formatLsTree did not use a tab where git does

Fixed that, and made parserLsTree accept the space as well as tab.

Fixes a reversion that made import of a tree from a special remote result in
a merge that deleted files that were not preferred content of that special
remote.

CHANGELOG

@@ -1,3 +1,11 @@
+git-annex (8.20210128) UNRELEASED; urgency=medium
+
+  * Fix a reversion that made import of a tree from a special remote
+    result in a merge that deleted files that were not preferred content
+    of that special remote.
+
+ -- Joey Hess <id@joeyh.name>  Thu, 28 Jan 2021 12:34:32 -0400
+
 git-annex (8.20210127) upstream; urgency=medium
 
   * Added mincopies configuration. This is like numcopies, but is

Git/LsTree.hs

@@ -101,6 +101,10 @@ parseLsTreeStrict b = go (AS.parse parserLsTree b)
 {- Parses a line of ls-tree output, in format:
  - mode SP type SP sha TAB file
  -
+ - The TAB can also be a space. Git does not use that, but an earlier
+ - version of formatLsTree did, and this keeps parsing what it output
+ - working.
+ -
  - (The --long format is not currently supported.) -}
 parserLsTree :: A.Parser TreeItem
 parserLsTree = TreeItem
@@ -111,8 +115,8 @@ parserLsTree = TreeItem
 	<*> A8.takeTill (== ' ')
 	<* A8.char ' '
 	-- sha
-	<*> (Ref <$> A8.takeTill (== '\t'))
+	<*> (Ref <$> A8.takeTill A8.isSpace)
-	<* A8.char '\t'
+	<* A8.space
 	-- file
 	<*> (asTopFilePath . Git.Filename.decode <$> A.takeByteString)
 
@@ -122,5 +126,4 @@ formatLsTree ti = unwords
 	[ showOct (mode ti) ""
 	, decodeBS (typeobj ti)
 	, fromRef (sha ti)
-	, fromRawFilePath (getTopFilePath (file ti))
+	] ++ ('\t' : fromRawFilePath (getTopFilePath (file ti)))
-	]

Logs/Export.hs

@@ -197,10 +197,12 @@ logExportExcluded u a = do
 getExportExcluded :: UUID -> Annex [Git.Tree.TreeItem]
 getExportExcluded u = do
 	logf <- fromRepo $ gitAnnexExportExcludeLog u
-	liftIO $ catchDefaultIO [] $ parser
+	liftIO $ catchDefaultIO [] $ exportExcludedParser
 		<$> L.readFile (fromRawFilePath logf)
   where
-	parser = map Git.Tree.lsTreeItemToTreeItem
+
+exportExcludedParser :: L.ByteString -> [Git.Tree.TreeItem]
+exportExcludedParser = map Git.Tree.lsTreeItemToTreeItem
 	. rights
 	. map Git.LsTree.parseLsTree
 	. L.split (fromIntegral $ ord '\n')

doc/bugs/exporttree_+_importtree_+_wanted___61___deleted_files.mdwn

@@ -41,3 +41,5 @@ Debian Buster
 ### Have you had any luck using git-annex before? (Sometimes we get tired of reading bug reports all day and a lil' positive end note does wonders)
 
 I use git-annex for all kinds of stuff. I love it!
+
+> [[fixed|done]] --[[Joey]]

doc/bugs/exporttree_+_importtree_+_wanted___61___deleted_files/comment_1_0611c7552e9660d8f3b5060fa114c2ef._comment

@@ -0,0 +1,50 @@
+[[!comment format=mdwn
+ username="joey"
+ subject="""comment 1"""
+ date="2021-01-28T15:52:03Z"
+ content="""
+Note that there's not data loss here, you can still check
+out branches with the deleted files, or revert the merge.
+
+Seems like importtree should add to the imported tree all files that
+were in the export but were not preferred content.
+
+Hmm, in Annex.Import.addBackExportExcluded, it tries to do just that. The
+implementation uses a log file in .git/annex/export.ex that lists the
+previously excluded files. There must be a bug in that.
+
+I can easily reproduce this bug:
+
+	joey@darkstar:/tmp/bench2>mkdir d
+	joey@darkstar:/tmp/bench2>git init r
+	joey@darkstar:/tmp/bench2>cd r
+	joey@darkstar:/tmp/bench2/r>git annex init
+	joey@darkstar:/tmp/bench2/r>git annex initremote d type=directory directory=../d exporttree=yes importtree=yes encryption=none
+	joey@darkstar:/tmp/bench2/r>git annex wanted d 'exclude=*.mp3'
+	joey@darkstar:/tmp/bench2/r>date > foo.bar
+	joey@darkstar:/tmp/bench2/r>date > foo.mp3
+	joey@darkstar:/tmp/bench2/r>git annex add
+	joey@darkstar:/tmp/bench2/r>git commit -m add
+	joey@darkstar:/tmp/bench2/r>git annex export master --to d
+	export d foo.bar ok
+	joey@darkstar:/tmp/bench2/r>git annex import master --from d
+	list d ok
+	update refs/remotes/d/master ok
+	(recording state in git...)
+	joey@darkstar:/tmp/bench2/r>git merge d/master
+	Updating f818c13..b1a0434
+	Fast-forward
+	 foo.mp3 | 1 -
+	 1 file changed, 1 deletion(-)
+	 delete mode 120000 foo.mp3
+	joey@darkstar:/tmp/bench2/r>cat .git/annex/export.ex/72c8d14c-af03-408c-845d-cac418d49e61 
+	120000 blob 64c3e1f1f81026cb8ab5a6593d4120a1d73044c3 foo.mp3
+
+So it seems adding back the exported file from the log is where the bug lies.
+And specifically, it seems when it tries to read this log, it silently fails
+to parse it, and so adds nothing back. 
+
+Aha! -- The parser is expecting a tab in the git ls-tree like log, but it's
+written with a space instead. It did used to work but the parser got rewritten
+for speed and was changed to only accept tab, not both space and tab.
+"""]]