2010-11-15 20:35:06 +00:00
|
|
|
{- git-annex command
|
|
|
|
-
|
2012-03-11 21:15:58 +00:00
|
|
|
- Copyright 2010-2012 Joey Hess <joey@kitenet.net>
|
2010-11-15 20:35:06 +00:00
|
|
|
-
|
|
|
|
- Licensed under the GNU GPL version 3 or higher.
|
|
|
|
-}
|
|
|
|
|
2011-08-31 23:13:02 +00:00
|
|
|
{-# LANGUAGE BangPatterns #-}
|
|
|
|
|
2010-11-15 20:35:06 +00:00
|
|
|
module Command.Unused where
|
|
|
|
|
2011-01-30 05:41:15 +00:00
|
|
|
import qualified Data.Set as S
|
2012-02-01 20:05:02 +00:00
|
|
|
import qualified Data.Text.Lazy as L
|
|
|
|
import qualified Data.Text.Lazy.Encoding as L
|
2012-03-12 02:00:49 +00:00
|
|
|
import Data.BloomFilter
|
|
|
|
import Data.BloomFilter.Easy
|
|
|
|
import Data.BloomFilter.Hash
|
2012-03-12 18:09:43 +00:00
|
|
|
import Control.Monad.ST
|
2010-11-15 20:35:06 +00:00
|
|
|
|
2011-10-05 20:02:51 +00:00
|
|
|
import Common.Annex
|
2010-11-15 20:35:06 +00:00
|
|
|
import Command
|
2012-05-02 18:59:05 +00:00
|
|
|
import Logs.Unused
|
2011-10-04 04:40:47 +00:00
|
|
|
import Annex.Content
|
2011-09-28 20:43:10 +00:00
|
|
|
import Utility.FileMode
|
2011-10-15 20:21:08 +00:00
|
|
|
import Logs.Location
|
2012-03-22 04:23:15 +00:00
|
|
|
import Config
|
2010-11-15 22:04:19 +00:00
|
|
|
import qualified Annex
|
2011-06-30 17:16:57 +00:00
|
|
|
import qualified Git
|
2011-12-14 19:56:11 +00:00
|
|
|
import qualified Git.Command
|
2011-12-12 22:23:24 +00:00
|
|
|
import qualified Git.Ref
|
2011-06-30 17:16:57 +00:00
|
|
|
import qualified Git.LsFiles as LsFiles
|
2011-09-28 20:43:10 +00:00
|
|
|
import qualified Git.LsTree as LsTree
|
2011-01-16 20:05:05 +00:00
|
|
|
import qualified Backend
|
2011-04-03 00:59:41 +00:00
|
|
|
import qualified Remote
|
2011-10-04 04:40:47 +00:00
|
|
|
import qualified Annex.Branch
|
2012-01-06 14:14:37 +00:00
|
|
|
import qualified Option
|
2011-10-04 04:40:47 +00:00
|
|
|
import Annex.CatFile
|
2010-11-15 20:35:06 +00:00
|
|
|
|
2011-10-29 19:19:05 +00:00
|
|
|
def :: [Command]
|
more command-specific options
Made --from and --to command-specific options.
Added generic storage for values of command-specific options,
which allows removing some of the special case fields in AnnexState.
(Also added generic storage for command-specific flags, although there are
not yet any.)
Note that this storage uses a Map, so repeatedly looking up the same value
is slightly more expensive than looking up an AnnexState field. But, the
value can be looked up once in the seek stage, transformed as necessary,
and passed in a closure to the start stage, and this avoids that overhead.
Still, I'm hesitant to use this for things like force or fast flags.
It's probably best to reserve it for flags that are only used by a few
commands, or options like --from and --to that it's important only be
allowed to be used with commands that implement them, to avoid user
confusion.
2012-01-06 07:06:25 +00:00
|
|
|
def = [withOptions [fromOption] $ command "unused" paramNothing seek
|
2011-03-19 22:58:49 +00:00
|
|
|
"look for unused file content"]
|
2010-12-30 19:06:26 +00:00
|
|
|
|
more command-specific options
Made --from and --to command-specific options.
Added generic storage for values of command-specific options,
which allows removing some of the special case fields in AnnexState.
(Also added generic storage for command-specific flags, although there are
not yet any.)
