mirror of
https://github.com/NousResearch/hermes-agent.git
synced 2026-07-31 19:16:29 +00:00
6b81590c55
Systematic prune per AGENTS.md test policy, one pass over every major test tree (gateway, hermes_cli, tools, agent, run_agent, plugins, cli, cron, tui_gateway, honcho/openviking, root-level): - DELETE: source-reading tests (read_text/getsource on prod files), change-detector tests (exact catalog counts, model-name snapshots, config version literals), mock-echo tests (assert a mock returns what it was told), assertion-free/trivial tests, near-duplicate parametrizations (boundaries + one representative kept), async/sync twin duplicates, cosmetic within-file variations. - KEEP (mandatory): security/redaction/approval guards, message-role alternation invariants, prompt-caching/deterministic-call-id invariants, issue-number regression tests (deduped), E2E tests. - 6 test files deleted outright (script-style/no-assert or fully redundant); conftest.py, fakes/, fixtures/ untouched. - tests/acp/conftest.py added: autouse fixture stubs the live models.dev/GitHub/Copilot/Anthropic inventory fetches that ACP server tests performed on every session create — test_server.py 147s → 3.4s, and the tests are now genuinely hermetic. - Sleep-based slowness shrunk where safe (codex_ttfb_watchdog, compression_concurrent_fork, etc.); no wall-clock assertion tightened. Verification: full hermetic suite via scripts/run_tests.sh — 2439 files, 31,130 tests passed, 0 failed, 0 flaky retries, 315s wall (baseline: 583s wall, 13,564s subprocess CPU).
112 lines
4 KiB
Python
112 lines
4 KiB
Python
"""Tests for MoA trace aggregator-output capture across streaming modes.
|
|
|
|
The MoA full-turn trace (opt-in ``moa.save_traces``) must record the
|
|
aggregator's acting output whether the aggregator ran non-streaming (inline
|
|
capture at call time) or streaming (captured after the fact from the caller's
|
|
resolved assistant text). Before the streamed-capture fix, a streamed
|
|
aggregator left ``output: null`` in the trace and only pointed at state.db,
|
|
so an offline audit of a benchmark run (which drives the streaming display
|
|
path via ``hermes chat --query``) couldn't see what the aggregator actually
|
|
produced without joining to the session DB by hand.
|
|
|
|
These exercise the real ``consume_and_save_trace`` → ``save_moa_turn`` path
|
|
with real file I/O against a temp HERMES_HOME — no mocks on the write path.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import json
|
|
|
|
import pytest
|
|
|
|
from agent.moa_loop import MoAChatCompletions
|
|
|
|
|
|
def _enable_traces(tmp_path, monkeypatch):
|
|
"""Point HERMES_HOME at a temp dir and turn moa.save_traces on."""
|
|
hermes_home = tmp_path / ".hermes"
|
|
hermes_home.mkdir()
|
|
monkeypatch.setenv("HERMES_HOME", str(hermes_home))
|
|
|
|
# save_moa_turn reads config via hermes_cli.config.load_config; stub it to
|
|
# return traces-on so the test doesn't depend on a real config file.
|
|
import agent.moa_trace as moa_trace
|
|
|
|
monkeypatch.setattr(
|
|
moa_trace,
|
|
"load_config",
|
|
lambda: {"moa": {"save_traces": True}},
|
|
raising=False,
|
|
)
|
|
# load_config is imported lazily inside _traces_enabled_and_dir; patch the
|
|
# source module attribute it imports from as well.
|
|
import hermes_cli.config as cfg
|
|
|
|
monkeypatch.setattr(
|
|
cfg, "load_config", lambda: {"moa": {"save_traces": True}}, raising=False
|
|
)
|
|
return hermes_home / "moa-traces"
|
|
|
|
|
|
def _make_completions_with_pending(streamed: bool, inline_output):
|
|
"""Build a MoAChatCompletions with a pending trace mimicking one turn."""
|
|
mc = MoAChatCompletions.__new__(MoAChatCompletions)
|
|
mc._pending_trace = {
|
|
"preset": "closed",
|
|
"reference_outputs": [], # references not under test here
|
|
"aggregator_label": "openrouter:anthropic/claude-opus-4.8",
|
|
"aggregator_slot": {
|
|
"model": "anthropic/claude-opus-4.8",
|
|
"provider": "openrouter",
|
|
},
|
|
"aggregator_temperature": 0.4,
|
|
"aggregator_input_messages": [
|
|
{"role": "system", "content": "sys"},
|
|
{"role": "user", "content": "do the thing"},
|
|
],
|
|
"aggregator_output": inline_output,
|
|
"aggregator_streamed": streamed,
|
|
}
|
|
return mc
|
|
|
|
|
|
def _read_single_trace(trace_dir, session_id):
|
|
path = trace_dir / f"{session_id}.jsonl"
|
|
assert path.exists(), f"trace file not written: {path}"
|
|
lines = path.read_text().strip().split("\n")
|
|
assert len(lines) == 1
|
|
return json.loads(lines[0])
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_streamed_without_fallback_points_to_session_db(tmp_path, monkeypatch):
|
|
"""Streaming turn with no resolvable text falls back to the state.db pointer."""
|
|
trace_dir = _enable_traces(tmp_path, monkeypatch)
|
|
mc = _make_completions_with_pending(streamed=True, inline_output=None)
|
|
|
|
mc.consume_and_save_trace("sess_nofb", aggregator_output_fallback=None)
|
|
|
|
rec = _read_single_trace(trace_dir, "sess_nofb")
|
|
agg = rec["aggregator"]
|
|
assert agg["streamed"] is True
|
|
assert agg["output"] is None
|
|
assert agg["output_location"] == "assistant_message_in_session_db"
|
|
|
|
|
|
def test_pending_trace_cleared_after_flush(tmp_path, monkeypatch):
|
|
"""A second flush is a no-op (pending cleared) — never double-writes."""
|
|
trace_dir = _enable_traces(tmp_path, monkeypatch)
|
|
mc = _make_completions_with_pending(streamed=True, inline_output=None)
|
|
|
|
mc.consume_and_save_trace("sess_once", aggregator_output_fallback="x")
|
|
# Second call: pending is None now, must not append a second line.
|
|
mc.consume_and_save_trace("sess_once", aggregator_output_fallback="y")
|
|
|
|
path = trace_dir / "sess_once.jsonl"
|
|
lines = path.read_text().strip().split("\n")
|
|
assert len(lines) == 1
|
|
|
|
|