1{- git-union-merge library
2 -
3 - Copyright 2011 Joey Hess <id@joeyh.name>
4 -
5 - Licensed under the GNU AGPL version 3 or higher.
6 -}
7
8module Git.UnionMerge (
9	merge,
10	mergeIndex
11) where
12
13import qualified Data.ByteString as S
14import qualified Data.ByteString.Lazy as L
15import qualified Data.ByteString.Lazy.Char8 as L8
16import qualified Data.ByteString.Char8 as S8
17import qualified Data.Set as S
18
19import Common
20import Git
21import Git.Sha
22import Git.CatFile
23import Git.Command
24import Git.UpdateIndex
25import Git.HashObject
26import Git.Types
27import Git.FilePath
28
29{- Performs a union merge between two branches, staging it in the index.
30 - Any previously staged changes in the index will be lost.
31 -
32 - Should be run with a temporary index file configured by useIndex.
33 -}
34merge :: Ref -> Ref -> Repo -> IO ()
35merge x y repo = do
36	hashhandle <- hashObjectStart True repo
37	ch <- catFileStart repo
38	streamUpdateIndex repo
39		[ lsTree x repo
40		, mergeTrees x y hashhandle ch repo
41		]
42	catFileStop ch
43	hashObjectStop hashhandle
44
45{- Merges a list of branches into the index. Previously staged changes in
46 - the index are preserved (and participate in the merge).
47 -
48 - update-index is run once per ref in turn, so that each ref is merged on
49 - top of the merge for the previous ref. It would be more efficient, but
50 - harder to calculate a single union merge involving all the refs, as well
51 - as the index.
52 -}
53mergeIndex :: HashObjectHandle -> CatFileHandle -> Repo -> [Ref] -> IO ()
54mergeIndex hashhandle ch repo bs = forM_ bs $ \b ->
55	streamUpdateIndex repo [mergeTreeIndex b hashhandle ch repo]
56
57{- For merging two trees. -}
58mergeTrees :: Ref -> Ref -> HashObjectHandle -> CatFileHandle -> Repo -> Streamer
59mergeTrees x y hashhandle ch = doMerge hashhandle ch
60	("diff-tree":diffOpts ++ [fromRef x, fromRef y, "--"])
61
62{- For merging a single tree into the index. -}
63mergeTreeIndex :: Ref -> HashObjectHandle -> CatFileHandle -> Repo -> Streamer
64mergeTreeIndex r hashhandle ch = doMerge hashhandle ch $
65	"diff-index" : diffOpts ++ ["--cached", fromRef r, "--"]
66
67diffOpts :: [String]
68diffOpts = ["--raw", "-z", "-r", "--no-renames", "-l0"]
69
70{- Streams update-index changes to perform a merge,
71 - using git to get a raw diff. -}
72doMerge :: HashObjectHandle -> CatFileHandle -> [String] -> Repo -> Streamer
73doMerge hashhandle ch differ repo streamer = do
74	(diff, cleanup) <- pipeNullSplit' (map Param differ) repo
75	go diff
76	void $ cleanup
77  where
78	go [] = noop
79	go (info:file:rest) = mergeFile info file hashhandle ch >>=
80		maybe (go rest) (\l -> streamer l >> go rest)
81	go (_:[]) = error $ "parse error " ++ show differ
82
83{- Given an info line from a git raw diff, and the filename, generates
84 - a line suitable for update-index that union merges the two sides of the
85 - diff. -}
86mergeFile :: S.ByteString -> RawFilePath -> HashObjectHandle -> CatFileHandle -> IO (Maybe L.ByteString)
87mergeFile info file hashhandle h = case S8.words info of
88	[_colonmode, _bmode, asha, bsha, _status] ->
89		case filter (`notElem` nullShas) [Ref asha, Ref bsha] of
90		[] -> return Nothing
91		(sha:[]) -> use sha
92		shas -> use
93			=<< either return (hashBlob hashhandle . L8.unlines)
94			=<< calcMerge . zip shas <$> mapM getcontents shas
95	_ -> return Nothing
96  where
97	use sha = return $ Just $
98		updateIndexLine sha TreeFile $ asTopFilePath file
99	-- Get file and split into lines to union merge.
100	-- The encoding of the file is assumed to be either ASCII or utf-8;
101	-- in either case it's safe to split on \n
102	getcontents s = L8.lines <$> catObject h s
103
104{- Calculates a union merge between a list of refs, with contents.
105 -
106 - When possible, reuses the content of an existing ref, rather than
107 - generating new content.
108 -}
109calcMerge :: [(Ref, [L8.ByteString])] -> Either Ref [L8.ByteString]
110calcMerge shacontents
111	| null reuseable = Right new
112	| otherwise = Left $ fst $ Prelude.head reuseable
113  where
114	reuseable = filter (\c -> sorteduniq (snd c) == new) shacontents
115	new = sorteduniq $ concat $ map snd shacontents
116	sorteduniq = S.toList . S.fromList
117