wehub-resource-sync
119 lines
4.4 KiB
Python
119 lines
4.4 KiB
Python
"""E2E: /model command in the Omnigent REPL under pexpect.
|
||
|
||
Migrated to mock LLM: drives ``/model`` against a mock ``omnigent run``
|
||
REPL and asserts the slash-command surface — show / set / show-after-set
|
||
/ reset — matches the design's contract end-to-end. No real Databricks
|
||
credentials required.
|
||
"""
|
||
|
||
from __future__ import annotations
|
||
|
||
from pathlib import Path
|
||
|
||
from tests.e2e.omnigent._pexpect_harness import (
|
||
clean_exit,
|
||
spawn_omnigent_run,
|
||
submit_prompt,
|
||
)
|
||
from tests.e2e.omnigent.conftest import configure_mock_llm
|
||
|
||
_MODEL = "mock-repl-model"
|
||
_HARNESS = "openai-agents"
|
||
# Override target. Any non-empty model id distinct from the spawn one
|
||
# is sufficient for the show/set assertions.
|
||
_OVERRIDE_MODEL = "mock-repl-model-override"
|
||
_SPAWN_TIMEOUT = 90.0
|
||
_BOOT_TIMEOUT = 60.0
|
||
_EXIT_TIMEOUT = 15.0
|
||
|
||
|
||
def _submit_slash_command(child, text: str) -> None: # type: ignore[no-untyped-def]
|
||
"""Submit a slash command under prompt-toolkit/pexpect."""
|
||
submit_prompt(child, text)
|
||
|
||
|
||
def test_repl_model_command_show_set_reset(
|
||
omnigent_python: Path,
|
||
omnigent_repo_root: Path,
|
||
mock_credentials_env: dict[str, str],
|
||
mock_llm_server_url: str,
|
||
) -> None:
|
||
"""Drive /model through its full state machine in the REPL.
|
||
|
||
Uses the mock LLM server so no real Databricks credentials are
|
||
needed. The /model slash commands are handled entirely within the
|
||
REPL process — no LLM turn is required to test the UI state machine.
|
||
|
||
Asserts each transition:
|
||
|
||
1. Initial ``/model`` renders the ``Active:`` readout with no
|
||
in-session override (``no model pinned``).
|
||
2. ``/model <name>`` confirms ``model set to <name>``.
|
||
3. Subsequent ``/model`` shows ``<name>`` in the ``Active:`` readout
|
||
(session override persisted).
|
||
4. ``/model default`` confirms ``model reset to agent default``.
|
||
|
||
:param omnigent_python: Interpreter with omnigent installed.
|
||
:param omnigent_repo_root: Working directory for the subprocess.
|
||
:param mock_credentials_env: Mock-LLM env vars.
|
||
:param mock_llm_server_url: Mock server URL.
|
||
"""
|
||
configure_mock_llm(
|
||
mock_llm_server_url,
|
||
[{"text": "ok"}],
|
||
key=_MODEL,
|
||
)
|
||
yaml_path = omnigent_repo_root / "tests" / "resources" / "examples" / "hello_world.yaml"
|
||
child = spawn_omnigent_run(
|
||
omnigent_python=omnigent_python,
|
||
yaml_path=yaml_path,
|
||
model=_MODEL,
|
||
harness=_HARNESS,
|
||
env=mock_credentials_env,
|
||
cwd=omnigent_repo_root,
|
||
timeout=_SPAWN_TIMEOUT,
|
||
)
|
||
try:
|
||
# Match the visible prompt marker rather than the bottom-
|
||
# toolbar state text: under pexpect the prompt-toolkit CPR
|
||
# handshake can suppress ``state: sleeping`` even when the
|
||
# REPL is ready.
|
||
child.expect(r"❯ ", timeout=_BOOT_TIMEOUT)
|
||
|
||
# No-arg /model renders the active-credential readout:
|
||
# "Active: <model | (no model pinned …)> · <provider> · <source>"
|
||
# (this replaced the legacy "model: (agent default)" line; see
|
||
# _build_model_readout_lines in omnigent/repl/_repl.py). With no
|
||
# in-session override yet, the model slot reads "no model pinned" —
|
||
# ``--model`` sets the routing model, not the /model session override
|
||
# the readout tracks (``session.model_override``).
|
||
_submit_slash_command(child, "/model")
|
||
child.expect("Active:", timeout=10)
|
||
child.expect("no model pinned", timeout=10)
|
||
|
||
_submit_slash_command(child, f"/model {_OVERRIDE_MODEL}")
|
||
# Set confirmation: "model set to <name> for future responses".
|
||
child.expect(f"model set to {_OVERRIDE_MODEL}", timeout=10)
|
||
|
||
# Nudge a fresh prompt to confirm the input buffer is clear before
|
||
# the next slash command.
|
||
child.send("\r")
|
||
child.expect(r"❯ ", timeout=10)
|
||
|
||
# After the set, the readout's model slot shows the override (prior
|
||
# occurrences — the set confirmation — are already consumed, so the
|
||
# next match is the fresh readout line).
|
||
_submit_slash_command(child, "/model")
|
||
child.expect("Active:", timeout=10)
|
||
child.expect(_OVERRIDE_MODEL, timeout=10)
|
||
|
||
_submit_slash_command(child, "/model default")
|
||
child.expect("model reset to agent default", timeout=10)
|
||
|
||
clean_exit(child, timeout=_EXIT_TIMEOUT)
|
||
assert child.exitstatus in (0, None)
|
||
assert child.signalstatus is None
|
||
finally:
|
||
if not child.closed:
|
||
child.close(force=True)
|