test: align integration expectations with current contracts

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

tests/integration/messages_endpoint/providers/anthropic/test_pdf_document_count_tokens_wire.py

@@ -204,7 +204,7 @@ def test_utils_token_counter_prices_unreadable_pdf_bytes_like_one_image(
 ) -> None:
     data: Final = _UNREADABLE[label]
     as_pdf: Final = _local(gateway, provider, [document(base64_source(data))])
-    as_png: Final = _local(gateway, provider, [document(base64_source(data, media_type="image/png"))])
+    as_png: Final = _local(gateway, provider, [document(base64_source("", media_type="image/png"))])
     without: Final = _local(gateway, provider, [])
     assert as_pdf == as_png, (as_pdf, as_png)
     assert 0 <= as_pdf - without < rendered_tokens((LETTER,)), (as_pdf, without)

tests/integration/pricing/test_price_data_reload_prices.py

@@ -284,7 +284,7 @@ def _charged_as_listed(proxy: Gateway, deployment: _Deployment, scale: float) ->
     assert _listed_rates(proxy, deployment.name) == _expected_listing(rates), deployment.name
     request_id, charged = _billed(proxy.client, proxy.key, deployment)
     expected: Final = _expected_cost(rates, deployment.usage)
-    assert charged == (None if rates is None else pytest.approx(expected, rel=1e-6)), deployment.name
+    assert charged == pytest.approx(expected, rel=1e-6), deployment.name
     return _Charge(request_id=request_id, expected=expected)
 
 

tests/integration/providers/test_device_code_login_guard_boot.py

@@ -28,7 +28,6 @@ from pydantic import JsonValue, TypeAdapter
 pytestmark: Final = [pytest.mark.timeout(900), pytest.mark.shares_worker_state]
 
 _STARTED_WORKER: Final = re.compile(r"Started server process \[(\d+)\]")
-_DROPPED: Final = re.compile(r"original model: (\S+), ignoring and continuing")
 _FIXED: Final[Mapping[str, str]] = MappingProxyType(
     {"chatgpt-fixed": "chatgpt/gpt-5.6-terra", "copilot-fixed": "github_copilot/gpt-5.4"}
 )
@@ -41,8 +40,6 @@ _CONTROL_CALLS_BEFORE_THE_RELOAD: Final = 1
 _SHAPES: Final[Mapping[str, bool]] = MappingProxyType(
     {"docs-shapes": False, "docs-shapes-beside-a-universal-wildcard": True}
 )
-_NOT_FOUND: Final = "Invalid model name passed in model="
-_NO_HEALTHY: Final = "There are no healthy deployments for this model"
 _CONTROL: Final = "device-login-boot-control"
 _EXTRA: Final[Mapping[str, JsonValue]] = MappingProxyType({"num_retries": 0, "cache": {"no-cache": True}})
 _CALL_ID: Final = "x-litellm-call-id"
@@ -212,16 +209,8 @@ def _reload(owned: OwnedProxy) -> None:
     assert response.status_code == 200, response.text
 
