feat(gix): stream small objects from native storage

Read object contents through GixPython and return independent byte streams from `Git.stream_object_data`. Partially reading one object must not corrupt another read. Retain CLI streaming above 8 MiB because the current native lookup buffers an entire object (GIX-2).

The staged version passes Ruff and Python syntax checks. Its SHA-1/SHA-256 native smoke and focused pytest selection pass: 9 passed in 2.40s.

Author: codex

git/_backend.py

@@ -7,6 +7,7 @@
 
 from collections import Counter
 from importlib import import_module
+import io
 import logging
 import os
 import re
@@ -96,11 +97,11 @@ def _oid(repo: Any, ref: Any) -> Any:
     return repo.rev_parse_single(ref)
 
 
-def object_data(command: Any, ref: bytes) -> Any:
+def object_data(command: Any, ref: bytes, *, stream: bool = False) -> Any:
     """Use native object reads without changing the public cat-file interface."""
     if gix is None:
         return NotImplemented
-    method = "get_object_header"
+    method = "stream_object_data" if stream else "get_object_header"
     command._require_version()
     try:
         repo = _repository(command, {})
@@ -109,6 +110,11 @@ def object_data(command: Any, ref: bytes) -> Any:
         if header is None:
             raise ValueError("SHA %s could not be resolved" % oid)
         result: Tuple[Any, ...] = (str(oid), header.kind(), header.size())
+        if stream:
+            # ponytail: native lookup buffers the object; use CLI above 8 MiB until GIX-2 provides streaming.
+            if header.size() > 8 * 1024 * 1024:
+                raise _Unsupported("large object streaming (GIX-2)")
+            result += (io.BytesIO(repo.find_object(oid).data),)
     except _Unsupported as exc:
         return _fallback(method, str(exc))
     except gix.Error as exc:

git/cmd.py

@@ -2093,6 +2093,9 @@ def stream_object_data(self, ref: Union[str, bytes]) -> Tuple[str, str, int, "Gi
             instance per thread to be safe!
         """
         request = self._prepare_ref(ref)
+        native = _backend.object_data(self, request[:-1], stream=True)
+        if native is not NotImplemented:
+            return native
         cmd = self._get_persistent_cmd("cat_file_all", "cat_file", batch=True, Z=True)
         hexsha, typename, size = self.__get_object_header(cmd, request)
         cmd_stdout = cmd.stdout if cmd.stdout is not None else io.BytesIO()

test/test_gix_backend.py

@@ -1,11 +1,13 @@
 """Exercise the installed backend, including proof that converted calls avoid Git."""
 
+from io import BytesIO
 from pathlib import Path
 from unittest.mock import patch
 
 import pytest
 
 from git import Git, Repo, _backend
+from gitdb import IStream
 
 gix = pytest.importorskip("gix")
 
@@ -29,3 +31,13 @@ def test_unknown_command_and_storage_environment_use_cli(repo):
         with repo.git.custom_environment(GIT_INDEX_FILE=".git/index"):
             assert repo.git._call_process_safe("ls_files", "--stage", "-v", "-z", "--full-name") == "fallback"
         assert cli.call_count == 2
+
+
+def test_partial_native_stream_does_not_corrupt_next_read(repo):
+    one = repo.odb.store(IStream("blob", 6, BytesIO(b"abcdef"))).binsha
+    two = repo.odb.store(IStream("blob", 3, BytesIO(b"xyz"))).binsha
+    with patch.object(Git, "execute", side_effect=AssertionError("unexpected CLI call")):
+        stream = repo.odb.stream(one)
+        assert stream.read(1) == b"a"
+        assert repo.odb.stream(two).read() == b"xyz"
+        assert stream.read() == b"bcdef"