Note that this storage uses a Map, so repeatedly looking up the same value
is slightly more expensive than looking up an AnnexState field. But, the
value can be looked up once in the seek stage, transformed as necessary,
and passed in a closure to the start stage, and this avoids that overhead.
Still, I'm hesitant to use this for things like force or fast flags.
It's probably best to reserve it for flags that are only used by a few
commands, or options like --from and --to that it's important only be
allowed to be used with commands that implement them, to avoid user
confusion.
2012-01-06 07:06:25 +00:00
|
|
|
fromOption :: Option
|
2012-01-06 14:14:37 +00:00
|
|
|
fromOption = Option.field ['f'] "from" paramRemote "remote to check for unused content"
|
more command-specific options
Made --from and --to command-specific options.
Added generic storage for values of command-specific options,
which allows removing some of the special case fields in AnnexState.
(Also added generic storage for command-specific flags, although there are
not yet any.)
Note that this storage uses a Map, so repeatedly looking up the same value
is slightly more expensive than looking up an AnnexState field. But, the
value can be looked up once in the seek stage, transformed as necessary,
and passed in a closure to the start stage, and this avoids that overhead.
Still, I'm hesitant to use this for things like force or fast flags.
It's probably best to reserve it for flags that are only used by a few
commands, or options like --from and --to that it's important only be
allowed to be used with commands that implement them, to avoid user
confusion.
2012-01-06 07:06:25 +00:00
|
|
|
|
2010-12-30 18:19:16 +00:00
|
|
|
seek :: [CommandSeek]
|
2012-02-16 04:41:30 +00:00
|
|
|
seek = [withNothing start]
|
2010-11-15 20:35:06 +00:00
|
|
|
|
|
|
|
{- Finds unused content in the annex. -}
|
2011-09-15 20:50:49 +00:00
|
|
|
start :: CommandStart
|
2011-10-29 23:16:45 +00:00
|
|
|
start = do
|
2012-01-06 14:14:37 +00:00
|
|
|
from <- Annex.getField $ Option.name fromOption
|
2011-05-29 02:37:17 +00:00
|
|
|
let (name, action) = case from of
|
|
|
|
Nothing -> (".", checkUnused)
|
|
|
|
Just "." -> (".", checkUnused)
|
2012-03-02 02:35:10 +00:00
|
|
|
Just "here" -> (".", checkUnused)
|
2011-05-29 02:37:17 +00:00
|
|
|
Just n -> (n, checkRemoteUnused n)
|
|
|
|
showStart "unused" name
|
|
|
|
next action
|
2010-11-15 20:35:06 +00:00
|
|
|
|
2011-05-29 02:20:22 +00:00
|
|
|
checkUnused :: CommandPerform
|
2012-03-12 01:08:48 +00:00
|
|
|
checkUnused = chain 0
|
|
|
|
[ check "" unusedMsg $ findunused =<< Annex.getState Annex.fast
|
|
|
|
, check "bad" staleBadMsg $ staleKeysPrune gitAnnexBadDir
|
|
|
|
, check "tmp" staleTmpMsg $ staleKeysPrune gitAnnexTmpDir
|
|
|
|
]
|
2011-04-29 17:59:00 +00:00
|
|
|
where
|
2012-03-11 16:38:59 +00:00
|
|
|
findunused True = do
|
|
|
|
showNote "fast mode enabled; only finding stale files"
|
|
|
|
return []
|
|
|
|
findunused False = do
|
|
|
|
showAction "checking for unused data"
|
2012-03-12 19:21:20 +00:00
|
|
|
excludeReferenced =<< getKeysPresent
|
2012-03-12 01:08:48 +00:00
|
|
|
chain _ [] = next $ return True
|
|
|
|
chain v (a:as) = do
