Fix V1 chat page HTTP 500 from missing reasoning resolver import

application/single_app/config.py

@@ -101,7 +101,7 @@
 EXECUTOR_TYPE = 'thread'
 EXECUTOR_MAX_WORKERS = 30
 SESSION_TYPE = 'filesystem'
-VERSION = "0.261.327"
+VERSION = "0.261.328"
 IS_DEVELOPMENT = is_development_env_enabled()
 
 # Opt-out for deployments where App Service Easy Auth is active but the platform

application/single_app/route_frontend_chats.py

@@ -19,6 +19,7 @@
 )
 from functions_model_endpoint_types import resolve_model_endpoint_request_model
 from functions_agent_catalog import build_accessible_agent_catalog
+from functions_model_capabilities import resolve_model_reasoning_policy
 from functions_ai_notice import get_ai_notice_config, is_ai_notice_dismissed
 from functions_collaboration import (
     assert_user_can_participate_in_collaboration_conversation,

docs/explanation/fixes/V1_CHAT_PAGE_BOOTSTRAP_FIX.md

@@ -0,0 +1,138 @@
+# V1 Chat Page Bootstrap Fix
+
+Fixed in version: **0.261.327**
+
+Cosmos review follow-up implemented in version: **0.261.328** (PR #1755).
+
+## Issue
+
+The classic V1 chat page could return HTTP 500 when navigating to `/chats`,
+before the composer or conversation list appeared. This affected nonempty
+legacy GPT, API Management (APIM), and multi-endpoint model catalogs.
+
+## Root Cause
+
+A model-catalog import cleanup removed the `functions_model_capabilities`
+import from `route_frontend_chats.py`. The V1-specific
+`_build_initial_chat_model_selection()` helper still called
+`resolve_model_reasoning_policy()` when serializing the initial model.
+
+The `/chats` route invokes that helper before rendering `chats.html`, outside
+the catalog-loading exception handlers. A nonempty catalog therefore raised
+`NameError: name 'resolve_model_reasoning_policy' is not defined`.
+An empty catalog returned `None` before reaching the missing dependency.
+
+Existing initial-selection tests checked source snippets but did not execute
+the helper or authenticated route, so they did not detect the regression.
+
+## Changes
+
+- `application/single_app/route_frontend_chats.py` explicitly imports the
+  reasoning resolver again. Its module uses standard-library dependencies and
+  the local capability catalog; the shared model-catalog extraction remains
+  intact.
+- `functional_tests/test_chat_v1_page_bootstrap.py` imports the real web app in
+  fresh normal and optimized Python processes with external I/O blocked.
+  It executes the real selector, catalog builder, authentication decorators,
+  settings sanitizer, and templates.
+- `application/single_app/config.py` advances the application version from
+  `0.261.326` to `0.261.327`.
+
+The fix does not change preferred model selection, legacy/APIM configured-first
+ordering, deployment aliases, reasoning policies, security requirements, or
+V2 routes. It does not catch the missing dependency or hide it by emptying the
+model catalog.
+
+## Validation
+
+Before the import was restored, the new page regressions raised the missing
+resolver `NameError` through the real authenticated `/chats` route in both
+normal and optimized Python.
+
+After the fix, cold and warm bootstrap paths render HTTP 200 for legacy, APIM,
+multi-endpoint, and empty catalogs. Coverage also checks selected deployments,
+canonical model names, supported/unsupported/unknown reasoning metadata,
+preferred model ids and selection keys, stale preferences, catalog immutability,
+and JSON serialization. Anonymous requests still redirect to login, users
+without the required role receive HTTP 403, and test credentials/private
+endpoint URLs do not appear in rendered HTML.
+
+The following commands ran with an isolated Python runtime for application
+integration tests; the route policy scripts also passed with the default
+interpreter:
+
+| Command | Result |
+| --- | --- |
+| `python -m pytest -q .\functional_tests\test_chat_v1_page_bootstrap.py .\functional_tests\test_chat_model_selector_initial_multiendpoint_bootstrap.py` | 7 passed, including the final run after the version update |
+| `python -m pytest -q .\functional_tests\test_model_reasoning_capability_resolution.py .\functional_tests\test_chat_retry_import_boundaries.py` | 31 passed, including 64 passing subtests |
+| `python -m pytest -q .\functional_tests\test_v2_ui_spa_route.py` | 6 passed |
+| `python .\functional_tests\route_tests\test_route_blueprint_policy_inventory.py` | 15 passed |
+| `python .\functional_tests\route_tests\test_route_unauthenticated_policy_contract.py` | 10 passed |
