fix(index): preserve undecodable symlink target bytes

Staging a symbolic link decoded its target through `os.readlink()` and then
strictly re-encoded it. Targets containing undecodable filesystem bytes raised
`UnicodeEncodeError` instead of being stored as the link blob.

Use `os.fsencode()` to recover the original target bytes. Git's
`object-file.c` stores the bytes read from a symlink directly; the scripted
regression stages `../target-\xff` and verifies those exact object bytes.
The fixture executes at its final location because native fixture copying
follows symlinks.

The regression fails before the fix and passes on both backends. Connected
index tests pass on CLI and Gix; the combined Gix Diff/tree/index run passes
253 tests and nine subtests, with two platform skips. Ruff and ShellCheck pass.

Author: codex

git/index/base.py

@@ -23,7 +23,7 @@
 from gitdb.base import IStream
 
 from git import _backend
-from git.compat import defenc, force_bytes, safe_decode
+from git.compat import defenc, safe_decode
 from git.cmd import Git
 import git.diff as git_diff
 from git.exc import CheckoutError, GitCommandError, GitError, InvalidGitRepositoryError, UnmergedEntriesError
@@ -754,8 +754,7 @@ def _store_path(self, filepath: PathLike, fprogress: Callable) -> BaseIndexEntry
 
         stream_size = st.st_size
         if S_ISLNK(st.st_mode):
-            # readlink is a string, but we need bytes.
-            target = force_bytes(os.readlink(filepath), encoding=defenc)
+            target = os.fsencode(os.readlink(filepath))
             stream_size = len(target)
 
             def open_stream() -> BinaryIO:

test/test_index.py

@@ -52,6 +52,17 @@ def test_staged_legacy_filename_and_blob_bytes():
         }
 
 
+@pytest.mark.skipif(os.name == "nt", reason="Windows symlink targets are Unicode paths")
+def test_staging_preserves_non_utf8_symlink_targets():
+    if not symlinks_supported():
+        pytest.skip("Symlinks unavailable")
+    # Native fixture copying follows symlinks instead of preserving their targets.
+    with fixture_repo("index-byte-symlink.sh", write=True, execute=True) as repo:
+        (entry,) = repo.index.add(["link"])
+        assert entry.mode == 0o120000
+        assert repo.odb.stream(entry.binsha).read() == b"../target-\xff"
+
+
 @ddt.ddt
 class TestIndex(TestBase):
     @with_rw_repo("HEAD")

tests/fixtures/index-byte-symlink.sh

@@ -0,0 +1,5 @@
+#!/usr/bin/env bash
+set -eu
+gix_testtools_require_symlinks
+git init --initial-branch=main repo
+ln -s "$(printf '../target-\377')" repo/link