|
|
|
|
v' <- a v
|
|
|
|
chain v' as
|
2011-01-28 18:10:50 +00:00
|
|
|
|
2011-05-29 02:20:22 +00:00
|
|
|
checkRemoteUnused :: String -> CommandPerform
|
2012-03-12 01:08:48 +00:00
|
|
|
checkRemoteUnused name = go =<< fromJust <$> Remote.byName (Just name)
|
|
|
|
where
|
|
|
|
go r = do
|
|
|
|
showAction "checking for unused data"
|
|
|
|
_ <- check "" (remoteUnusedMsg r) (remoteunused r) 0
|
|
|
|
next $ return True
|
|
|
|
remoteunused r =
|
|
|
|
excludeReferenced =<< loggedKeysFor (Remote.uuid r)
|
|
|
|
|
|
|
|
check :: FilePath -> ([(Int, Key)] -> String) -> Annex [Key] -> Int -> Annex Int
|
|
|
|
check file msg a c = do
|
|
|
|
l <- a
|
|
|
|
let unusedlist = number c l
|
|
|
|
unless (null l) $ showLongNote $ msg unusedlist
|
2012-05-02 18:59:05 +00:00
|
|
|
writeUnusedLog file unusedlist
|
2012-03-12 01:08:48 +00:00
|
|
|
return $ c + length l
|
2011-05-15 06:49:43 +00:00
|
|
|
|
2012-03-12 01:08:48 +00:00
|
|
|
number :: Int -> [a] -> [(Int, a)]
|
|
|
|
number _ [] = []
|
|
|
|
number n (x:xs) = (n+1, x) : number (n+1) xs
|
2011-04-03 00:59:41 +00:00
|
|
|
|
|
|
|
table :: [(Int, Key)] -> [String]
|
2011-07-15 16:47:14 +00:00
|
|
|
table l = " NUMBER KEY" : map cols l
|
2010-11-15 20:35:06 +00:00
|
|
|
where
|
2011-03-12 19:30:17 +00:00
|
|
|
cols (n,k) = " " ++ pad 6 (show n) ++ " " ++ show k
|
2010-11-15 22:04:19 +00:00
|
|
|
pad n s = s ++ replicate (n - length s) ' '
|
|
|
|
|
2011-04-03 00:59:41 +00:00
|
|
|
staleTmpMsg :: [(Int, Key)] -> String
|
|
|
|
staleTmpMsg t = unlines $
|
|
|
|
["Some partially transferred data exists in temporary files:"]
|
|
|
|
++ table t ++ [dropMsg Nothing]
|
2011-04-29 17:59:00 +00:00
|
|
|
|
|
|
|
staleBadMsg :: [(Int, Key)] -> String
|
|
|
|
staleBadMsg t = unlines $
|
|
|
|
["Some corrupted files have been preserved by fsck, just in case:"]
|
|
|
|
++ table t ++ [dropMsg Nothing]
|
|
|
|
|
2011-04-03 00:59:41 +00:00
|
|
|
unusedMsg :: [(Int, Key)] -> String
|
|
|
|
unusedMsg u = unusedMsg' u
|
2011-09-23 22:04:38 +00:00
|
|
|
["Some annexed data is no longer used by any files:"]
|
|
|
|
[dropMsg Nothing]
|
2011-04-03 00:59:41 +00:00
|
|
|
unusedMsg' :: [(Int, Key)] -> [String] -> [String] -> String
|
|
|
|
unusedMsg' u header trailer = unlines $
|
|
|
|
header ++
|
|
|
|
table u ++
|
|
|
|
["(To see where data was previously used, try: git log --stat -S'KEY')"] ++
|
|
|
|
trailer
|
|
|
|
|
2011-12-31 08:11:39 +00:00
|
|
|
remoteUnusedMsg :: Remote -> [(Int, Key)] -> String
|
2011-09-23 22:04:38 +00:00
|
|
|
remoteUnusedMsg r u = unusedMsg' u
|
|
|
|
["Some annexed data on " ++ name ++ " is not used by any files:"]
|
|
|
|
[dropMsg $ Just r]
|
|
|
|
where
|
|
|
|
name = Remote.name r
|
|
|
|
|
2011-12-31 08:11:39 +00:00
|
|
|
dropMsg :: Maybe Remote -> String
|
2011-04-03 00:59:41 +00:00
|
|
|
dropMsg Nothing = dropMsg' ""
|
|
|
|
dropMsg (Just r) = dropMsg' $ " --from " ++ Remote.name r
|
|
|
|
dropMsg' :: String -> String
|
2011-06-23 16:23:25 +00:00
|
|
|
dropMsg' s = "\nTo remove unwanted data: git-annex dropunused" ++ s ++ " NUMBER\n"
|
2011-04-03 00:59:41 +00:00
|
|
|
|
2012-03-12 19:21:20 +00:00
|
|
|
{- Finds keys in the list that are not referenced in the git repository.