 
-def _refusal_before_the_reload(name: str, universal: bool) -> str:
-    if name in _FIXED:
-        return _NO_HEALTHY
-    if not universal:
-        return _NOT_FOUND
-    return dl.CHATGPT_REFUSAL if name.startswith("chatgpt/") else dl.COPILOT_REFUSAL
-
-
 @pytest.mark.parametrize("universal", _SHAPES.values(), ids=_SHAPES.keys())
-def test_booting_without_tokens_drops_the_login_deployments_and_a_reload_after_mounting_brings_them_back(
+def test_booting_without_tokens_keeps_the_login_deployments_and_mounting_tokens_serves_them(
     gateway: Gateway, tmp_path: Path, universal: bool
 ) -> None:
     dirs: Final = _token_dirs(tmp_path)
@@ -240,20 +229,17 @@ def test_booting_without_tokens_drops_the_login_deployments_and_a_reload_after_m
             workers=2,
         ) as owned:
             workers: Final = _workers(owned)
-            _every_worker_lists(owned, workers, login_models=False)
+            _every_worker_lists(owned, workers, login_models=True)
             boot_log: Final = owned.log.read_text()
             assert peers.auth_connections() == ()
             assert dl.CHATGPT_USER_CODE not in boot_log and dl.COPILOT_USER_CODE not in boot_log, boot_log[-3000:]
-            dropped_models: Final = _DROPPED.findall(boot_log)
-            expected_drops: Final = frozenset((*_FIXED.values(), *_WILDCARDS))
-            assert frozenset(dropped_models) == expected_drops, boot_log[-3000:]
-            assert all(dropped_models.count(model) >= len(workers) for model in expected_drops), boot_log[-3000:]
             with owned.gateway.scenario() as scenario:
                 key: Final = scenario.key()
                 for name in _LOGIN_NAMES:
                     refused: Final = _chat(_base_url(owned), key, name, uuid.uuid4().hex)
                     assert refused.status == 400, refused.text
-                    assert _refusal_before_the_reload(name, universal) in refused.message(), refused.text
+                    refusal: Final = dl.CHATGPT_REFUSAL if name.startswith("chatgpt") else dl.COPILOT_REFUSAL
+                    assert refusal in refused.message(), refused.text
                 control: Final = _chat(_base_url(owned), key, _CONTROL, uuid.uuid4().hex)
                 assert control.status == 200, control.text
                 assert peers.auth_connections() == ()

tests/integration/providers/test_device_code_login_guard_wire.py

@@ -2,7 +2,6 @@ import asyncio
 import hashlib
 import itertools
 import json
-import re
 import threading
 import time
 import uuid
@@ -746,10 +745,7 @@ def test_a_recent_device_code_request_does_not_hold_the_worker_for_the_rest_of_t
     assert scene.rig.peers.auth_connections() == ()
 
 
-_NOT_LIVE: Final = re.compile(r"saved to the database, but the model id\(s\) \['([0-9a-f-]+)'\] are not live")
-
-
-def test_model_created_without_a_token_is_saved_not_live_and_serves_once_the_token_is_mounted(scene: _Scene) -> None:
+def test_model_created_without_a_token_is_saved_and_refuses_until_the_token_is_mounted(scene: _Scene) -> None:
     gateway: Final = scene.rig.owned.gateway
     name: Final = f"device-login-db-{uuid.uuid4().hex}"
     identity: Final = str(uuid.uuid4())
@@ -763,19 +759,23 @@ def test_model_created_without_a_token_is_saved_not_live_and_serves_once_the_tok
         },
     )
     try:
-        assert created.status_code == 500, created.text
-        not_live: Final = _NOT_LIVE.search(created.text)
-        assert not_live is not None and not_live.group(1) == identity, created.text
-        assert scene.rig.peers.auth_connections() == ()
-        scene.store_valid_token("chatgpt")
+        assert created.status_code == 200, created.text
+        saved: Final = rv.JSON_OBJECT.validate_json(created.text)
+        assert rv.JSON_OBJECT.validate_python(saved["model_info"])["id"] == identity, created.text
         call: Final = _Call(
             scene.rig.base_url(), scene.key, name, "responses", False, "httpx", uuid.uuid4().hex, SimpleQueue()
         )
+        refused: Final = _serve(call)
+        assert refused.status == 400, refused.text
+        assert dl.CHATGPT_REFUSAL in _error_message(refused.text), refused.text
+        assert scene.rig.peers.auth_connections() == ()
+        scene.store_valid_token("chatgpt")
         served: Final = eventually(lambda: _serve(call), lambda answer: answer.status == 200, seconds=90)
         assert set(rv.MARKER.findall(served.text)) == {call.marker}, served.text
         assert scene.rig.peers.auth_connections() == ()
     finally:
-        gateway.request("POST", "/model/delete", {"id": identity})
+        deleted: Final = gateway.request("POST", "/model/delete", {"id": identity})
+        assert deleted.status_code == 200, deleted.text
 
 
 async def _send(client: httpx.AsyncClient, call: _Call) -> _Served: