fix(config): preserve bytes in subsection names and values

Git accepts configuration subsection names and values containing bytes that
are not valid in the filesystem encoding. Strict decoding made configuration
enumeration, remote names and URLs fail with `UnicodeDecodeError`; byte-valued
writes also decoded strictly before reaching Git.

Use the shared lossless decoder for Git's NUL-framed records and incoming
byte values. Keep existing key validation, mutation framing and Git-managed
locking. File and BytesIO regressions compare against `git config --null`,
exercise remote names/URLs and verify exact bytes after writes.

Four regressions fail before the fix. All 66 configuration tests pass on each
backend with Git 2.54.0 and GixPython 0.3.0. Ruff, mypy and whitespace checks
pass. Windows byte-valued argument writes remain excluded because Windows
command-line arguments cannot preserve arbitrary original bytes.

Author: codex

doc/source/changes.rst

@@ -125,6 +125,7 @@ API changes
   including updates when ``logmsg`` is omitted.
 * ``GitConfigParser`` uses Git's configuration syntax and canonical key spelling,
   and no longer subclasses ``configparser.RawConfigParser``.
+  Subsection names and values preserve undecodable bytes with ``surrogateescape``.
   Common getters, typed values, duplicate values, file/stream inputs and mutation
   methods remain, including ``GitConfigParser[section] = values`` for replacing
   nonempty sections. Invalid Git syntax previously accepted by Python is rejected.

git/config.py

@@ -17,7 +17,7 @@
 from typing import Any, Dict, Generic, Iterator, List, OrderedDict, Sequence, Tuple, TypeVar, Union, TYPE_CHECKING
 from typing import Mapping
 
-from git.compat import defenc, force_text
+from git.compat import safe_decode
 from git.exc import GitCommandError
 from git.types import _T, ConfigLevels_Tup, Lit_config_levels, PathLike, assert_never
 
@@ -368,11 +368,11 @@ def read(self) -> None:
                 if not record:
                     continue
                 key, separator, value = record.partition(b"\n")
-                prefix, option = key.decode(defenc).rsplit(".", 1)
+                prefix, option = safe_decode(key).rsplit(".", 1)
                 section = self._public_section(prefix)
                 if section not in sections:
                     sections[section] = _OMD()
-                sections[section].add(option, value.decode(defenc) if separator else None)
+                sections[section].add(option, safe_decode(value) if separator else None)
         self._sections = sections
         self._is_initialized = True
 
@@ -476,7 +476,7 @@ def _mutate(self, operation: str, *args: str) -> None:
     def _value(value: Any) -> str:
         if value is None:
             raise ValueError("Writing valueless Git configuration options is unsupported")
-        value = str(value) if isinstance(value, (int, float, bool)) else force_text(value)
+        value = str(value) if isinstance(value, (int, float, bool)) else safe_decode(value)
         if any(c in value for c in "\0\r"):
             raise ValueError("Git configuration values must not contain CR or NUL")
         return value

test/test_config.py

@@ -69,6 +69,42 @@ def test_config_values_and_mutations(config_directory, stream):
     assert (source.getvalue() if stream else path.read_bytes()).startswith(b"[CoRe]\nFlag\n")
 
 
+@pytest.mark.parametrize("stream", [False, True])
+def test_config_preserves_non_utf8_subsections_and_values(stream):
+    path = read_only_fixture("config-bytes.sh")
+    source = BytesIO((path / "config").read_bytes()) if stream else path / "config"
+    assert (
+        Git()
+        .config("list", "--file", str(path / "config"), "--null", stdout_as_string=False)
+        .endswith(b"remote.origin-\xff.url\ntarget-\xfe\0")
+    )
+    with GitConfigParser(source) as config:
+        assert 'remote "origin-\udcff"' in config.sections()
+        assert config.get('remote "origin-\udcff"', "url") == "target-\udcfe"
+    with Repo(path / "repo") as repo:
+        remote = repo.remotes[0]
+        assert remote.name == "origin-\udcff"
+        assert remote.url == "target-\udcfe"
+
+
+@pytest.mark.skipif(os.name == "nt", reason="Windows command arguments cannot carry arbitrary bytes")
+@pytest.mark.parametrize("stream", [False, True])
+def test_config_mutations_preserve_non_utf8_values(stream):
+    with writable_fixture("config-bytes.sh") as fixture:
+        path = fixture.path() / "config"
+        source = BytesIO(path.read_bytes()) if stream else path
+        with GitConfigParser(source, read_only=False) as config:
+            section = 'remote "origin-\udcff"'
+            assert config.get(section, "url") == "target-\udcfe"
+            config.set_value(section, "url", "updated-\udcfd")
+            config.add_value(section, "url", b"bytes-\xfc")
+            assert config.get_values(section, "url") == ["updated-\udcfd", "bytes-\udcfc"]
+        content = source.getvalue() if stream else path.read_bytes()
+        assert b'"origin-\xff"' in content
+        assert b"updated-\xfd" in content
+        assert b"bytes-\xfc" in content
+
+
 def test_config_includes_and_repository_conditions():
     path = read_only_fixture("config.sh", "includes")
     with Repo(path / "repo") as repo:

tests/fixtures/config-bytes.sh

@@ -0,0 +1,5 @@
+#!/usr/bin/env bash
+set -eu
+git init -q repo
+printf '[remote "origin-\377"]\n\turl = target-\376\n' >> repo/.git/config
+cp repo/.git/config config