darcs-2.16.1: src/Darcs/Repository/Packs.hs
{-|
License : GPL-2
Packs are an optimization that enable faster repository cloning over HTTP.
A pack is actually a @tar.gz@ file that contains many files that would otherwise
have to be transfered one by one (which is much slower over HTTP).
Two packs are created at the same time by 'createPacks':
1. The basic pack, contains the latest recorded version of the working tree.
2. The patches pack, contains the set of patches of the repository.
The paths of these files are @_darcs\/packs\/basic.tar.gz@ and
@_darcs\/packs\/patches.tar.gz@. There is also @_darcs\/packs\/pristine@ which
indicates the pristine hash at the moment of the creation of the packs. This
last file is useful to determine whether the basic pack is in sync with the
current pristine of the repository.
-}
module Darcs.Repository.Packs
( fetchAndUnpackBasic
, fetchAndUnpackPatches
, packsDir
, createPacks
) where
import qualified Codec.Archive.Tar as Tar
import Codec.Archive.Tar.Entry ( fileEntry, toTarPath )
import Codec.Compression.GZip as GZ ( compress, decompress )
import Control.Concurrent.Async ( withAsync )
import Control.Exception ( Exception, IOException, throwIO, catch, finally )
import Control.Monad ( void, when, unless )
import System.IO.Error ( isAlreadyExistsError )
import System.IO.Unsafe ( unsafeInterleaveIO )
import qualified Data.ByteString.Lazy.Char8 as BLC
import Data.List ( isPrefixOf, sort )
import System.Directory ( createDirectoryIfMissing
, renameFile
, removeFile
, doesFileExist
, getModificationTime
, listDirectory
)
import System.FilePath ( (</>)
, (<.>)
, takeFileName
, splitPath
, joinPath
, takeDirectory
)
import System.Posix.Files ( createLink )
import Darcs.Prelude
import Darcs.Util.ByteString ( gzReadFilePS )
import Darcs.Util.Lock ( withTemp )
import Darcs.Util.External ( Cachable(..), fetchFileLazyPS )
import Darcs.Util.Global ( darcsdir )
import Darcs.Util.Progress ( debugMessage, progressList )
import Darcs.Patch ( IsRepoType, RepoPatch )
import Darcs.Patch.PatchInfoAnd ( extractHash )
import Darcs.Patch.Progress ( progressFL )
import Darcs.Patch.Witnesses.Ordered ( mapFL )
import Darcs.Patch.Set ( patchSet2FL )
import Darcs.Repository.Traverse ( listInventories )
import Darcs.Repository.InternalTypes ( Repository )
import Darcs.Repository.Hashed ( readRepo )
import Darcs.Repository.Inventory ( getValidHash )
import Darcs.Repository.Format
( identifyRepoFormat, formatHas, RepoProperty ( HashedInventory ) )
import Darcs.Repository.Cache ( fetchFileUsingCache
, HashedDir(..)
, Cache
, closestWritableDirectory
, hashedDir
, bucketFolder
)
import Darcs.Repository.Old ( oldRepoFailMsg )
import Darcs.Repository.Pristine ( readHashedPristineRoot )
packsDir, basicPack, patchesPack :: String
packsDir = "packs"
basicPack = "basic.tar.gz"
patchesPack = "patches.tar.gz"
fetchAndUnpack :: FilePath
-> HashedDir
-> Cache
-> FilePath
-> IO ()
fetchAndUnpack filename dir cache remote = do
unpackTar cache dir . Tar.read . GZ.decompress =<<
fetchFileLazyPS (remote </> darcsdir </> packsDir </> filename) Uncachable
fetchAndUnpackPatches :: [String] -> Cache -> FilePath -> IO ()
fetchAndUnpackPatches paths cache remote =
-- Patches pack can miss some new patches of the repository.
-- So we download pack asynchonously and alway do a complete pass
-- of individual patch files.
withAsync (fetchAndUnpack patchesPack HashedInventoriesDir cache remote) $ \_ -> do
fetchFilesUsingCache cache HashedPatchesDir paths
fetchAndUnpackBasic :: Cache -> FilePath -> IO ()
fetchAndUnpackBasic = fetchAndUnpack basicPack HashedPristineDir
unpackTar :: Exception e => Cache -> HashedDir -> Tar.Entries e -> IO ()
unpackTar _ _ Tar.Done = return ()
unpackTar _ _ (Tar.Fail e) = throwIO e
unpackTar c dir (Tar.Next e es) = case Tar.entryContent e of
Tar.NormalFile bs _ -> do
let p = Tar.entryPath e
if "meta-" `isPrefixOf` takeFileName p
then unpackTar c dir es -- just ignore them
else do
ex <- doesFileExist p
if ex
then debugMessage $ "TAR thread: exists " ++ p ++ "\nStopping TAR thread."
