nisse/nisse-dev/tests/test_loop_conversations.py

"""Conversations on the new loop (M2.5, PR A): the `general` scope, a
thread per conversation, idle close and continuation (L1, L5), and agreed
work filed as a record issue whose item starts as agreed (D1, L8). Real git
on the throwaway build host; a fake Gitea."""

import re
from typing import Any

from test_loop_milestones import DON, Forge, ScriptedModel, call, loop_agent
from test_loop_milestones import world as world  # noqa: PLC0414 -- the fixture

from nisse_core.gitea import Issue
from nisse_core.log.canonical import Fact, ToolResult
from nisse_dev.loop.agent import FILE_ISSUE
from nisse_dev.loop.items import CHAT, GO_AHEAD, MERGE, WAITING, loop_of

PROPOSAL = call(
    "propose_work",
    project="sandbox",
    title="Add count_lines",
    summary="Add count_lines(text) to text.py, counting newline-separated lines, "
    "with tests. First step: the function and its tests.",
)


def conversations(world: dict[str, Any]) -> list[Any]:
    return [i for i in world["store"].work_items() if i.scope == "general"]


def facts(world: dict[str, Any], item: Any) -> list[str]:
    return [
        e.event.text
        for e in world["store"].transcript(item.id, "orchestrator").read()
        if isinstance(e.event, Fact)
    ]


def quiet(world: dict[str, Any]) -> None:
    """No assigned issue: only conversations in these tests."""
    world["forge"].issues.clear()


def test_a_top_level_message_is_a_conversation_in_its_own_thread(
    world: dict[str, Any],
) -> None:
    quiet(world)
    model = ScriptedModel(
        [call("reply", text="A 4090 runs a 27B model at 4-bit comfortably.")],
        [call("reply", text="Yes, with a smaller context.")],
    )
    agent, transport = loop_agent(world, model)
    asked = transport.receive(DON, "Could a 4090 run a 27B model?")
    [chat] = conversations(world)
    loop = loop_of(chat)
    assert loop.thread == asked  # the thread is under the operator's message
    assert (chat.position, loop.wait) == (WAITING, CHAT)
    reply = transport.outbound[-1]
    assert reply.thread == asked and "4-bit" in reply.text
    assert 'conversation "Could a 4090 run a 27B model?": talking' in agent.status()

    transport.receive(DON, "And a 32B?", thread=asked)
    assert len(conversations(world)) == 1  # the same conversation
    assert transport.outbound[-1].text == "Yes, with a smaller context."


def test_agreed_work_is_filed_and_starts_as_agreed(world: dict[str, Any]) -> None:
    quiet(world)
    forge: Forge = world["forge"]
    model = ScriptedModel(
        [PROPOSAL],
        [  # the filed item's first run: straight to work, no restatement
            call("start_milestone", name="count_lines"),
            call("write_file", path="lines.py", content="L = 1\n"),
            call("finish_milestone", summary="count_lines", final="yes"),
        ],
    )
    _, transport = loop_agent(world, model)
    root = transport.receive(DON, "I'd like a line counter in the sandbox.")
    assert transport.outbound[-1].text.startswith(
        'rev-1: I\'d open an issue in don/sandbox, "Add count_lines"'
    )
    transport.receive(DON, "go", thread=root)

    [filed] = forge.issues
    marker = re.search(r"<!-- nisse-item: ([\w-]+) -->", filed.body)
    assert (
        marker and "count newline-separated lines" not in filed.body
    )  # it's the summary
    assert "counting newline-separated lines" in filed.body
    assert forge.assigned == [filed.number]
    work = world["store"].work_item(marker.group(1))
    assert work is not None and work.scope == "sandbox"
    assert loop_of(work).issue == filed.number and loop_of(work).started
    assert (work.position, loop_of(work).wait) == (WAITING, MERGE)  # it worked
    [pr] = forge.prs
    assert f"Closes #{filed.number}" in forge.prs[pr]["body"]
    # The conversation says where the work went and goes on.
    [chat] = conversations(world)
    assert loop_of(chat).became == (f"don/sandbox#{filed.number}",)
    assert loop_of(chat).wait == CHAT
    assert any("filed from the conversation" in f for f in facts(world, work))


def test_intake_never_takes_a_record_issue(world: dict[str, Any]) -> None:
    quiet(world)
    forge: Forge = world["forge"]
    body = "brief\n\n<!-- nisse-item: sandbox-cabc123 -->"
    forge.issues.append(Issue(30, "a record", body, "nisse-dev", ()))
    agent, _ = loop_agent(world, ScriptedModel())
    agent.intake()
    assert world["store"].work_items() == []


