huskies/server/src/agents/pool/auto_assign/merge_failure_subscriber.rs

//! TransitionFired subscriber that auto-spawns mergemaster on ConflictDetected merge failures.
//!
//! Listens on the pipeline transition broadcast channel and schedules a
//! mergemaster agent whenever a story enters
//! `Stage::MergeFailure { kind: ConflictDetected(_), .. }`.
//! Other [`MergeFailureKind`] variants require human intervention and are
//! intentionally ignored here.

use std::path::{Path, PathBuf};
use std::sync::Arc;

use crate::pipeline_state::{MergeFailureKind, Stage};
use crate::slog;
use crate::slog_warn;

use super::super::super::PipelineStage;
use super::super::AgentPool;
use super::scan::{find_free_agent_for_stage, is_story_assigned_for_stage};

/// Spawn a background task that auto-spawns mergemaster agents on
/// `Stage::MergeFailure { kind: ConflictDetected(_) }` transitions.
///
/// The task subscribes to the pipeline transition broadcast channel and calls
/// [`AgentPool::start_agent`] with the first free mergemaster agent whenever a
/// story transitions into a recoverable conflict state.  All other
/// [`MergeFailureKind`] variants are silently skipped — they need a human.
pub(crate) fn spawn_merge_failure_subscriber(pool: Arc<AgentPool>, project_root: PathBuf) {
    let mut rx = crate::pipeline_state::subscribe_transitions();
    tokio::spawn(async move {
        loop {
            match rx.recv().await {
                Ok(fired) => {
                    on_merge_failure_transition(&pool, &project_root, &fired).await;
                }
                Err(tokio::sync::broadcast::error::RecvError::Lagged(n)) => {
                    slog_warn!(
                        "[merge-failure-sub] Subscriber lagged, skipped {n} event(s). \
                         ConflictDetected stories may need manual mergemaster spawn."
                    );
                }
                Err(tokio::sync::broadcast::error::RecvError::Closed) => break,
            }
        }
    });
}

async fn on_merge_failure_transition(
    pool: &AgentPool,
    project_root: &Path,
    fired: &crate::pipeline_state::TransitionFired,
) {
    let Stage::MergeFailure { ref kind, .. } = fired.after else {
        return;
    };

    let story_id = &fired.story_id.0;

    match kind {
        MergeFailureKind::ConflictDetected(_) => {
            let config = match crate::config::ProjectConfig::load(project_root) {
                Ok(c) => c,
                Err(e) => {
                    slog_warn!("[merge-failure-sub] Failed to load config for '{story_id}': {e}");
                    return;
                }
            };

            let agent_name = {
                let agents = match pool.agents.lock() {
                    Ok(a) => a,
                    Err(e) => {
                        slog_warn!(
                            "[merge-failure-sub] Failed to lock agent pool for '{story_id}': {e}"
                        );
                        return;
                    }
                };
                if is_story_assigned_for_stage(
                    &config,
                    &agents,
                    story_id,
                    &PipelineStage::Mergemaster,
                ) {
                    return; // mergemaster already running for this story
                }
                find_free_agent_for_stage(&config, &agents, &PipelineStage::Mergemaster)
                    .map(str::to_string)
            };

            if let Some(agent) = agent_name {
                slog!(
                    "[merge-failure-sub] ConflictDetected on '{story_id}'; \
                     auto-spawning mergemaster '{agent}'."
                );
                if let Err(e) = pool
                    .start_agent(project_root, story_id, Some(&agent), None, None)
                    .await
                {
                    slog!("[merge-failure-sub] Failed to spawn '{agent}' for '{story_id}': {e}");
                }
            } else {
                slog!(
                    "[merge-failure-sub] ConflictDetected on '{story_id}'; \
                     no free mergemaster agent available."
                );
            }
        }
        // GatesFailed, EmptyDiff, NoCommits, Other — all require human intervention.
        MergeFailureKind::GatesFailed(_)
        | MergeFailureKind::EmptyDiff
        | MergeFailureKind::NoCommits
        | MergeFailureKind::Other(_) => {}
    }
}