+| `python .\functional_tests\route_tests\test_route_policy_test_coverage.py` | 3 passed |
+| `python .\functional_tests\test_docs_app_surface_coverage.py` | 7 passed; generated inventory remains current |
+| `python .\functional_tests\test_docs_site_quality.py` | 6 passed |
+| `python -m pytest -q .\ui_tests\test_chat_page_multi_endpoint_notice_render.py .\ui_tests\test_staging_chat_smoke.py` | 2 skipped; an authenticated deployment test environment was not configured |
+| `git -c core.whitespace=blank-at-eol,blank-at-eof,space-before-tab,cr-at-eol diff --check` | Passed |
+
+The default local interpreter had an unrelated PyOpenSSL/cryptography
+compatibility failure during Semantic Kernel imports. Integration checks used
+a session-local virtual environment with the repository's Flask, Werkzeug,
+and Semantic Kernel versions and a compatible TLS dependency. No shared
+interpreter packages or production dependency manifests were changed.
+
+The investigation also found an unrelated stale expectation in
+`test_chat_template_json_bootstrap_safety.py`: it still requires the removed
+multi-endpoint migration notice, while the notice-removal regression requires
+its absence. That baseline failure is outside this fix.
+
+## Cosmos Contract Review Follow-up
+
+PR #1755's `cosmos-contracts` check failed with three
+`Changed owners requires review` diagnostics for `_build_new_chat_conversation`,
+`_resolve_chat_upload_context`, and `register_route_frontend_chats.upload_file`,
+plus `Stale source-only owner review` and
+`New/changed owner needs owner-runtime SDK evidence` for the upload context owner.
+
+The gate fingerprints the entire module as well as each storage function.
+Restoring the resolver import changed that module fingerprint, although all
+three function bodies remained unchanged. Their base-commit boundaries matched
+the committed inventory. This is a PR-caused review requirement, not a transient
+runner failure or a defect in Cosmos mapping compatibility.
+
+The inventory is refreshed and each consumer is reviewed. Both write owners
+discard point-write responses; the upload context owner reads mapping fields
+and returns the mapping without sending it to a strict plain-dictionary sink.
+It remains `sdk-compatible`, now with fresh real-owner runtime evidence instead
+of an expanded source-only baseline.
+
+`functional_tests/test_chat_upload_cosmos_response_contract.py` uses the existing
+SDK-default offline container and actual installed SDK `CosmosDict` responses.
+Fresh normal and optimized web imports exercise an owned personal conversation,
+cross-user denial without writes, new/missing conversation creation, retained
+fields and ETags, and discarded upsert responses. No new Cosmos double,
+production normalization, relaxed assertion, or CI configuration change is needed.
+The follow-up advances `application/single_app/config.py` to `0.261.328`.
+
+Follow-up validation uses the session-local interpreter with the application
+pin `azure-cosmos==4.9.0`. An initial full run exposed an inherited local SDK
+version mismatch (`4.16.3` versus `4.9.0`); only the virtual environment was
+corrected. Dependency assertions, production pins, and CI configuration remain
+unchanged.
+
+| Command (session-local Python where applicable) | Result |
+| --- | --- |
+| `python -m pytest -q -p test_support.cosmos_network_guard .\functional_tests\test_cosmos_document_boundaries.py .\functional_tests\test_chat_upload_cosmos_response_contract.py .\functional_tests\test_chat_v1_page_bootstrap.py .\functional_tests\test_chat_model_selector_initial_multiendpoint_bootstrap.py --disable-warnings --tb=short` | 59 passed |
+| `python .\scripts\check_cosmos_response_boundaries.py --run-contracts` | 1,051 owners, 394 doubles, zero review gaps; 683 passed and 231 subtests passed |
+| `python .\functional_tests\test_docs_app_surface_coverage.py` | 7 passed; inventory remains current |
+| `python .\functional_tests\test_docs_site_quality.py` | 6 passed |
+
+## Deployment Verification
+
+The missing import is reproduced and fixed locally. The supplied deployed
+HTTP 500 screenshot did not include a server traceback, so the deployed
+exception has not been independently matched to this regression.
+
+After deploying the fixed build, verify authenticated V1 `/chats` navigation
+and a complete send/response cycle with the configured model. Offline route
+coverage does not establish that the deployed model service can complete a
+chat request.

