tests/test_inference_seam.py

"""Configurable inference seam (server/model.py) — env-selected, network-free."""
from server import model


def _fake_resp(content):
    class R:
        def raise_for_status(self):
            pass

        def json(self):
            return {"choices": [{"message": {"content": content}}], "usage": {}}

    return R()


def test_remote_branch_used_when_base_url_set(monkeypatch):
    monkeypatch.setattr(model, "INFERENCE_BASE_URL", "http://localhost:9/v1")
    # if the local llama.cpp path were taken, get_llm would be called -> blow up
    monkeypatch.setattr(model, "get_llm", lambda: (_ for _ in ()).throw(AssertionError("local used")))

    calls = {}

    def fake_post(url, json, headers, timeout):
        calls["url"] = url
        calls["body"] = json
        return _fake_resp('{"events": []}')

    monkeypatch.setattr("requests.post", fake_post)

    out = model.complete_json([{"role": "user", "content": "hi"}], {"type": "object"})
    assert out == '{"events": []}'
    assert calls["url"].endswith("/chat/completions")
    assert calls["body"]["response_format"]["type"] == "json_schema"


def test_local_branch_when_base_url_unset(monkeypatch):
    monkeypatch.setattr(model, "INFERENCE_BASE_URL", "")

    def no_network(*a, **k):
        raise AssertionError("remote must not be called when INFERENCE_BASE_URL is unset")

    monkeypatch.setattr("requests.post", no_network)

    class FakeLLM:
        def create_chat_completion(self, **kw):
            return {"choices": [{"message": {"content": "{}"}}], "usage": {}}

    monkeypatch.setattr(model, "get_llm", lambda: FakeLLM())
    assert model.complete_json([{"role": "user", "content": "hi"}], {"type": "object"}) == "{}"