metadata: Fix encoding problem that led to mojibake when storing metadata strings that contained both unicode characters and a space (or '!') character.

The fix is to stop using w82s, which does not properly reconstitute unicode
strings. Instrad, use utf8 bytestring to get the [Word8] to base64. This
passes unicode through perfectly, including any invalid filesystem encoded
characters.

Note that toB64 / fromB64 are also used for creds and cipher
embedding. It would be unfortunate if this change broke those uses.

For cipher embedding, note that ciphers can contain arbitrary bytes (should
really be using ByteString.Char8 there). Testing indicated it's not safe to
use the new fromB64 there; I think that characters were incorrectly
combined.

For credpair embedding, the username or password could contain unicode.
Before, that unicode would fail to round-trip through the b64.
So, I guess this is not going to break any embedded creds that worked
before.

This bug may have affected some creds before, and if so,
this change will not fix old ones, but should fix new ones at least.

Remote/Helper/Encryptable.hs

@@ -20,13 +20,14 @@ module Remote.Helper.Encryptable (
 ) where
 
 import qualified Data.Map as M
+import qualified "dataenc" Codec.Binary.Base64 as B64
+import Data.Bits.Utils
 
 import Common.Annex
 import Types.Remote
 import Crypto
 import Types.Crypto
 import qualified Annex
-import Utility.Base64
 
 -- Used to ensure that encryption has been set up before trying to
 -- eg, store creds in the remote config that would need to use the
@@ -137,9 +138,9 @@ cipherKey c = fmap make <$> remoteCipher c
 
 {- Stores an StorableCipher in a remote's configuration. -}
 storeCipher :: RemoteConfig -> StorableCipher -> RemoteConfig
-storeCipher c (SharedCipher t) = M.insert "cipher" (toB64 t) c
+storeCipher c (SharedCipher t) = M.insert "cipher" (toB64bs t) c
 storeCipher c (EncryptedCipher t _ ks) =
-	M.insert "cipher" (toB64 t) $ M.insert "cipherkeys" (showkeys ks) c
+	M.insert "cipher" (toB64bs t) $ M.insert "cipherkeys" (showkeys ks) c
   where
 	showkeys (KeyIds l) = intercalate "," l
 
@@ -149,11 +150,11 @@ extractCipher c = case (M.lookup "cipher" c,
 			M.lookup "cipherkeys" c,
 			M.lookup "encryption" c) of
 	(Just t, Just ks, encryption) | maybe True (== "hybrid") encryption ->
-		Just $ EncryptedCipher (fromB64 t) Hybrid (readkeys ks)
+		Just $ EncryptedCipher (fromB64bs t) Hybrid (readkeys ks)
 	(Just t, Just ks, Just "pubkey") ->
-		Just $ EncryptedCipher (fromB64 t) PubKey (readkeys ks)
+		Just $ EncryptedCipher (fromB64bs t) PubKey (readkeys ks)
 	(Just t, Nothing, encryption) | maybe True (== "shared") encryption ->
-		Just $ SharedCipher (fromB64 t)
+		Just $ SharedCipher (fromB64bs t)
 	_ -> Nothing
   where
 	readkeys = KeyIds . split ","
@@ -169,3 +170,14 @@ describeEncryption c = case extractCipher c of
 			PubKey -> Nothing
 			Hybrid -> Just "(hybrid mode)"
 		]
+
+{- Not using Utility.Base64 because these "Strings" are really
+ - bags of bytes and that would convert to unicode and not roung-trip
+ - cleanly. -}
+toB64bs :: String -> String
+toB64bs = B64.encode . s2w8
+
+fromB64bs :: String -> String
+fromB64bs s = fromMaybe bad $ w82s <$> B64.decode s
+  where
+	bad = error "bad base64 encoded data"

Test.hs

@@ -71,6 +71,7 @@ import qualified Utility.Hash
 import qualified Utility.Scheduled
 import qualified Utility.HumanTime
 import qualified Utility.ThreadScheduler
+import qualified Utility.Base64
 import qualified Command.Uninit
 import qualified CmdLine.GitAnnex as GitAnnex
 #ifndef mingw32_HOST_OS
@@ -163,6 +164,7 @@ properties = localOption (QuickCheckTests 1000) $ testGroup "QuickCheck"
 	, testProperty "prop_branchView_legal" Logs.View.prop_branchView_legal
 	, testProperty "prop_view_roundtrips" Annex.View.prop_view_roundtrips
 	, testProperty "prop_viewedFile_rountrips" Annex.View.ViewedFile.prop_viewedFile_roundtrips
+	, testProperty "prop_b64_roundtrips" Utility.Base64.prop_b64_roundtrips
 	]
 
 {- These tests set up the test environment, but also test some basic parts

Types/MetaData.hs

@@ -224,6 +224,7 @@ data ModMeta
 	| DelMeta MetaField MetaValue
 	| SetMeta MetaField MetaValue -- removes any existing values
 	| MaybeSetMeta MetaField MetaValue -- when field has no existing value
+	deriving (Show)
 
 {- Applies a ModMeta, generating the new MetaData.
  - Note that the new MetaData does not include all the 

Utility/Base64.hs

@@ -1,24 +1,28 @@
-{- Simple Base64 access
+{- Simple Base64 encoding of Strings
  -
  - Copyright 2011 Joey Hess <id@joeyh.name>
  -
  - License: BSD-2-clause
  -}
 