functional_tests/cosmos_response_boundaries.json

@@ -1043,7 +1043,6 @@
     "application/single_app/route_enhanced_citations.py::_get_authorized_chat_artifact_message": "b2d9c0fc54c4cf9545a6c46bb93b71c0e1a69d429c8ef1c05ba3fb071582bc4b",
     "application/single_app/route_enhanced_citations.py::register_enhanced_citations_routes.get_enhanced_citation_tabular": "ff7e5776314ddaf82390e3ac0817fb8270307d32da2b29a7bf4f4c7ef529d61e",
     "application/single_app/route_external_public_documents.py::register_route_external_public_documents.external_get_public_documents": "7a2cab5fcb36b8d80f2733e2a3fa2df33e04fbeeea7095bec72bf185778aef02",
-    "application/single_app/route_frontend_chats.py::_resolve_chat_upload_context": "ffacf10e9ecdb47901fa0deb6665e82bcdf06b7cb701feb024f76ab6db06d854",
     "application/single_app/route_frontend_control_center.py::get_control_center_statistics": "6fd3ab6692fe40b5e63490edcdae51eae7ebeb8067996c361cde4cac14c296e6",
     "application/single_app/route_frontend_conversations.py::_authorize_frontend_personal_conversation_access": "a8650a20c098eb598306d3827dce49f4b4c1501404cbd3b200d4887b1fbde552",
     "application/single_app/route_frontend_conversations.py::register_route_frontend_conversations.conversations": "a1d6c7ed62f7b68cb59058ec65e83993ef1f9756d129e439a9bccdc2377ea365",
@@ -24387,12 +24386,14 @@
         ],
         "queries": [],
         "batches": [],
-        "module_sha256": "8ebafa6d3ad84d52c1f94fefab84dd67ce8c270142c20114a1ed49b11369676f"
+        "module_sha256": "0641427633cbce91db00156c6934580b2a5c9578f65b4816bce1fbdeaf9eb97f"
       },
       "classification": "response-unused",
-      "contract": "_build_new_chat_conversation: point response values are discarded; query rows, if present, are separate plain SDK JSON. Existing write errors remain visible.",
+      "contract": "Constructs and returns its own plain conversation dictionary; the SDK upsert response is discarded. Real-owner new/missing upload context tests verify persisted SDK responses retain ETags without changing the returned domain document.",
       "evidence_scope": "source-review",
-      "tests": []
+      "tests": [
+        "functional_tests/test_chat_upload_cosmos_response_contract.py::test_chat_upload_context_real_owner_sdk_responses"
+      ]
     },
     "application/single_app/route_frontend_chats.py::_resolve_chat_upload_context": {
       "boundary": {
@@ -24406,13 +24407,13 @@
         ],
         "queries": [],
         "batches": [],
-        "module_sha256": "8ebafa6d3ad84d52c1f94fefab84dd67ce8c270142c20114a1ed49b11369676f"
+        "module_sha256": "0641427633cbce91db00156c6934580b2a5c9578f65b4816bce1fbdeaf9eb97f"
       },
       "classification": "sdk-compatible",
-      "contract": "_resolve_chat_upload_context: point documents use Mapping field reads/JSON encoding/metadata projections; batch status/body lists are separate, never top-level-normalized as documents. Strict domain sinks remain with their storage owners.",
-      "evidence_scope": "sdk-api-and-source-review",
+      "contract": "Point conversation responses use mapping field access for backing detection, collaboration identity, and personal ownership; the returned upload context retains the mapping and ETag. Upload consumers read/mutate fields and upsert the mapping, without an exact-dictionary sink. Real-owner runtime checks cover SDK personal reads, cross-user denial, and new/missing context creation.",
+      "evidence_scope": "owner-runtime",
       "tests": [
-        "functional_tests/test_m365_cosmos_sdk_contract.py::test_real_sdk_point_documents_are_wrapped_but_nested_json_and_query_rows_are_plain"
+        "functional_tests/test_chat_upload_cosmos_response_contract.py::test_chat_upload_context_real_owner_sdk_responses"
       ]
     },
     "application/single_app/route_frontend_chats.py::register_route_frontend_chats.upload_file": {
@@ -24478,10 +24479,10 @@
           }
         ],
         "batches": [],
-        "module_sha256": "8ebafa6d3ad84d52c1f94fefab84dd67ce8c270142c20114a1ed49b11369676f"
+        "module_sha256": "0641427633cbce91db00156c6934580b2a5c9578f65b4816bce1fbdeaf9eb97f"
       },
       "classification": "response-unused",
-      "contract": "register_route_frontend_chats.upload_file: point response values are discarded; query rows, if present, are separate plain SDK JSON. Existing write errors remain visible.",
+      "contract": "All six message/conversation upsert responses are discarded; upload responses project constructed message/workspace fields and conversation title/id, not the SDK write response. Query rows supply thread ids and counts separately. The unchanged body reads/mutates the SDK-compatible upload context mapping.",
       "evidence_scope": "source-review",
       "tests": []
     },