|
|
|
|
-
|
|
|
|
- Strategy:
|
|
|
|
-
|
|
|
|
- * Build a bloom filter of all keys referenced by symlinks. This
|
|
|
|
- is the fastest one to build and will filter out most keys.
|
|
|
|
- * If keys remain, build a second bloom filter of keys referenced by
|
|
|
|
- all branches.
|
|
|
|
- * The list is streamed through these bloom filters lazily, so both will
|
|
|
|
- exist at the same time. This means that twice the memory is used,
|
|
|
|
- but they're relatively small, so the added complexity of using a
|
|
|
|
- mutable bloom filter does not seem worthwhile.
|
|
|
|
- * Generating the second bloom filter can take quite a while, since
|
|
|
|
- it needs enumerating all keys in all git branches. But, the common
|
|
|
|
- case, if the second filter is needed, is for some keys to be globally
|
|
|
|
- unused, and in that case, no short-circuit is possible.
|
|
|
|
- Short-circuiting if the first filter filters all the keys handles the
|
|
|
|
- other common case.
|
|
|
|
-}
|
2011-09-28 21:35:47 +00:00
|
|
|
excludeReferenced :: [Key] -> Annex [Key]
|
2012-03-12 19:21:20 +00:00
|
|
|
excludeReferenced ks = runfilter firstlevel ks >>= runfilter secondlevel
|
2011-09-28 21:35:47 +00:00
|
|
|
where
|
2012-03-12 19:21:20 +00:00
|
|
|
runfilter _ [] = return [] -- optimisation
|
|
|
|
runfilter a l = bloomFilter show l <$> genBloomFilter show a
|
|
|
|
firstlevel = withKeysReferencedM
|
|
|
|
secondlevel = withKeysReferencedInGit
|
2011-09-28 21:35:47 +00:00
|
|
|
|
2011-04-03 00:59:41 +00:00
|
|
|
{- Finds items in the first, smaller list, that are not
|
|
|
|
- present in the second, larger list.
|
|
|
|
-
|
|
|
|
- Constructing a single set, of the list that tends to be
|
|
|
|
- smaller, appears more efficient in both memory and CPU
|
|
|
|
- than constructing and taking the S.difference of two sets. -}
|
|
|
|
exclude :: Ord a => [a] -> [a] -> [a]
|
|
|
|
exclude [] _ = [] -- optimisation
|
|
|
|
exclude smaller larger = S.toList $ remove larger $ S.fromList smaller
|
2010-11-15 20:35:06 +00:00
|
|
|
where
|
2011-01-30 05:41:15 +00:00
|
|
|
remove a b = foldl (flip S.delete) b a
|
2011-01-16 20:05:05 +00:00
|
|
|
|
2012-03-12 20:18:14 +00:00
|
|
|
{- A bloom filter capable of holding half a million keys with a
|
|
|
|
- false positive rate of 1 in 1000 uses around 8 mb of memory,
|
|
|
|
- so will easily fit on even my lowest memory systems.
|
|
|
|
-}
|
|
|
|
bloomCapacity :: Annex Int
|
|
|
|
bloomCapacity = fromMaybe 500000 . readish
|
2012-05-06 00:15:32 +00:00
|
|
|
<$> getConfig (annexConfig "bloomcapacity") ""
|
2012-03-12 20:18:14 +00:00
|
|
|
bloomAccuracy :: Annex Int
|
|
|
|
bloomAccuracy = fromMaybe 1000 . readish
|
2012-05-06 00:15:32 +00:00
|
|
|
<$> getConfig (annexConfig "bloomaccuracy") ""
|
2012-03-12 20:18:14 +00:00
|
|
|
bloomBitsHashes :: Annex (Int, Int)
|
|
|
|
bloomBitsHashes = do
|
|
|
|
capacity <- bloomCapacity
|
|
|
|
accuracy <- bloomAccuracy
|
|
|
|
return $ suggestSizing capacity (1/ fromIntegral accuracy)
|
|
|
|
|
2012-03-12 19:21:20 +00:00
|
|
|
{- Creates a bloom filter, and runs an action, such as withKeysReferenced,
|
2012-03-12 18:09:43 +00:00
|
|
|
- to populate it.