-module Utility.Base64 (toB64, fromB64Maybe, fromB64) where
+module Utility.Base64 (toB64, fromB64Maybe, fromB64, prop_b64_roundtrips) where
 
-import "dataenc" Codec.Binary.Base64
+import qualified "dataenc" Codec.Binary.Base64 as B64
-import Data.Bits.Utils
 import Control.Applicative
 import Data.Maybe
+import qualified Data.ByteString.Lazy as L
+import Data.ByteString.Lazy.UTF8 (fromString, toString)
 
 toB64 :: String -> String	
-toB64 = encode . s2w8
+toB64 = B64.encode . L.unpack . fromString
 
 fromB64Maybe :: String -> Maybe String
-fromB64Maybe s = w82s <$> decode s
+fromB64Maybe s = toString . L.pack <$> B64.decode s
 
 fromB64 :: String -> String
 fromB64 = fromMaybe bad . fromB64Maybe
   where
 	bad = error "bad base64 encoded data"
+
+prop_b64_roundtrips :: String -> Bool
+prop_b64_roundtrips s = s == fromB64 (toB64 s)

Utility/FileSystemEncoding.hs

@@ -14,6 +14,8 @@ module Utility.FileSystemEncoding (
 	decodeBS,
 	decodeW8,
 	encodeW8,
+	encodeW8NUL,
+	decodeW8NUL,
 	truncateFilePath,
 ) where
 
@@ -25,6 +27,7 @@ import System.IO.Unsafe
 import qualified Data.Hash.MD5 as MD5
 import Data.Word
 import Data.Bits.Utils
+import Data.List.Utils
 import qualified Data.ByteString.Lazy as L
 #ifdef mingw32_HOST_OS
 import qualified Data.ByteString.Lazy.UTF8 as L8
@@ -89,6 +92,9 @@ decodeBS = L8.toString
  - w82c produces a String, which may contain Chars that are invalid
  - unicode. From there, this is really a simple matter of applying the
  - file system encoding, only complicated by GHC's interface to doing so.
+ -
+ - Note that the encoding stops at any NUL in the input. FilePaths
+ - do not normally contain embedded NUL, but Haskell Strings may.
  -}
 {-# NOINLINE encodeW8 #-}
 encodeW8 :: [Word8] -> FilePath
@@ -101,6 +107,17 @@ encodeW8 w8 = unsafePerformIO $ do
 decodeW8 :: FilePath -> [Word8]
 decodeW8 = s2w8 . _encodeFilePath
 
+{- Like encodeW8 and decodeW8, but NULs are passed through unchanged. -}
+encodeW8NUL :: [Word8] -> FilePath
+encodeW8NUL = join nul . map encodeW8 . split (s2w8 nul)
+  where
+	nul = ['\NUL']
+
+decodeW8NUL :: FilePath -> [Word8]
+decodeW8NUL = join (s2w8 nul) . map decodeW8 . split nul
+  where
+	nul = ['\NUL']
+
 {- Truncates a FilePath to the given number of bytes (or less),
  - as represented on disk.
  -

debian/changelog

@@ -22,6 +22,11 @@ git-annex (5.2015022) UNRELEASED; urgency=medium
   * When re-execing git-annex, use current program location, rather than
     ~/.config/git-annex/program, when possible.
   * Submodules are now supported by git-annex!
+  * metadata: Fix encoding problem that led to mojibake when storing
+    metadata strings that contained both unicode characters and a space
+    (or '!') character.
+  * Also potentially fixes encoding problem when embedding credentials
+    that contain unicode characters.
 
  -- Joey Hess <id@joeyh.name>  Thu, 19 Feb 2015 14:16:03 -0400
 

doc/bugs/Unicode_characters_lost__47__converted_in_metadata.mdwn

@@ -13,3 +13,5 @@ Unicode characters in metadata are pruned/converted/lost:
 ### What version of git-annex are you using? On what operating system?
 
 5.20141125 Debian
+
+> [[fixed|done]]; test pass. --[[Joey]]

doc/bugs/Unicode_characters_lost__47__converted_in_metadata/comment_1_799fd2dc62dc51a6a8fa8a6d15d7a1f9._comment

@@ -0,0 +1,32 @@
+[[!comment format=mdwn
+ username="joey"
+ subject="""comment 1"""
+ date="2015-03-04T14:31:21Z"
+ content="""
+What I'm seeing is the unicode arrow is replaced with 0092 and the elipsis
+with &. It's losing the other byte.
+
+The problem seems to be in the base64 encoding that's done, when the metadata
+value contains spaces or a few other problem characters. These same
+unicode characters roundtrip through without a problem when not embedded
+in a string with spaces.
+
+<pre>
+*Utility.Base64> let s = "…"
+*Utility.Base64> (s, fromB64 $ toB64 s)
+("\8230","&")
+</pre>
+
+git-annex also uses base64 for encoding some creds
+(and for tagged pushes over XMPP, but only the JID is encoded).
+
+The real culprit is the use of `w82s`, which doesn't handle multi-byte
+characters. I can easily fix this by using `encodeW8` instead.
+Audited git-annex for other problem w82s uses and don't see any, so will
+only need to fix this once.
+
+Added a quickcheck test for fromB64 . toB64 roundtripping.
+
+Unfortunately, the entered unicode characters didn't get saved right,
+so git-annex can do nothing to fix data that was already entered.
+"""]]