// ── Tests ──────────────────────────────────────────────────────────────────

#[cfg(test)]
mod tests {
    use super::*;
    use crate::agents::{AgentPool, AgentStatus};
    use crate::io::watcher::WatcherEvent;
    use std::sync::Arc;
    use tokio::sync::broadcast;

    fn setup_project(tmp: &tempfile::TempDir) {
        let sk = tmp.path().join(".huskies");
        std::fs::create_dir_all(&sk).unwrap();
        std::fs::write(
            sk.join("project.toml"),
            "[[agent]]\nname = \"mergemaster\"\nstage = \"mergemaster\"\n",
        )
        .unwrap();
    }

    fn seed_at_merge(story_id: &str) {
        crate::crdt_state::init_for_test();
        crate::db::ensure_content_store();
        crate::db::write_item_with_content(
            story_id,
            "4_merge",
            "---\nname: Test\n---\n",
            crate::db::ItemMeta::named("Test"),
        );
    }

    fn make_pool(port: u16) -> Arc<AgentPool> {
        let (tx, _) = broadcast::channel::<WatcherEvent>(4);
        Arc::new(AgentPool::new(port, tx))
    }

    fn make_fired(
        story_id: &str,
        kind: MergeFailureKind,
    ) -> crate::pipeline_state::TransitionFired {
        use crate::pipeline_state::{BranchName, PipelineEvent, StoryId, TransitionFired};
        use std::num::NonZeroU32;
        TransitionFired {
            story_id: StoryId(story_id.to_string()),
            before: crate::pipeline_state::Stage::Merge {
                feature_branch: BranchName("feature/test".to_string()),
                commits_ahead: NonZeroU32::new(1).unwrap(),
                claim: None,
            },
            after: crate::pipeline_state::Stage::MergeFailure {
                kind: kind.clone(),
                feature_branch: BranchName("feature/test".to_string()),
                commits_ahead: NonZeroU32::new(1).unwrap(),
            },
            event: PipelineEvent::MergeFailed { kind },
            at: chrono::Utc::now(),
        }
    }

    // ── AC4: each MergeFailureKind variant ──────────────────────────────────

    /// ConflictDetected → on_merge_failure_transition must spawn mergemaster.
    ///
    /// Calls the handler directly (not via the broadcast subscriber) to avoid
    /// cross-test channel contamination from the global TRANSITION_TX.
    #[tokio::test]
    async fn conflict_detected_spawns_mergemaster_via_subscriber() {
        let tmp = tempfile::tempdir().unwrap();
        setup_project(&tmp);
        let story_id = "998_sub_conflict";
        seed_at_merge(story_id);

        let pool = make_pool(3998);
        let fired = make_fired(
            story_id,
            MergeFailureKind::ConflictDetected(Some("CONFLICT (content): src/lib.rs".to_string())),
        );
        on_merge_failure_transition(&pool, tmp.path(), &fired).await;

        let agents = pool.agents.lock().unwrap();
        assert!(
            agents.iter().any(|(key, a)| {
                key.contains(story_id)
                    && a.agent_name == "mergemaster"
                    && matches!(a.status, AgentStatus::Pending | AgentStatus::Running)
            }),
            "mergemaster must be spawned for ConflictDetected"
        );
    }

    /// GatesFailed → subscriber must NOT spawn mergemaster (human intervention needed).
    #[tokio::test]
    async fn gates_failed_does_not_spawn_mergemaster() {
        let tmp = tempfile::tempdir().unwrap();
        setup_project(&tmp);
        let story_id = "998_sub_gates";
        seed_at_merge(story_id);

        let pool = make_pool(3997);
        spawn_merge_failure_subscriber(Arc::clone(&pool), tmp.path().to_path_buf());

        crate::agents::lifecycle::transition_to_merge_failure(
            story_id,
            MergeFailureKind::GatesFailed("error[E0308]: mismatched types".to_string()),
        )
        .expect("transition must succeed");

        // Give the subscriber time to run (it should do nothing).
        tokio::time::sleep(std::time::Duration::from_millis(100)).await;

        let agents = pool.agents.lock().unwrap();
        let spawned = agents.iter().any(|(key, a)| {
            key.contains(story_id)
                && a.agent_name == "mergemaster"
                && matches!(a.status, AgentStatus::Pending | AgentStatus::Running)
        });
        assert!(!spawned, "mergemaster must NOT be spawned for GatesFailed");
    }

