NFD file names appear and disappear repeatedly when scanning

[PVince81]

Steps

1. Setup SMB external storage (don't enable NFD compatibility mode)
2. Put a file with NFD-encoded file name called "bête.txt" (make sure to copy-paste this exact string as it contains an NFD encoded accented character)
3. Run occ files:scan --all
4. Check oc_filecache entries: file is there
5. Run occ files:scan --all again
6. Check oc_filecache entries: file is not there
7. Run occ files:scan --all again
8. Check oc_filecache entries: file is there again with a different file id

Expected result

Rescanning repeatedly must give the same results

Actual result

File appears, disappears, appears, disappears after each scan run respectively.
Also see slightly related: #29438

Versions

Tested with v21.0.4

[PVince81]

same behavior after enabling NFD compatibility mode 😦
at least in that mode it should keep the files visible

[PVince81]

also happening on master for both scenarios with and without compatibility mode on

[PVince81]

the bug is happening in the scanner where we use the NFD name in https://github.com/nextcloud/server/blob/master/lib/private/Files/Cache/Scanner.php#L422 while the $existingChildren contains names normalized to NFD https://github.com/nextcloud/server/blob/master/lib/private/Files/Cache/Scanner.php#L426 so we never find the existing ones again

one possible solution would be to normalize $file already
not sure if we had this before and this got remove for some reason like performance

[PVince81]

it looks like this used to be normalized: 9735b5d#diff-2b1f228ea5cf1c94347251d18c7a94f86bfb229573277dd160cb57cd7b16c864L381 called by 9735b5d#diff-2b1f228ea5cf1c94347251d18c7a94f86bfb229573277dd160cb57cd7b16c864L428

so this is a regression from NC < 19

[PVince81]

PR is here with a working fix #29605