Switch from using regex-compat to regex-tdfa, as the C regex library is rather buggy.

Limit.hs

@@ -13,7 +13,8 @@ import Data.Time.Clock.POSIX
 import qualified Data.Set as S
 import qualified Data.Map as M
 import System.Path.WildMatch
-import Text.Regex
+import Text.Regex.TDFA
+import Text.Regex.TDFA.String
 
 import Common.Annex
 import qualified Annex
@@ -83,12 +84,17 @@ limitExclude :: MkLimit
 limitExclude glob = Right $ const $ return . not . matchglob glob
 
 {- Could just use wildCheckCase, but this way the regex is only compiled
- - once. -}
+ - once. Also, we use regex-TDFA because it's less buggy in its support
+ - of non-unicode characters. -}
 matchglob :: String -> Annex.FileInfo -> Bool
 matchglob glob (Annex.FileInfo { Annex.matchFile = f }) =
-	isJust $ matchRegex cregex f
+	case cregex of
+		Right r -> case execute r f of
+			Right (Just _) -> True
+			_ -> False
+		Left _ -> error $ "failed to compile regex: " ++ regex
   where
-	cregex = mkRegex regex
+	cregex = compile defaultCompOpt defaultExecOpt regex
 	regex = '^':wildToRegex glob
 
 {- Adds a limit to skip files not believed to be present

Makefile

@@ -19,7 +19,7 @@ fast: dist/caballog
 	@$$(grep 'ghc --make' dist/caballog | head -n 1)
 	@ln -sf dist/build/git-annex/git-annex git-annex
 
-dist/caballog:
+dist/caballog: git-annex.cabal
 	$(CABAL) configure -f"-Production" -O0
 	$(CABAL) build -v2 | tee $@
 

debian/changelog

@@ -34,6 +34,8 @@ git-annex (4.20130228) UNRELEASED; urgency=low
     should add files in the current directory, but not act on unlocked files
     elsewhere in the tree.
   * assistant: Sync with all git remotes on startup.
+  * Switch from using regex-compat to regex-tdfa, as the C regex library
+    is rather buggy.
 
  -- Joey Hess <joeyh@debian.org>  Wed, 27 Feb 2013 23:20:40 -0400
 

debian/control

@@ -9,6 +9,7 @@ Build-Depends:
 	libghc-hslogger-dev,
 	libghc-pcre-light-dev,
 	libghc-sha-dev,
+	libghc-regex-tdfa-dev,
 	libghc-dataenc-dev,
 	libghc-utf8-string-dev,
 	libghc-hs3-dev (>= 0.5.6),

doc/bugs/git-annex_doesn__39__t_list_files_containing_ISO8859-15_characters.mdwn

@@ -33,3 +33,16 @@ git-annex 4.20130227, on Debian GNU/Linux (sid, i386).
     LC_ALL=
 </code></pre>
 
+> Tracked this back to a bug in either the C library or the haskell
+> regex-posix wrpaper around it. I'm not sure which, but I emailed the
+> maintainer of the haskell library. It just doesn't think these
+> things are characters; even `.` fails to match them! Everything should
+> match that...
+> 
+> There are apparently quite a lot of bugs on POSIX regex libraries
+> as implemented on different systems:
+> <http://www.haskell.org/haskellwiki/Regex_Posix>
+> 
+> It seemed best to jettison this dependency entirely; I've switched it to
+> haskell's pure regex-tdfa library, which works nicely. [[done]]
+> --[[Joey]]

doc/install/fromscratch.mdwn

@@ -19,7 +19,7 @@ quite a lot.
   * [DAV](http://hackage.haskell.org/package/DAV) (optional)
   * [SafeSemaphore](http://hackage.haskell.org/package/SafeSemaphore)
   * [UUID](http://hackage.haskell.org/package/uuid)
-  * [Glob](http://hackage.haskell.org/package/Glob)
+  * [regex-tdfa](http://hackage.haskell.org/package/regex-tdfa)
 * Optional haskell stuff, used by the [[assistant]] and its webapp (edit Makefile to disable)
   * [stm](http://hackage.haskell.org/package/stm)
     (version 2.3 or newer)

git-annex.cabal

@@ -70,7 +70,7 @@ Executable git-annex
    extensible-exceptions, dataenc, SHA, process, json,
    base (>= 4.5 && < 4.8), monad-control, transformers-base, lifted-base,
    IfElse, text, QuickCheck >= 2.1, bloomfilter, edit-distance, process,
-   SafeSemaphore, uuid, random, regex-compat
+   SafeSemaphore, uuid, random, regex-tdfa
   -- Need to list these because they're generated from .hsc files.
   Other-Modules: Utility.Touch Utility.Mounts
   Include-Dirs: Utility