Files
isparkclaw-webui/tests/test_nous_portal_routing.py
nesquena-hermes 666d385c03 fix: Nous static models use @nous: prefix — v0.50.164 (#885)
fix: Nous static models use @nous: prefix — v0.50.164 (#885)

Follow-up to #854 / PR #870. The previous fix made Nous static IDs
slash-prefixed and added a portal-guard branch to resolve_model_provider().
This tightens the static list to use the explicit @nous: prefix, matching
the format of live-fetched models after ui.js's _fetchLiveModels() portal-
prefix step.

The @provider:model branch in resolve_model_provider() is more explicit and
reliable than the portal-guard fallback. Both static and live-fetched paths
now converge on the same resolver output — and as a side effect, the dedup
check in _fetchLiveModels() now correctly identifies static entries as already
present, eliminating duplicate entries in the dropdown for Nous users.

Verified: all 29 Nous models in the browser dropdown carry @nous: prefix,
routing confirmed correct via resolve_model_provider() for all 4 static IDs,
1941 tests passing.

Closes #854.
2026-04-22 22:56:21 -07:00

177 lines
8.0 KiB
Python

"""Regression tests for Nous portal model routing bugs (issue #854).
Two bugs fixed:
1. Nous static model IDs were bare names (claude-opus-4.6) instead of
slash-prefixed (anthropic/claude-opus-4.6), causing Nous to reject them.
2. resolve_model_provider() routed slash-prefixed cross-namespace models
through OpenRouter instead of the configured portal provider.
Invariant: when a portal provider (Nous, OpenCode) is active, the full
slash-prefixed model ID MUST be preserved end-to-end — portals use the
provider/model path as the canonical name at their inference endpoint.
Stripping the prefix to a bare name is exactly Bug 1, so the fix for Bug 2
must not reintroduce it.
"""
import sys
import types
def _models_with_provider(provider, monkeypatch):
"""Patch config.cfg to simulate an active provider, return resolve_model_provider."""
import api.config as config
old = dict(config.cfg)
config.cfg.clear()
config.cfg["model"] = {"provider": provider}
try:
config._cfg_mtime = config.Path(config._get_config_path()).stat().st_mtime
except Exception:
config._cfg_mtime = 0.0
try:
from api.config import resolve_model_provider
return resolve_model_provider
finally:
config.cfg.clear()
config.cfg.update(old)
class TestNousModelIds:
"""Nous static model IDs must use @nous: prefix for explicit portal routing."""
def test_nous_models_use_at_prefix(self):
"""All Nous static models must carry the @nous: explicit provider prefix.
This ensures they route through the @provider:model branch of
resolve_model_provider() — identical to the live-fetched path — rather
than relying on the slash-only portal provider guard.
"""
from api.config import _PROVIDER_MODELS
nous_models = _PROVIDER_MODELS.get("nous", [])
assert nous_models, "Nous must have at least one static model"
for m in nous_models:
mid = m["id"]
assert mid.startswith("@nous:"), (
f"Nous model '{mid}' must start with '@nous:' "
f"(e.g. @nous:anthropic/claude-opus-4.6) so it routes through "
f"the explicit provider hint branch, not the weaker portal guard."
)
def test_nous_known_models_present(self):
"""Key Nous models must be present with correct @nous:-prefixed IDs."""
from api.config import _PROVIDER_MODELS
nous_ids = {m["id"] for m in _PROVIDER_MODELS.get("nous", [])}
assert "@nous:anthropic/claude-opus-4.6" in nous_ids, (
"@nous:anthropic/claude-opus-4.6 must be in Nous model list"
)
assert "@nous:anthropic/claude-sonnet-4.6" in nous_ids, (
"@nous:anthropic/claude-sonnet-4.6 must be in Nous model list"
)
assert "@nous:openai/gpt-5.4-mini" in nous_ids, (
"@nous:openai/gpt-5.4-mini must be in Nous model list"
)
def test_nous_models_no_bare_or_slash_only(self):
"""No Nous static model should be bare or slash-only without @nous: prefix."""
from api.config import _PROVIDER_MODELS
bad_ids = {
"claude-opus-4.6", "claude-sonnet-4.6", "gpt-5.4-mini",
"gemini-3.1-pro-preview",
"anthropic/claude-opus-4.6", "anthropic/claude-sonnet-4.6",