|
|
|
|
-
|
|
|
|
- The action is passed a callback that it can use to feed values into the
|
|
|
|
- bloom filter.
|
|
|
|
-
|
|
|
|
- Once the action completes, the mutable filter is frozen
|
|
|
|
- for later use.
|
|
|
|
-}
|
2012-03-12 19:21:20 +00:00
|
|
|
genBloomFilter :: Hashable t => (v -> t) -> ((v -> Annex ()) -> Annex b) -> Annex (Bloom t)
|
2012-03-12 18:09:43 +00:00
|
|
|
genBloomFilter convert populate = do
|
2012-03-12 20:18:14 +00:00
|
|
|
(numbits, numhashes) <- bloomBitsHashes
|
2012-03-12 18:09:43 +00:00
|
|
|
bloom <- lift $ newMB (cheapHashes numhashes) numbits
|
2012-03-12 19:21:20 +00:00
|
|
|
_ <- populate $ \v -> lift $ insertMB bloom (convert v)
|
2012-03-12 18:09:43 +00:00
|
|
|
lift $ unsafeFreezeMB bloom
|
|
|
|
where
|
|
|
|
lift = liftIO . stToIO
|
|
|
|
|
2012-03-12 19:21:20 +00:00
|
|
|
bloomFilter :: Hashable t => (v -> t) -> [v] -> Bloom t -> [v]
|
|
|
|
bloomFilter convert l bloom = filter (\k -> convert k `notElemB` bloom) l
|
2012-03-12 18:09:43 +00:00
|
|
|
|
2012-03-12 19:21:20 +00:00
|
|
|
{- Given an initial value, folds it with each key referenced by
|
|
|
|
- symlinks in the git repo. -}
|
2012-03-11 19:19:07 +00:00
|
|
|
withKeysReferenced :: v -> (Key -> v -> v) -> Annex v
|
2012-03-12 19:21:20 +00:00
|
|
|
withKeysReferenced initial a = withKeysReferenced' initial folda
|
|
|
|
where
|
|
|
|
folda k v = return $ a k v
|
|
|
|
|
|
|
|
{- Runs an action on each referenced key in the git repo. -}
|
|
|
|
withKeysReferencedM :: (Key -> Annex ()) -> Annex ()
|
|
|
|
withKeysReferencedM a = withKeysReferenced' () calla
|
2012-03-12 18:09:43 +00:00
|
|
|
where
|
2012-03-12 19:21:20 +00:00
|
|
|
calla k _ = a k
|
|
|
|
|
2012-03-12 18:09:43 +00:00
|
|
|
withKeysReferenced' :: v -> (Key -> v -> Annex v) -> Annex v
|
|
|
|
withKeysReferenced' initial a = go initial =<< files
|
2012-03-11 19:19:07 +00:00
|
|
|
where
|
2012-03-11 21:15:58 +00:00
|
|
|
files = do
|
|
|
|
top <- fromRepo Git.workTree
|
|
|
|
inRepo $ LsFiles.inRepo [top]
|
2012-03-11 19:19:07 +00:00
|
|
|
go v [] = return v
|
|
|
|
go v (f:fs) = do
|
|
|
|
x <- Backend.lookupFile f
|
|
|
|
case x of
|
|
|
|
Nothing -> go v fs
|
|
|
|
Just (k, _) -> do
|
2012-03-12 18:09:43 +00:00
|
|
|
!v' <- a k v
|
2012-03-11 19:19:07 +00:00
|
|
|
go v' fs
|
|
|
|
|
2012-03-12 19:21:20 +00:00
|
|
|
|
|
|
|
withKeysReferencedInGit :: (Key -> Annex ()) -> Annex ()
|
|
|
|
withKeysReferencedInGit a = do
|
|
|
|
rs <- relevantrefs <$> showref
|
|
|
|
forM_ rs (withKeysReferencedInGitRef a)
|
|
|
|
where
|
|
|
|
showref = inRepo $ Git.Command.pipeRead [Param "show-ref"]
|
|
|
|
relevantrefs = map (Git.Ref . snd) .
|
|
|
|
nubBy uniqref .
|
|
|
|
filter ourbranches .