else do
if p == darcsdir </> "hashed_inventory"
then writeFile' Nothing p bs
else writeFile' (closestWritableDirectory c) p $ GZ.compress bs
debugMessage $ "TAR thread: GET " ++ p
unpackTar c dir es
_ -> fail "Unexpected non-file tar entry"
where
writeFile' Nothing path content = withTemp $ \tmp -> do
BLC.writeFile tmp content
renameFile tmp path
writeFile' (Just ca) path content = do
let fileFullPath = case splitPath path of
_:hDir:hFile:_ -> joinPath [ca, hDir, bucketFolder hFile, hFile]
_ -> fail "Unexpected file path"
createDirectoryIfMissing True $ takeDirectory path
createLink fileFullPath path `catch` (\(ex :: IOException) -> do
if isAlreadyExistsError ex then
return () -- so much the better
else
-- ignore cache if we cannot link
writeFile' Nothing path content)
-- | Similar to @'mapM_' ('void' 'fetchFileUsingCache')@, exepts
-- it stops execution if file it's going to fetch already exists.
fetchFilesUsingCache :: Cache -> HashedDir -> [FilePath] -> IO ()
fetchFilesUsingCache cache dir = mapM_ go where
go path = do
ex <- doesFileExist $ darcsdir </> hashedDir dir </> path
if ex
then debugMessage $ "FILE thread: exists " ++ path
else void $ fetchFileUsingCache cache dir path
-- | Create packs from the current recorded version of the repository.
createPacks :: (IsRepoType rt, RepoPatch p)
=> Repository rt p wR wU wT -> IO ()
createPacks repo = flip finally (mapM_ removeFileIfExists
[ darcsdir </> "meta-filelist-inventories"
, darcsdir </> "meta-filelist-pristine"
, basicTar <.> "part"
, patchesTar <.> "part"
]) $ do
rf <- identifyRepoFormat "."
-- function is exposed in API so could be called on non-hashed repo
unless (formatHas HashedInventory rf) $ fail oldRepoFailMsg
createDirectoryIfMissing False (darcsdir </> packsDir)
-- pristine hash
Just hash <- readHashedPristineRoot repo
writeFile ( darcsdir </> packsDir </> "pristine" ) $ getValidHash hash
-- pack patchesTar
ps <- mapFL hashedPatchFileName . progressFL "Packing patches" . patchSet2FL <$> readRepo repo
is <- map ((darcsdir </> "inventories") </>) <$> listInventories
writeFile (darcsdir </> "meta-filelist-inventories") . unlines $
map takeFileName is
-- Note: tinkering with zlib's compression parameters does not make
-- any noticeable difference in generated archive size;
-- switching to bzip2 would provide ~25% gain OTOH.
BLC.writeFile (patchesTar <.> "part") . GZ.compress . Tar.write =<<
mapM fileEntry' ((darcsdir </> "meta-filelist-inventories") : ps ++ reverse is)
renameFile (patchesTar <.> "part") patchesTar
-- pack basicTar
pr <- sortByMTime =<< dirContents (darcsdir </> "pristine.hashed")
writeFile (darcsdir </> "meta-filelist-pristine") . unlines $
map takeFileName pr
BLC.writeFile (basicTar <.> "part") . GZ.compress . Tar.write =<< mapM fileEntry' (
[ darcsdir </> "meta-filelist-pristine"
, darcsdir </> "hashed_inventory"
] ++ progressList "Packing pristine" (reverse pr))
renameFile (basicTar <.> "part") basicTar
where
basicTar = darcsdir </> packsDir </> basicPack
patchesTar = darcsdir </> packsDir </> patchesPack
fileEntry' x = unsafeInterleaveIO $ do
content <- BLC.fromChunks . return <$> gzReadFilePS x
tp <- either fail return $ toTarPath False x
return $ fileEntry tp content
dirContents dir = map (dir </>) <$> listDirectory dir
hashedPatchFileName x = case extractHash x of
Left _ -> fail "unexpected unhashed patch"
Right h -> darcsdir </> "patches" </> h
sortByMTime xs = map snd . sort <$> mapM (\x -> (\t -> (t, x)) <$>
getModificationTime x) xs
removeFileIfExists x = do
ex <- doesFileExist x
when ex $ removeFile x