def test_a_proposal_for_an_unknown_project_is_sent_back(world: dict[str, Any]) -> None:
    quiet(world)
    model = ScriptedModel(
        [call("propose_work", project="nope", title="t", summary="s.")],
        [call("reply", text="Which project should it go in?")],
    )
    _, transport = loop_agent(world, model)
    transport.receive(DON, "Let's build a thing")
    results = [
        e.event.result
        for e in world["store"].transcript(conversations(world)[0].id, "driver").read()
        if isinstance(e.event, ToolResult)
    ]
    assert any("no project 'nope': the projects are sandbox" in r for r in results)
    assert loop_of(conversations(world)[0]).wait == CHAT


def test_changes_to_a_proposal_go_back_to_the_driver(world: dict[str, Any]) -> None:
    quiet(world)
    model = ScriptedModel([PROPOSAL], [PROPOSAL])
    _, transport = loop_agent(world, model)
    root = transport.receive(DON, "a line counter please")
    transport.receive(DON, "call it line_count instead", thread=root)
    loop = loop_of(conversations(world)[0])
    assert (loop.wait, loop.rev) == (GO_AHEAD, 2)
    assert world["forge"].issues == []


def test_an_idle_conversation_closes_and_a_reply_continues_it(
    world: dict[str, Any],
) -> None:
    quiet(world)
    model = ScriptedModel(
        [call("reply", text="Sure.")], [call("reply", text="Picking up again.")]
    )
    agent, transport = loop_agent(world, model)
    clock = [1000.0]
    agent._clock = lambda: clock[0]
    agent._idle = 3600
    root = transport.receive(DON, "thinking about a backup tool")
    clock[0] += 3599
    agent.intake()
    assert conversations(world)[0].status == "open"
    clock[0] += 1
    agent.intake()
    [old] = conversations(world)
    assert (old.status, old.outcome) == ("closed", "idle")

    transport.receive(DON, "back to that backup tool", thread=root)
    new = [c for c in conversations(world) if c.status == "open"]
    assert len(new) == 1 and loop_of(new[0]).follows == old.id
    assert loop_of(new[0]).thread == root
    assert any("this continues the conversation" in f for f in facts(world, new[0]))


def test_a_plain_yes_files_the_work(world: dict[str, Any]) -> None:
    quiet(world)

    class Reader:
        def agrees(self, proposal: str, reply: str, model: str = "") -> bool:
            return reply == "sounds right"

    model = ScriptedModel([PROPOSAL], [call("ask_human", question="anything else?")])
    agent, transport = loop_agent(world, model)
    agent._reader = Reader()
    root = transport.receive(DON, "a line counter please")
    transport.receive(DON, "sounds right", thread=root)
    assert len(world["forge"].issues) == 1


def test_a_filing_cut_off_by_a_crash_is_settled_at_boot(world: dict[str, Any]) -> None:
    quiet(world)
    forge: Forge = world["forge"]
    model = ScriptedModel([PROPOSAL])
    _, transport = loop_agent(world, model)
    root = transport.receive(DON, "a line counter please")
    chat = conversations(world)[0]
    store = world["store"]
    # The issue was opened, then the agent died before its item existed.
    intent = store.begin_intent(
        FILE_ISSUE,
        {"repo": "don/sandbox", "item": "sandbox-cdead01", "title": "Add count_lines"},
        cause="filing",
        work_item=chat.id,
    )
    forge.issues.append(
        Issue(
            40, "Add count_lines", "b\n\n<!-- nisse-item: sandbox-cdead01 -->", "x", ()
        )
    )
    agent2, _ = loop_agent(world, ScriptedModel([call("ask_human", question="ok?")]))
    agent2.boot()
    work = store.work_item("sandbox-cdead01")
    assert work is not None and loop_of(work).issue == 40 and loop_of(work).started
    assert not [i for i in store.pending_intents() if i.id == intent]
    assert root  # the conversation itself is untouched


def test_a_filing_that_never_reached_gitea_is_put_up_again(
    world: dict[str, Any],
) -> None:
    quiet(world)
    _, transport = loop_agent(world, ScriptedModel([PROPOSAL]))
    transport.receive(DON, "a line counter please")
    chat = conversations(world)[0]
    world["store"].begin_intent(
        FILE_ISSUE,
        {"repo": "don/sandbox", "item": "sandbox-cnever1", "title": "Add count_lines"},
        cause="filing",
        work_item=chat.id,
    )
    agent2, _ = loop_agent(world, ScriptedModel())
    agent2.boot()
    loop = loop_of(world["store"].work_item(chat.id))
    assert loop.wait == GO_AHEAD and "cut off by a restart" in loop.note
    assert world["store"].work_item("sandbox-cnever1") is None


def test_commands_alone_at_top_level_are_not_conversations(
    world: dict[str, Any],
) -> None:
    quiet(world)
    _, transport = loop_agent(world, ScriptedModel())
    transport.receive(DON, "status")
    transport.receive(DON, "go")
    assert conversations(world) == []