run git status before enabling clean filter

Avoids annex.largefiles inconsitency and also avoids a lot of
unneccessary calls to the clean filter when a large repo's clone
is being initialized.

This commit was supported by the NSF-funded DataLad project.

Annex/Init.hs

@@ -35,6 +35,7 @@ import Annex.UUID
 import Annex.Link
 import Annex.WorkTree
 import Config
+import Config.Smudge
 import Annex.Direct
 import Annex.AdjustedBranch
 import Annex.Environment

Config.hs

@@ -102,21 +102,3 @@ setCrippledFileSystem :: Bool -> Annex ()
 setCrippledFileSystem b = do
 	setConfig (annexConfig "crippledfilesystem") (Git.Config.boolConfig b)
 	Annex.changeGitConfig $ \c -> c { annexCrippledFileSystem = b }
-
-configureSmudgeFilter :: Annex ()
-configureSmudgeFilter = do
-	setConfig (ConfigKey "filter.annex.smudge") "git-annex smudge %f"
-	setConfig (ConfigKey "filter.annex.clean") "git-annex smudge --clean %f"
-	lf <- Annex.fromRepo Git.attributesLocal
-	gf <- Annex.fromRepo Git.attributes
-	lfs <- readattr lf
-	gfs <- readattr gf
-	liftIO $ unless ("filter=annex" `isInfixOf` (lfs ++ gfs)) $ do
-		createDirectoryIfMissing True (takeDirectory lf)
-		writeFile lf (lfs ++ "\n" ++ stdattr)
-  where
-	readattr = liftIO . catchDefaultIO "" . readFileStrict
-	stdattr = unlines
-		[ "* filter=annex"
-		, ".* !filter"
-		]

Config/Smudge.hs

@@ -0,0 +1,41 @@
+{- Git smudge filter configuration
+ -
+ - Copyright 2011-2018 Joey Hess <id@joeyh.name>
+ -
+ - Licensed under the GNU GPL version 3 or higher.
+ -}
+
+module Config.Smudge where
+
+import Annex.Common
+import qualified Annex
+import qualified Git
+import qualified Git.Command
+import Config
+
+configureSmudgeFilter :: Annex ()
+configureSmudgeFilter = do
+	-- If this is run in a newly cloned repository, git may not have
+	-- cached file information in the index yet, and so after
+	-- configuring the clean filter, the next git status would want to
+	-- run it on every file. That is expensive and can also result in
+	-- unexpected changes when the file is checked into git or annex
+	-- counter to the annex.largefiles configuration.
+	-- Avoid that problem by running git status now.
+	inRepo $ Git.Command.runQuiet [Param "status", Param "--porcelain"]
+
+	setConfig (ConfigKey "filter.annex.smudge") "git-annex smudge %f"
+	setConfig (ConfigKey "filter.annex.clean") "git-annex smudge --clean %f"
+	lf <- Annex.fromRepo Git.attributesLocal
+	gf <- Annex.fromRepo Git.attributes
+	lfs <- readattr lf
+	gfs <- readattr gf
+	liftIO $ unless ("filter=annex" `isInfixOf` (lfs ++ gfs)) $ do
+		createDirectoryIfMissing True (takeDirectory lf)
+		writeFile lf (lfs ++ "\n" ++ stdattr)
+  where
+	readattr = liftIO . catchDefaultIO "" . readFileStrict
+	stdattr = unlines
+		[ "* filter=annex"
+		, ".* !filter"
+		]

Upgrade/V5.hs

@@ -10,6 +10,7 @@ module Upgrade.V5 where
 import Annex.Common
 import qualified Annex
 import Config
+import Config.Smudge
 import Annex.InodeSentinal
 import Annex.Link
 import Annex.Direct

doc/bugs/Too_difficult_if_not_impossible_to_explicitly_add__47__keep_file_under_git___40__not_annex__41___in_v6_without_employing_.gitattributes/comment_13_5f11262639d60ef63a44de09b191e0b6._comment

@@ -0,0 +1,21 @@
+[[!comment format=mdwn
+ username="joey"
+ subject="""comment 13"""
+ date="2018-08-28T14:02:37Z"
+ content="""
+However, that leaves the case where .gitattributes configures
+annex.largefiles, but that's been overridden for a file to add it to git,
+and then the repo is cloned and initted with --version=6 (or upgraded).
+
+Turns out that calling git status before enabling the smudge filter
+prevents git from getting confused about the file being modified in this
+case.
+
+In the fresh clone, git has not populated the index with stat info
+yet, and so it later runs the clean filter on the file, and that
+respects the largefiles configuration, so the way the file is
+stored in git is not taken into account.
+
+Worked around this by adding a `git status` call to the v6
+initialization/upgrade.
+"""]]

git-annex.cabal

@@ -809,6 +809,7 @@ Executable git-annex
     Config.Files
     Config.DynamicConfig
     Config.GitConfig
+    Config.Smudge
     Creds
     Crypto
     Database.Export