functional_tests/test_chat_upload_cosmos_response_contract.py

@@ -0,0 +1,87 @@
+# test_chat_upload_cosmos_response_contract.py
+"""Real chat upload context owners with the installed SDK's point responses.
+
+Version: 0.261.328
+Implemented in: 0.261.328
+
+Fresh normal/optimized processes import the real web application offline and
+exercise existing SDK-default storage doubles, without replacing the owners.
+"""
+
+import pytest
+
+from test_orchestration_result_imports import run_probe
+
+
+PROBE = r'''
+import sys
+
+sys.path[:0] = sys.argv[1:3]
+from azure.cosmos._cosmos_responses import CosmosDict
+from test_support.offline_bootstrap import offline_app_imports
+
+
+def check(condition, message):
+    if not condition:
+        raise AssertionError(message)
+
+
+with offline_app_imports() as environment:
+    import app as web
+    import route_frontend_chats as route
+
+    container = route.cosmos_conversations_container
+    saved = container.upsert_item({
+        "id": "upload-owned", "user_id": "upload-user",
+        "context": [], "tags": ["retained"], "strict": False,
+    })
+    check(type(saved) is CosmosDict, "Point writes must return the installed SDK wrapper.")
+    context = route._resolve_chat_upload_context(
+        " upload-owned ", "upload-user", {"userId": "upload-user"},
+    )
+    document = context["conversation_item"]
+    check(type(document) is CosmosDict, "The real owner should retain its compatible SDK mapping.")
+    check(document["_etag"] == saved["_etag"], "Point read lost its ETag.")
+    check(document["tags"] == ["retained"], "Point read lost domain fields.")
+    check(type(document["context"]) is list, "Nested JSON should remain plain.")
+    check(context["conversation_id"] == "upload-owned", "Partition-routed identity changed.")
+    check(context["response_conversation_id"] == "upload-owned", "Personal response identity changed.")
+    check(context["collaboration_conversation"] is None, "Personal read invented collaboration.")
+
+    before = dict(container.items)
+    try:
+        route._resolve_chat_upload_context(
+            "upload-owned", "other-user", {"userId": "other-user"},
+        )
+    except PermissionError:
+        pass
+    else:
+        raise AssertionError("SDK mapping bypassed personal conversation ownership.")
+    check(container.items == before, "Forbidden upload context changed storage.")
+
+    with web.app.test_request_context():
+        for requested_id in ("", "missing-upload"):
+            created = route._resolve_chat_upload_context(
+                requested_id, "upload-user", {"userId": "upload-user"},
+            )
+            document = created["conversation_item"]
+            check(type(document) is dict, "New conversation must return the constructed domain document.")
+            check(document["user_id"] == "upload-user", "New conversation lost its owner.")
+            check(created["conversation_id"] == document["id"], "Created conversation identity changed.")
+            check(created["response_conversation_id"] == document["id"], "Created response identity changed.")
+            check(created["collaboration_conversation"] is None, "Missing context invented collaboration.")
+            stored = container.read_item(item=document["id"], partition_key=document["id"])
+            check(type(stored) is CosmosDict, "Persisted new conversation is not SDK-shaped.")
+            check(bool(stored["_etag"]), "New conversation write lost its ETag.")
+            check(stored["title"] == "New Conversation", "New conversation defaults changed.")
+            check("_etag" not in document, "Unused write response leaked into the constructed document.")
+
+    check(not environment.network_attempts, "Upload context attempted external I/O.")
+print("PASS: chat upload SDK point response contract")
+'''
+
+
+@pytest.mark.parametrize("optimized", [False, True])
+def test_chat_upload_context_real_owner_sdk_responses(optimized):
+    """Verify mapping compatibility, ownership denial, and new/missing contexts."""
+    run_probe(PROBE, (), optimized)