"openai/gpt-5.4-mini", "google/gemini-3.1-pro-preview",
}
nous_ids = {m["id"] for m in _PROVIDER_MODELS.get("nous", [])}
for bad in bad_ids:
assert bad not in nous_ids, (
f"Model ID '{bad}' found in Nous static list without @nous: prefix. "
f"Use '@nous:{bad}' so routing matches the live-fetched path."
)
class TestPortalProviderRouting:
"""Portal providers (Nous, OpenCode) must route cross-namespace models
through themselves, not through OpenRouter."""
def _resolve(self, model_id, provider):
import api.config as config
old = dict(config.cfg)
old_mtime = config._cfg_mtime
config.cfg.clear()
config.cfg["model"] = {"provider": provider}
try:
config._cfg_mtime = config.Path(config._get_config_path()).stat().st_mtime
except Exception:
config._cfg_mtime = 0.0
try:
from api.config import resolve_model_provider
return resolve_model_provider(model_id)
finally:
config.cfg.clear()
config.cfg.update(old)
config._cfg_mtime = old_mtime
def test_nous_routes_anthropic_model(self):
"""anthropic/claude-opus-4.6 with nous provider must route to nous with
the full slash-prefixed ID preserved — Nous rejects bare names."""
model, provider, _ = self._resolve("anthropic/claude-opus-4.6", "nous")
assert provider == "nous", (
f"Expected provider='nous', got '{provider}'. "
f"Nous portal must handle cross-namespace models directly."
)
assert model == "anthropic/claude-opus-4.6", (
f"Expected full slash-prefixed 'anthropic/claude-opus-4.6', got '{model}'. "
f"Portals need the provider/model path to route upstream (Bug 1)."
)
def test_nous_routes_openai_model(self):
"""openai/gpt-5.4-mini with nous provider must route to nous with slash preserved."""
model, provider, _ = self._resolve("openai/gpt-5.4-mini", "nous")
assert provider == "nous", f"Expected provider='nous', got '{provider}'."
assert model == "openai/gpt-5.4-mini", (
f"Expected 'openai/gpt-5.4-mini', got '{model}' — portal must preserve namespace."
)
def test_nous_routes_google_model(self):
"""google/gemini-3.1-pro-preview with nous provider must route to nous with slash preserved."""
model, provider, _ = self._resolve("google/gemini-3.1-pro-preview", "nous")
assert provider == "nous", f"Expected provider='nous', got '{provider}'."
assert model == "google/gemini-3.1-pro-preview", (
f"Expected 'google/gemini-3.1-pro-preview', got '{model}'."
)
def test_opencode_zen_routes_cross_namespace(self):
"""opencode-zen is also a portal — cross-namespace models must route through it
with the slash-prefixed ID preserved."""
model, provider, _ = self._resolve("anthropic/claude-sonnet-4.6", "opencode-zen")
assert provider == "opencode-zen", f"Expected provider='opencode-zen', got '{provider}'."
assert model == "anthropic/claude-sonnet-4.6", (
f"Expected 'anthropic/claude-sonnet-4.6', got '{model}'."
)
def test_portal_path_matches_at_prefix_path(self):
"""Static dropdown (bare slash) and live-fetched (@provider: prefix) paths
must produce identical resolver output for the same model — otherwise
Nous receives different forms depending on catalog source.
"""
# Static dropdown form
m1, p1, _ = self._resolve("anthropic/claude-opus-4.6", "nous")
# Live-fetched form (after ui.js _fetchLiveModels prefixes with @nous:)
m2, p2, _ = self._resolve("@nous:anthropic/claude-opus-4.6", "nous")
assert (m1, p1) == (m2, p2), (
f"Static path {m1, p1} and live path {m2, p2} must match — "
f"both should send the same model ID to Nous."
)
def test_non_portal_still_routes_to_openrouter(self):
"""Non-portal providers (anthropic) must still route cross-namespace to OpenRouter."""
model, provider, _ = self._resolve("openai/gpt-5.4-mini", "anthropic")
assert provider == "openrouter", (
f"Expected provider='openrouter' for cross-namespace with anthropic config, "
f"got '{provider}'."
)
def test_openrouter_config_keeps_full_path(self):
"""OpenRouter config must always keep the full provider/model path."""
model, provider, _ = self._resolve("anthropic/claude-sonnet-4.6", "openrouter")
assert provider == "openrouter"
assert model == "anthropic/claude-sonnet-4.6"