leftmost, @@ -102,8 +106,6 @@ GeneralizedNewtypeDeriving, OverloadedStrings GHC-Options: - If !flag(Criterion) - Buildable: False Source-Repository head type: git