reproduced

doc/bugs/SQlite_failed_when_copying_to_remote_repository/comment_7_3d4a7d330d80060275c34d0a13416fb7._comment

@@ -0,0 +1,78 @@
+[[!comment format=mdwn
+ username="joey"
+ subject="""comment 7"""
+ date="2023-03-14T17:00:06Z"
+ content="""
+Oho, I reproduced it!
+
+First I made a git-annex repo named "fooé", and added a file "foo" to it,
+and that is unlocked.
+
+	joey@darkstar:/tmp>git clone fooé bar
+	Cloning into 'bar'...
+	done.
+	joey@darkstar:/tmp>cd bar
+	joey@darkstar:/tmp/bar>git-annex move --from origin
+	move foo (from origin...)
+	(recording state in git...)
+	ok
+	(recording state in git...)
+	joey@darkstar:/tmp/bar>git-annex copy --to origin
+	copy foo (to origin...)
+	(recording state in git...)
+	ok
+	(recording state in git...)
+	joey@darkstar:/tmp/bar>git-annex move --from origin
+	move foo (from origin...) (recording state in git...)
+	ok
+	(recording state in git...)
+	joey@darkstar:/tmp/bar>LANG=C git-annex copy --to origin
+	copy foo (to origin...)
+	
+	  sqlite worker thread crashed: SQLite3 returned ErrorCan'tOpen while attempting to perform open "/tmp/foo\65533\65533/.git/annex/keysdb/db".
+	  CallStack (from HasCallStack):
+	    error, called at ./Database/Handle.hs:87:25 in main:Database.Handle
+	
+	ok
+
+Note that using git-annex in repo fooé with LANG=C works. The problem
+seems limited to a remote with a unicode character in its name when not in a
+unicode locale.
+
+In strace I see this:
+
+	3451696 openat(AT_FDCWD, "/tmp/foo\357\277\275\357\277\275/.git/annex/keysdb/db", O_RDWR|O_CREAT|O_NOFOLLOW|O_CLOEXEC, 0644 <unfinished ...>
+	3451694 <... close resumed>)            = 0
+	3451696 <... openat resumed>)           = -1 ENOENT (No such file or directory)
+
+That doesn't look like the correct encoding for "é" does it?
+`"/tmp/foo\303\251"` would be correct and is what git-annex otherwise uses
+when accessing that repo.
+
+I think the reason for this is simply that persistent-sqlite uses Text
+for the location of the database. And Text is unicode encoded. So when the non
+unicode locale results in a FilePath that is encoded using the filesystem encoding,
+with surrogate characters, and that gets fed to Data.Text.pack, it replaces the
+"invalid scalar values" with "\65533".
+
+The same thing would happen in a unicode locale if the remote's path was not
+valid unicode.
+
+Filed an issue to get persistent-sqlite to not use Text for the FilePath
+<https://github.com/yesodweb/persistent/issues/1481>
+
+I don't think that non-unicode FilePaths can be generally squeezed into a Text,
+so we may need to wait persistent getting fixed. Although it should be possible,
+in a non-unicode locale, to convert a non-unicode FilePath like "fooé"
+to a Text.
+
+----
+
+Also notice that git-annex succeeds despite this error. Which is reasonable
+since it was only unable to update the remote's keys db, and the remote
+can and will just update it itself next time git-annex is used over there.
+Which will work since that git-annex will be running in the directory 
+and will use relative paths.
+
+So perhaps the error message could just be suppressed?
+"""]]