Work around git 2.50 bug that caused it to crash when there is a merge conflict with an unlocked annexed file

This fixes several test suite failures with git 2.50.

See the bug report for the full, gory details.

CHANGELOG

@@ -6,6 +6,8 @@ git-annex (10.20250606) UNRELEASED; urgency=medium
     filesystems like VFAT that don't support such filenames.
   * webapp: Rename "Upgrade Repository" to "Convert Repository"
     to avoid confusion with git-annex upgrade.
+  * Work around git 2.50 bug that caused it to crash when there is a merge
+    conflict with an unlocked annexed file.
 
  -- Joey Hess <id@joeyh.name>  Mon, 23 Jun 2025 11:11:29 -0400
 

Database/Keys.hs

@@ -1,6 +1,6 @@
 {- Sqlite database of information about Keys
  -
- - Copyright 2015-2022 Joey Hess <id@joeyh.name>
+ - Copyright 2015-2025 Joey Hess <id@joeyh.name>
  -
  - Licensed under the GNU AGPL version 3 or higher.
  -}
@@ -260,7 +260,7 @@ isInodeKnown i s = or <$> runReaderIO ContentTable
  - is an associated file.
  -}
 reconcileStaged :: Bool -> H.DbQueue -> Annex DbTablesChanged
-reconcileStaged dbisnew qh = ifM isBareRepo
+reconcileStaged dbisnew qh = ifM notneeded
 	( return mempty
 	, do
 		gitindex <- inRepo currentIndexFile
@@ -335,6 +335,14 @@ reconcileStaged dbisnew qh = ifM isBareRepo
 	getindextree = inRepo $ \r -> writeTreeQuiet $ r
 		{ gitGlobalOpts = gitGlobalOpts r ++ bypassSmudgeConfig }
 	
+	notneeded = isBareRepo
+		-- Avoid doing anything when run by the 
+	 	-- smudge clean filter. When that happens in a conflicted
+		-- merge situation, running git write-tree
+		-- here would cause git merge to fail with an internal
+		-- error. This works around around that bug in git.
+		<||> Annex.getState Annex.insmudgecleanfilter
+	
 	diff old new =
 		-- Avoid running smudge clean filter, since we want the
 		-- raw output, and it would block trying to access the

doc/bugs/test_suite_fail_with_git_2.50.mdwn

@@ -82,3 +82,15 @@ So, something that reconcileStaged does is making git unhappy when it
 runs the smudge clean filter while creating a stash. It seems logical that
 the problem would involve the index file, which `reconcileStaged` touches,
 and which gets updated when stashing..
+
+> Made reconcileStaged run `git write-tree` but not do anything else, and
+> that is sufficient to make git stash fail. This must be a bug in git,
+> `git write-tree` should be able to be run at any time, even if it exits
+> 1 due to the index being in conflict. Having `git write-tree` affect
+> another process that was already running is not good behavior for git.
+> Since `git write-tree` does need to sometimes update the index,
+> this feels like lacking locking.
+> 
+> I have worked around this by making reconcileStaged avoid doing anything
+> when called by the smudge clean filter. Which I don't think will cause
+> any other problems, fingers crossed. [[done]] --[[Joey]]