    /// EmptyDiff → subscriber must NOT spawn mergemaster.
    #[tokio::test]
    async fn empty_diff_does_not_spawn_mergemaster() {
        let tmp = tempfile::tempdir().unwrap();
        setup_project(&tmp);
        let story_id = "998_sub_emptydiff";
        seed_at_merge(story_id);

        let pool = make_pool(3996);
        spawn_merge_failure_subscriber(Arc::clone(&pool), tmp.path().to_path_buf());

        crate::agents::lifecycle::transition_to_merge_failure(
            story_id,
            MergeFailureKind::EmptyDiff,
        )
        .expect("transition must succeed");

        tokio::time::sleep(std::time::Duration::from_millis(100)).await;

        let agents = pool.agents.lock().unwrap();
        let spawned = agents.iter().any(|(key, a)| {
            key.contains(story_id)
                && a.agent_name == "mergemaster"
                && matches!(a.status, AgentStatus::Pending | AgentStatus::Running)
        });
        assert!(!spawned, "mergemaster must NOT be spawned for EmptyDiff");
    }

    /// NoCommits → subscriber must NOT spawn mergemaster.
    #[tokio::test]
    async fn no_commits_does_not_spawn_mergemaster() {
        let tmp = tempfile::tempdir().unwrap();
        setup_project(&tmp);
        let story_id = "998_sub_nocommits";
        seed_at_merge(story_id);

        let pool = make_pool(3995);
        spawn_merge_failure_subscriber(Arc::clone(&pool), tmp.path().to_path_buf());

        crate::agents::lifecycle::transition_to_merge_failure(
            story_id,
            MergeFailureKind::NoCommits,
        )
        .expect("transition must succeed");

        tokio::time::sleep(std::time::Duration::from_millis(100)).await;

        let agents = pool.agents.lock().unwrap();
        let spawned = agents.iter().any(|(key, a)| {
            key.contains(story_id)
                && a.agent_name == "mergemaster"
                && matches!(a.status, AgentStatus::Pending | AgentStatus::Running)
        });
        assert!(!spawned, "mergemaster must NOT be spawned for NoCommits");
    }

    /// Other(_) → subscriber must NOT spawn mergemaster.
    #[tokio::test]
    async fn other_does_not_spawn_mergemaster() {
        let tmp = tempfile::tempdir().unwrap();
        setup_project(&tmp);
        let story_id = "998_sub_other";
        seed_at_merge(story_id);

        let pool = make_pool(3994);
        spawn_merge_failure_subscriber(Arc::clone(&pool), tmp.path().to_path_buf());

        crate::agents::lifecycle::transition_to_merge_failure(
            story_id,
            MergeFailureKind::Other("unknown error".to_string()),
        )
        .expect("transition must succeed");

        tokio::time::sleep(std::time::Duration::from_millis(100)).await;

        let agents = pool.agents.lock().unwrap();
        let spawned = agents.iter().any(|(key, a)| {
            key.contains(story_id)
                && a.agent_name == "mergemaster"
                && matches!(a.status, AgentStatus::Pending | AgentStatus::Running)
        });
        assert!(!spawned, "mergemaster must NOT be spawned for Other");
    }

    /// ConflictDetected self-loop — handler must NOT spawn a second mergemaster
    /// when one is already Pending/Running for the story.
    ///
    /// Calls the handler twice directly (no broadcast subscriber) so there is no
    /// timing window: the first call sets the agent to Pending synchronously,
    /// and the second call sees that Pending entry and returns early.
    #[tokio::test]
    async fn conflict_detected_self_loop_does_not_double_spawn() {
        let tmp = tempfile::tempdir().unwrap();
        setup_project(&tmp);
        let story_id = "998_sub_selfloop";
        seed_at_merge(story_id);

        let pool = make_pool(3993);
        let fired = make_fired(
            story_id,
            MergeFailureKind::ConflictDetected(Some("CONFLICT".to_string())),
        );

        // First call — spawns mergemaster (agent enters Pending).
        on_merge_failure_transition(&pool, tmp.path(), &fired).await;
        {
            let agents = pool.agents.lock().unwrap();
            assert!(
                agents.iter().any(|(key, a)| {
                    key.contains(story_id)
                        && a.agent_name == "mergemaster"
                        && matches!(a.status, AgentStatus::Pending | AgentStatus::Running)
                }),
                "mergemaster must be Pending after first ConflictDetected"
            );
        }

        // Second call (self-loop) — agent is still Pending; guard must prevent double-spawn.
        on_merge_failure_transition(&pool, tmp.path(), &fired).await;

        let agents = pool.agents.lock().unwrap();
        let active_count = agents
            .iter()
            .filter(|(key, a)| {
                key.contains(story_id)
                    && a.agent_name == "mergemaster"
                    && matches!(a.status, AgentStatus::Pending | AgentStatus::Running)
            })
            .count();
        assert_eq!(
            active_count, 1,
            "mergemaster must not be double-spawned on ConflictDetected self-loop"
        );
    }
}