fix(files): reduce any scheme or case wrapping files/<id> the way gemini does before the ownership lookup

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

litellm/proxy/openai_files_endpoints/common_utils.py

@@ -3,7 +3,7 @@ import itertools
 import logging
 import mimetypes
 import re
-from collections.abc import Mapping, Sequence
+from collections.abc import Iterator, Mapping, Sequence
 from dataclasses import dataclass, field
 from datetime import datetime, timezone
 from types import MappingProxyType
@@ -457,11 +457,27 @@ def canonical_provider_file_id(file_id: str) -> str:
 _GEMINI_FILES_URL: Final = "https://generativelanguage.googleapis.com/v1beta/files/"
 
 
+def _gemini_files_tail(file_id: str) -> str:
+    """The id Gemini resolves ``file_id`` to, whatever scheme or case wraps it: the text after its ``files/``."""
+    files_index: Final = file_id.find("files/")
+    return file_id[files_index:].strip("/").removeprefix("files/").strip("/") if files_index != -1 else ""
+
+
+def _iter_stored_forms(bare_ids: Sequence[str]) -> Iterator[str]:
+    for bare_id in bare_ids:
+        yield bare_id
+        yield f"files/{bare_id}"
+        yield f"{_GEMINI_FILES_URL}{bare_id}"
+
+
 def provider_file_id_spellings(file_id: str) -> tuple[str, ...]:
     """Every spelling an ownership row may hold for ``file_id``: as sent, unwrapped from a model-encoded id and
-    canonicalized, plus the ``files/`` and Gemini URL forms that upload rows stored before ids were canonicalized."""
-    canonical: Final = canonical_provider_file_id(get_original_file_id(file_id))
-    return tuple(dict.fromkeys((file_id, canonical, f"files/{canonical}", f"{_GEMINI_FILES_URL}{canonical}")))
+    canonicalized, reduced the way Gemini reduces it, plus the ``files/`` and Gemini URL forms that upload rows
+    stored before ids were canonicalized."""
+    original: Final = get_original_file_id(file_id)
+    reductions: Final = (canonical_provider_file_id(original), _gemini_files_tail(unquote(original)))
+    bare_ids: Final = tuple(dict.fromkeys(bare for bare in reductions if bare))
+    return tuple(dict.fromkeys((file_id, *_iter_stored_forms(bare_ids))))
 
 
 def is_model_embedded_id(file_id: str) -> bool:

tests/unit/enterprise/proxy/hooks/test_managed_files.py

@@ -61,8 +61,7 @@ class _InMemoryManagedFileTable:
         or_filters: Final = where.get("OR")
         if isinstance(or_filters, list):
             for clause in or_filters:
-                matched: Final = await self.find_first(clause)
-                if matched is not None:
+                if (matched := await self.find_first(clause)) is not None:
                     return matched
             return None
         unified_file_id: Final = where.get("unified_file_id")
@@ -6581,6 +6580,8 @@ _GEMINI_OUTPUT_SPELLINGS: Final = (
     "files%2Fbatch-out1",
     "batch-out1/",
     "https://generativelanguage.googleapis.com/v1beta/files/batch-out1",
+    "HTTPS://generativelanguage.googleapis.com/v1beta/files/batch-out1",
+    "x://h/files/batch-out1",
 )
 
 

tests/unit/proxy/openai_files_endpoint/test_files_common_utils.py

@@ -1003,3 +1003,16 @@ def test_provider_file_id_spellings_unwraps_a_model_encoded_id_and_covers_the_st
     )
     for spelling in ("batch-out1", "files%2Fbatch-out1", "batch-out1/"):
         assert set(provider_file_id_spellings(spelling)) >= set(provider_file_id_spellings(wrapped)[1:]), spelling
+
+
+@pytest.mark.parametrize(
+    "respelling",
+    (
+        "HTTPS://generativelanguage.googleapis.com/v1beta/files/batch-out1",
+        "x://h/files/batch-out1",
+        "S3://b/files/batch-out1",
+        "x%3A%2F%2Fh%2Ffiles%2Fbatch-out1",
+    ),
+)
+def test_provider_file_id_spellings_reduce_any_scheme_the_way_gemini_does(respelling: str):
+    assert "batch-out1" in provider_file_id_spellings(respelling)