Teknium
2a95791992
run_agent.py: delete the `# noqa: F401` re-export block (agent.process_bootstrap
OpenAI/_SafeWriter/_get_proxy_*, model_tools get_tool_definitions/
handle_function_call/check_toolset_requirements, FailoverReason,
_qwen_portal_headers/_routermint_headers, session_persistence names,
estimate_request_tokens_rough, ContextCompressor + friends, jittered_backoff,
prompt_builder names, message_sanitization names, tool_dispatch_helpers
names) — 41 names run_agent never used itself — and the `_STREAM_DIAG_HEADERS`
back-compat class alias (no in-tree reader). run_agent now imports only what
it uses (get_toolset_for_tool, is_local_endpoint, coalesce/uniquify tool-call
ids, cleanup_vm/get_active_env from terminal_tool_lifecycle).
agent/*: `_ra().X` late-binds that only reached a re-export now import the
defining module directly (agent_runtime_helpers -> process_bootstrap.OpenAI,
model_tools.handle_function_call, session_persistence._safe_session_filename_component;
agent_init -> model_tools.get_tool_definitions/check_toolset_requirements,
_lazy_headers("agent.client_lifecycle", ...) for qwen/routermint;
system_prompt -> agent.prompt_builder / model_tools directly, dropping its
own _ra() shim and the `_r` parameter threading). `_ra()` stays for
run_agent-resident names (logger, AIAgent, _hermes_home, _set_interrupt, ...).
toolsets.py: remove resolve_multiple_toolsets (shim-only, restored by
34abf954bd); tests/test_toolsets.py pins the same union behavior via
resolve_toolset over each name.
providers/__init__.py: drop the OMIT_TEMPERATURE re-export (no callers via the
package); ProviderProfile stays because __init__ uses it for annotations —
2 tests repointed to providers.base.
agent/iteration_budget.py: drop the "run_agent re-exports the class"
docstring pointer; 4 tests import IterationBudget from its home.
model_tools.py (arg_coercion names), agent/tool_executor.py, and
hermes_cli/cli_session_mixin.py repoints landed via a sibling commit on this
shared worktree.
Callers repointed: gateway/run.py, hermes_cli/cli_chat_turn_mixin.py,
hermes_cli/cli_tui_mixin.py, tui_gateway/session_workdir.py,
agent/transports/codex.py (one-line imports) + comment pointers in
tools/file_state.py, tools/schema_sanitizer.py, scripts/tool_search_livetest.py.
Tests: patch("run_agent.X") / monkeypatch.setattr(run_agent, "X") /
`from run_agent import X` -> defining module across 99 test files.
60 lines
1.9 KiB
Python
60 lines
1.9 KiB
Python
"""Tests for tool call argument repair in the streaming assembly path.
|
|
|
|
The streaming path (run_agent._call_chat_completions) assembles tool call
|
|
deltas into full arguments. When a model truncates or malforms the JSON
|
|
(e.g. GLM-5.1 via Ollama), the assembly path used to pass the broken JSON
|
|
straight through — setting has_truncated_tool_args but NOT repairing it.
|
|
That triggered the truncation handler to kill the session with /new required.
|
|
|
|
The fix: repair arguments in the streaming assembly path using
|
|
_repair_tool_call_arguments() so repairable malformations (trailing commas,
|
|
unclosed brackets, Python None) don't kill the session.
|
|
"""
|
|
|
|
import json
|
|
|
|
from agent.message_sanitization import _repair_tool_call_arguments
|
|
class TestStreamingAssemblyRepair:
|
|
"""Verify that _repair_tool_call_arguments is applied to streaming tool
|
|
call arguments before they're assembled into mock_tool_calls.
|
|
|
|
These tests verify the REPAIR FUNCTION itself works correctly for the
|
|
cases that arise during streaming assembly. Integration tests that
|
|
exercise the full streaming path are in run_agent.py's streaming tests.
|
|
"""
|
|
|
|
# -- Truncation cases (most common streaming failure) --
|
|
|
|
def test_truncated_object_no_close_brace(self):
|
|
"""Model stops mid-JSON, common with output length limits."""
|
|
raw = '{"command": "ls -la", "timeout": 30'
|
|
result = _repair_tool_call_arguments(raw, "terminal")
|
|
parsed = json.loads(result)
|
|
assert parsed["command"] == "ls -la"
|
|
assert parsed["timeout"] == 30
|
|
|
|
|
|
|
|
# -- Trailing comma cases (Ollama/GLM common) --
|
|
|
|
|
|
|
|
# -- Python None from model output --
|
|
|
|
|
|
# -- Empty arguments (some models emit empty string) --
|
|
|
|
def test_empty_string(self):
|
|
assert _repair_tool_call_arguments("", "test") == "{}"
|
|
|
|
|
|
# -- Already-valid JSON passes through unchanged --
|
|
|
|
|
|
# -- Extra closing brackets (rare but happens) --
|
|
|
|
|
|
# -- Real-world GLM-5.1 truncation pattern --
|
|
|
|
|