learnharness/src/server/chats.ts

import { abortExtraction, extractionStatuses } from "../agent/subagents/extraction.ts";
import { deleteChatTraces, type ChatDeleteMode, type ChatDeletionStats } from "../core/graph-deletion.ts";
import { titleFromPrompt } from "../core/contracts.ts";
import { graphRevision } from "./graph.ts";
import { stores } from "./stores.ts";
import { isTurnRunning } from "./turn-liveness.ts";

/**
 * Server-side chat entries for the transport layer (app/api/chats*,
 * app/chat/[id]). Today these delegate straight to the ChatStore port; the
 * services stage carves the real orchestration behind this same surface.
 *
 * deleteChatSession (wave 6, session deletion — the FIRST destructive
 * operation on the graph ledgers) is the one real orchestration here:
 * gates → extraction wind-down → one synchronous deletion block. See
 * core/graph-deletion.ts for the ledger semantics and the atomicity
 * argument.
 */

export const { createChat, getChat, listChats } = stores.chatStore;
export { titleFromPrompt };
export type { ChatDeleteMode, ChatDeletionStats } from "../core/graph-deletion.ts";

/** How long the wind-down wait may take before the caller gets a retryable
 *  503 (an aborted extraction ends at its next model-call boundary —
 *  provider cancellation is normally sub-second, but a wedged provider
 *  must never wedge deletion). */
const WIND_DOWN_TIMEOUT_MS = 10_000;
const WIND_DOWN_POLL_MS = 50;

/** What deleteChatSession did — the route maps it straight to HTTP. */
export type ChatDeleteOutcome =
	| { ok: false; status: 404; code: "not_found"; error: string }
	| { ok: false; status: 409; code: "turn_in_flight" | "jobs_running"; error: string; jobIds: string[] }
	| { ok: false; status: 503; code: "extraction_winding_down"; error: string }
	| {
			ok: true;
			mode: ChatDeleteMode;
			stats: ChatDeletionStats;
			jobsRemoved: number;
			graphRevision: number;
	  };

/**
 * Delete a session (owner rulings 2026-09-22/23): gates first, then the
 * purge. The owner-locked choreography:
 *
 *   1. GATES — a turn in flight or a running background job refuses with
 *      409 (they write the chat record and steer into it; deleting
 *      mid-flight would corrupt the write-ahead flow). The UI retries once
 *      work drains.
 *   2. EXTRACTION WIND-DOWN — the chat's own pipeline run is TRUE-
 *      CANCELLED (abortExtraction: spend stops, nothing merges), then
 *      awaited off the registry. A timeout is a retryable 503, never a
 *      forced race.
 *   3. THE SYNC BLOCK — everything below runs without a single await:
 *      no extraction append can interleave (the event loop never yields
 *      mid-block), so the read-filter-write of the ledger rewrites is
 *      atomic against the pipeline. Ledgers first, jobs next, the chat
 *      record LAST (a crash mid-block leaves a deletable chat with a
 *      purged graph — re-running DELETE fixes it — never a deleted chat
 *      with live traces). The pipeline's per-batch liveness guard
 *      (extraction.ts) closes the last window: a run that slipped past
 *      the wind-down finds the chat gone at its merge and aborts with
 *      nothing merged.
 */
export async function deleteChatSession(chatId: string, mode: ChatDeleteMode): Promise<ChatDeleteOutcome> {
	if (!stores.chatStore.getChat(chatId)) {
		return { ok: false, status: 404, code: "not_found", error: "No such session" };
	}
	if (isTurnRunning(chatId)) {
		return {
			ok: false,
			status: 409,
			code: "turn_in_flight",
			error: "A turn is running in this session — try again when it finishes",
			jobIds: [],
		};
	}
	const runningJobs = stores.jobStore.listJobs(chatId).filter((job) => job.status === "running");
	if (runningJobs.length > 0) {
		return {
			ok: false,
			status: 409,
			code: "jobs_running",
			error: "Background jobs are still running in this session — try again when they finish",
			jobIds: runningJobs.map((job) => job.id),
		};
	}
	if (extractionStatuses().some((run) => run.chatId === chatId)) {
		abortExtraction(chatId); // true cancellation — spend stops, nothing merges
		const deadline = Date.now() + WIND_DOWN_TIMEOUT_MS;
		while (extractionStatuses().some((run) => run.chatId === chatId)) {
			if (Date.now() > deadline) {
				return {
					ok: false,
					status: 503,
					code: "extraction_winding_down",
					error: "The session's extraction is still winding down — try again in a moment",
				};
			}
			await new Promise((resolve) => setTimeout(resolve, WIND_DOWN_POLL_MS));
		}
	}

	// THE SYNC BLOCK — no awaits below (see the doc comment). Order:
	// ledgers → jobs → the chat record.
	const stats = deleteChatTraces(stores.graphStore, chatId, mode);
	const jobsRemoved = stores.jobStore.deleteJobsForChat(chatId);
	stores.chatStore.deleteChat(chatId);
	return { ok: true, mode, stats, jobsRemoved, graphRevision: graphRevision() };
}