Preserve metadata when staging a new version of an annexed file.

Performance impact: When adding a large tree of new files, this needs
to do some git cat-file queries to check if any of the files already
existed and might need a metadata copy. I tried a benchmark in a copy
of my sound repository (so there was already a significant git tree
to check against.

Adding 10000 small files, with a cold cache:
  before: 1m48.539s
  after:  1m52.791s

So, impact is 0.0004 seconds per file added. Which seems acceptable, so did
not add some kind of configuration to enable/disable this.

This commit was sponsored by Lisa Feilen.

5 files changed, 42 insertions, 15 deletions

diff --git a/Annex/CatFile.hs b/Annex/CatFile.hs
index 54a4d1099..87d179a62 100644
--- a/Annex/CatFile.hs
+++ b/Annex/CatFile.hs
@@ -87,8 +87,7 @@ catKey' modeguaranteed ref mode
 		| modeguaranteed = catObject ref
 		| otherwise = L.take 8192 <$> catObject ref
 
-{- Looks up the file mode corresponding to the Ref using the running
- - cat-file.
+{- Looks up the key corresponding to the Ref using the running cat-file.
  -
  - Currently this always has to look in HEAD, because cat-file --batch
  - does not offer a way to specify that we want to look up a tree object
diff --git a/Annex/MetaData.hs b/Annex/MetaData.hs
index ef235b51f..b7850a868 100644
--- a/Annex/MetaData.hs
+++ b/Annex/MetaData.hs
@@ -11,6 +11,7 @@ import Common.Annex
 import qualified Annex
 import Types.MetaData
 import Logs.MetaData
+import Annex.CatFile
 
 import qualified Data.Set as S
 import qualified Data.Map as M
@@ -27,18 +28,27 @@ yearMetaField = MetaField "year"
 monthMetaField :: MetaField
 monthMetaField = MetaField "month"
 
-{- Generates metadata for a file that has just been ingested into the
- - annex. Passed the FileStatus of the content file.
+{- Adds metadata for a file that has just been ingested into the
+ - annex, but has not yet been committed to git.
  -
- - Does not overwrite any existing metadata values for the key.
+ - When the file has been modified, the metadata is copied over
+ - from the old key to the new key. Note that it looks at the old key as
+ - committed to HEAD -- the new key may or may not have already been staged
+ - in th annex.
+ -
+ - Also, can generate new metadata, if configured to do so.
  -}
-genMetaData :: Key -> FileStatus -> Annex ()
-genMetaData key status = whenM (annexGenMetaData <$> Annex.getGitConfig) $ do
-	metadata <- getCurrentMetaData key
-	let metadata' = genMetaData' status metadata
-	unless (metadata' == emptyMetaData) $
-		addMetaData key metadata'
-
+genMetaData :: Key -> FilePath -> FileStatus -> Annex ()
+genMetaData key file status = do
+	maybe noop (flip copyMetaData key) =<< catKeyFileHEAD file
+	whenM (annexGenMetaData <$> Annex.getGitConfig) $ do
+		metadata <- getCurrentMetaData key
+		let metadata' = genMetaData' status metadata
+		unless (metadata' == emptyMetaData) $
+			addMetaData key metadata'
+
+{- Generates metadata from the FileStatus.
+ - Does not overwrite any existing metadata values. -}
 genMetaData' :: FileStatus -> MetaData -> MetaData
 genMetaData' status old = MetaData $ M.fromList $ filter isnew
 	[ (yearMetaField, S.singleton $ toMetaValue $ show y)
diff --git a/Command/Add.hs b/Command/Add.hs
index 0906ae531..662ce4242 100644
--- a/Command/Add.hs
+++ b/Command/Add.hs
@@ -161,14 +161,14 @@ ingest (Just source) = do
 	goindirect (Just (key, _)) mcache ms = do
 		catchAnnex (moveAnnex key $ contentLocation source)
 			(undo (keyFilename source) key)
-		maybe noop (genMetaData key) ms
+		maybe noop (genMetaData key (keyFilename source)) ms
 		liftIO $ nukeFile $ keyFilename source
 		return $ (Just key, mcache)
 	goindirect _ _ _ = failure "failed to generate a key"
 
 	godirect (Just (key, _)) (Just cache) ms = do
 		addInodeCache key cache
-		maybe noop (genMetaData key) ms
+		maybe noop (genMetaData key (keyFilename source)) ms
 		finishIngestDirect key source
 		return $ (Just key, Just cache)
 	godirect _ _ _ = failure "failed to generate a key"
diff --git a/Logs/MetaData.hs b/Logs/MetaData.hs
index 63314bcef..6702c3733 100644
--- a/Logs/MetaData.hs
+++ b/Logs/MetaData.hs
@@ -28,10 +28,10 @@
 
 module Logs.MetaData (
 	getCurrentMetaData,
-	getMetaData,
 	addMetaData,
 	addMetaData',
 	currentMetaData,
+	copyMetaData,
 ) where
 
 import Common.Annex
@@ -135,3 +135,20 @@ simplifyLog s = case sl of
 	  where
 		older = value l
 		unique = older `differenceMetaData` newer
+
+{- Copies the metadata from the old key to the new key.
+ -
+ - The exact content of the metadata file is copied, so that the timestamps
+ - remain the same, and because this is more space-efficient in the git
+ - repository.
+ - 
+ - Any metadata already attached to the new key is not preserved.
+ -}
+copyMetaData :: Key -> Key -> Annex ()
+copyMetaData oldkey newkey
+	| oldkey == newkey = noop
+	| otherwise = do
+		l <- getMetaData oldkey
+		unless (S.null l) $
+			Annex.Branch.change (metaDataLogFile newkey) $
+				const $ showLog l
diff --git a/debian/changelog b/debian/changelog
index 543504c12..8c157aeb2 100644
--- a/debian/changelog
+++ b/debian/changelog
@@ -11,6 +11,7 @@ git-annex (5.20140222) UNRELEASED; urgency=medium
     tag/showname.
   * annex.genmetadata can be set to make git-annex automatically set
     metadata (year and month) when adding files.
+  * Preserve metadata when staging a new version of an annexed file.
   * metadata: Field names limited to alphanumerics and a few whitelisted
     punctuation characters to avoid issues with views, etc.
   * metadata: Support --json