|
|
|
|
map (separate (== ' ')) . lines
|
|
|
|
uniqref (x, _) (y, _) = x == y
|
|
|
|
ourbranchend = '/' : show Annex.Branch.name
|
|
|
|
ourbranches (_, b) = not $ ourbranchend `isSuffixOf` b
|
|
|
|
|
|
|
|
withKeysReferencedInGitRef :: (Key -> Annex ()) -> Git.Ref -> Annex ()
|
|
|
|
withKeysReferencedInGitRef a ref = do
|
2011-12-12 22:23:24 +00:00
|
|
|
showAction $ "checking " ++ Git.Ref.describe ref
|
2012-03-12 19:21:20 +00:00
|
|
|
go =<< inRepo (LsTree.lsTree ref)
|
2011-09-28 20:43:10 +00:00
|
|
|
where
|
2012-04-22 03:32:33 +00:00
|
|
|
go [] = noop
|
2012-03-12 19:21:20 +00:00
|
|
|
go (l:ls)
|
2011-09-29 00:12:11 +00:00
|
|
|
| isSymLink (LsTree.mode l) = do
|
2012-02-01 20:05:02 +00:00
|
|
|
content <- L.decodeUtf8 <$> catFile ref (LsTree.file l)
|
2011-11-12 21:45:12 +00:00
|
|
|
case fileKey (takeFileName $ L.unpack content) of
|
2012-03-12 19:21:20 +00:00
|
|
|
Nothing -> go ls
|
2012-03-11 19:19:07 +00:00
|
|
|
Just k -> do
|
2012-03-12 19:21:20 +00:00
|
|
|
a k
|
|
|
|
go ls
|
|
|
|
| otherwise = go ls
|
2011-01-28 18:10:50 +00:00
|
|
|
|
2011-04-29 17:59:00 +00:00
|
|
|
{- Looks in the specified directory for bad/tmp keys, and returns a list
|
2012-03-11 16:38:59 +00:00
|
|
|
- of those that might still have value, or might be stale and removable.
|
2011-04-29 17:59:00 +00:00
|
|
|
-
|
2012-03-11 16:38:59 +00:00
|
|
|
- Also, stale keys that can be proven to have no value are deleted.
|
2011-04-29 17:59:00 +00:00
|
|
|
-}
|
2012-03-11 16:38:59 +00:00
|
|
|
staleKeysPrune :: (Git.Repo -> FilePath) -> Annex [Key]
|
|
|
|
staleKeysPrune dirspec = do
|
2011-05-17 01:18:34 +00:00
|
|
|
contents <- staleKeys dirspec
|
2011-09-23 22:13:24 +00:00
|
|
|
|
2012-03-11 16:38:59 +00:00
|
|
|
dups <- filterM inAnnex contents
|
|
|
|
let stale = contents `exclude` dups
|
2011-04-29 17:59:00 +00:00
|
|
|
|
2011-11-08 19:34:10 +00:00
|
|
|
dir <- fromRepo dirspec
|
2011-10-04 02:24:57 +00:00
|
|
|
liftIO $ forM_ dups $ \t -> removeFile $ dir </> keyFile t
|
2011-04-29 17:59:00 +00:00
|
|
|
|
|
|
|
return stale
|
|
|
|
|
2011-05-17 01:18:34 +00:00
|
|
|
staleKeys :: (Git.Repo -> FilePath) -> Annex [Key]
|
|
|
|
staleKeys dirspec = do
|
2011-11-08 19:34:10 +00:00
|
|
|
dir <- fromRepo dirspec
|
2012-03-14 21:43:34 +00:00
|
|
|
ifM (liftIO $ doesDirectoryExist dir)
|
|
|
|
( do
|
2011-04-29 17:59:00 +00:00
|
|
|
contents <- liftIO $ getDirectoryContents dir
|
2011-01-28 18:10:50 +00:00
|
|
|
files <- liftIO $ filterM doesFileExist $
|
2011-04-29 17:59:00 +00:00
|
|
|
map (dir </>) contents
|
2011-07-15 16:47:14 +00:00
|
|
|
return $ mapMaybe (fileKey . takeFileName) files
|
2012-03-14 21:43:34 +00:00
|
|
|
, return []
|
|
|
|
)
|