diff --git a/Annex/VectorClock.hs b/Annex/VectorClock.hs
index 951145ca14..5e9e2cb89b 100644
--- a/Annex/VectorClock.hs
+++ b/Annex/VectorClock.hs
@@ -17,6 +17,7 @@ import Prelude
 import Utility.Env
 import Utility.TimeStamp
 import Utility.QuickCheck
+import qualified Data.Attoparsec.ByteString.Lazy as A
 
 -- | Some very old logs did not have any time stamp at all;
 -- Unknown is used for those.
@@ -44,3 +45,6 @@ formatVectorClock (VectorClock t) = show t
 
 parseVectorClock :: String -> Maybe VectorClock
 parseVectorClock t = VectorClock <$> parsePOSIXTime t
+
+vectorClockParser :: A.Parser VectorClock
+vectorClockParser = VectorClock <$> parserPOSIXTime
diff --git a/Logs/Line.hs b/Logs/Line.hs
index a7e17190e2..716ee8be87 100644
--- a/Logs/Line.hs
+++ b/Logs/Line.hs
@@ -36,6 +36,12 @@ DAMAGE.
 
 module Logs.Line where
 
+import Common
+
+import qualified Data.Attoparsec.ByteString.Lazy as A
+import Data.Attoparsec.ByteString.Char8 (isEndOfLine)
+import qualified Data.DList as D
+
 -- This is the same as Data.List.lines, with \r added.
 -- This works around some versions of git-annex which wrote \r
 -- into git-annex branch files on Windows. Those \r's sometimes
@@ -49,3 +55,27 @@ splitLines s =  cons (case break (\c -> c == '\n' || c == '\r') s of
 		_:s''   -> splitLines s''))
   where
 	cons ~(h, t) = h : t
+
+{- Applies a parser to each line of a log file.
+ -
+ - If the parser fails to parse a line, that line is skipped, instead of
+ - the overall parse failing. This is generally a good idea in case a newer
+ - version of git-annex somehow changed the format of the log file.
+ -
+ - Any combination of \r and \n are taken to be the end of the line.
+ - (Some versions of git-annex on Windows wrote \r into git-annex branch
+ - files, and multiple \r's sometimes accumulated.)
+ -
+ - The parser does not itself need to avoid parsing beyond the end of line;
+ - this is implemented only pass the content of a line to the parser.
+ -}
+parseLogLines :: A.Parser a -> A.Parser [a]
+parseLogLines parser = go D.empty
+  where
+	go dl = do
+		line <- A.takeTill isEndOfLine
+		A.skipWhile isEndOfLine
+		let dl' = case A.parseOnly parser line of
+			Left _ -> dl
+			Right v -> D.snoc dl v
+		(A.endOfInput *> return (D.toList dl')) <|> go dl'
diff --git a/Logs/Presence/Pure.hs b/Logs/Presence/Pure.hs
index 4411fa0176..776ae42018 100644
--- a/Logs/Presence/Pure.hs
+++ b/Logs/Presence/Pure.hs
@@ -15,6 +15,8 @@ import Utility.QuickCheck
 import qualified Data.Map as M
 import qualified Data.ByteString.Lazy as L
 import qualified Data.ByteString as S
+import qualified Data.Attoparsec.ByteString.Lazy as A
+import Data.Attoparsec.ByteString.Char8 (char, anyChar)
 import Data.ByteString.Builder
 
 newtype LogInfo = LogInfo { fromLogInfo :: S.ByteString }
@@ -32,17 +34,25 @@ instance Show LogLine where
 data LogStatus = InfoPresent | InfoMissing | InfoDead
 	deriving (Eq, Show, Bounded, Enum)
 
-{- Parses a log file. Unparseable lines are ignored. -}
 parseLog :: L.ByteString -> [LogLine]
-parseLog = mapMaybe parseline . splitLines . decodeBL
-  where
-	parseline l = LogLine
-		<$> parseVectorClock c
-		<*> parseStatus s
-		<*> pure (LogInfo (encodeBS rest))
-	  where
-		(c, pastc) = separate (== ' ') l
-		(s, rest) = separate (== ' ') pastc
+parseLog = fromMaybe [] . A.maybeResult . A.parse (logParser <* A.endOfInput)
+
+logParser :: A.Parser [LogLine]
+logParser = parseLogLines $ LogLine
+	<$> vectorClockParser
+	<* char ' '
+	<*> statusParser
+	<* char ' '
+	<*> (LogInfo <$> A.takeByteString)
+
+statusParser :: A.Parser LogStatus
+statusParser = do
+	c <- anyChar
+	case c of
+		'1' -> return InfoPresent
+		'0' -> return InfoMissing
+		'X' -> return InfoDead
+		_ -> fail "unknown status character"
 
 parseStatus :: String -> Maybe LogStatus
 parseStatus "1" = Just InfoPresent