orca/src/main/ai-vault/session-scanner-opencode-sqlite.ts

import type { AiVaultSession, AiVaultSessionPreviewMessage } from '../../shared/ai-vault-types'
import {
  addPreviewMessage,
  createAccumulator,
  finalizeSession,
  updateTimeline
} from './session-scanner-accumulator'
import {
  normalizeFullFirstUserPromptText,
  shouldCaptureFullFirstUserPrompt
} from './session-scanner-first-user-prompt'
import { readOpenCodeDatabase } from './session-scanner-opencode-sqlite-open'
import { normalizeTitleText } from './session-scanner-values'
import type SyncDatabase from '../sqlite/sync-database'
import { columnExists, tableExists } from '../opencode-usage/schema-helpers'

// Why: OpenCode 1.17.x migrated session storage from per-session JSON files
// to a single SQLite DB at ~/.local/share/opencode/opencode.db. This module
// parses individual sessions from the DB into AiVaultSession objects. The
// discovery layer (listing candidates) lives in
// session-scanner-opencode-sqlite-discovery.ts.

const OPENCODE_SQLITE_PREVIEW_LIMIT = 5
// Why (#8864): a heavy session can hold ~10K parts (25-150 KB tool-output
// blobs). Join preview parts against only the newest N messages instead of
// scanning every part of the session, using the real (session_id, time_created,
// id) index. Bounds the read to those messages' parts; the 15 s parse timeout
// caps the residual for a single pathological giant part.
const OPENCODE_SQLITE_PREVIEW_MESSAGE_WINDOW = 100
// Bounds a pathological single message; a real typed prompt is a handful of parts.
const FIRST_USER_PROMPT_PART_LIMIT = 512

type SessionRow = {
  id: string
  title: string | null
  directory: string | null
  time_created: number
  time_updated: number
  model_json: string | null
  agent: string | null
  tokens_input: number
  tokens_output: number
  tokens_reasoning: number
  tokens_cache_read: number
  cost: number
  message_count: number
}

type PreviewRow = {
  role: string | null
  part_data: string
  time_created: number
  summary_title: string | null
  summary_body: string | null
}

function canReadOpenCodeSessions(db: SyncDatabase): boolean {
  return (
    tableExists(db, 'session') &&
    columnExists(db, 'session', 'time_created') &&
    columnExists(db, 'session', 'time_updated')
  )
}

function sessionColumnSelect(db: SyncDatabase, columnName: string): string {
  return columnExists(db, 'session', columnName) ? `s.${columnName}` : 'NULL'
}

function sessionNumberColumnSelect(db: SyncDatabase, columnName: string): string {
  return columnExists(db, 'session', columnName) ? `s.${columnName}` : '0'
}

function canCountOpenCodeMessages(db: SyncDatabase): boolean {
  return (
    tableExists(db, 'message') &&
    columnExists(db, 'message', 'session_id') &&
    columnExists(db, 'message', 'data')
  )
}

function buildSessionQuery(db: SyncDatabase): string {
  const messageCountSubquery = canCountOpenCodeMessages(db)
    ? `(SELECT COUNT(*) FROM message m
        WHERE m.session_id = s.id
          AND json_extract(m.data, '$.role') IN ('user','assistant'))`
    : '0'
  return `SELECT s.id,
                 ${sessionColumnSelect(db, 'title')} AS title,
                 ${sessionColumnSelect(db, 'directory')} AS directory,
                 s.time_created,
                 s.time_updated,
                 ${sessionColumnSelect(db, 'model')} AS model_json,
                 ${sessionColumnSelect(db, 'agent')} AS agent,
                 ${sessionNumberColumnSelect(db, 'tokens_input')} AS tokens_input,
                 ${sessionNumberColumnSelect(db, 'tokens_output')} AS tokens_output,
                 ${sessionNumberColumnSelect(db, 'tokens_reasoning')} AS tokens_reasoning,
                 ${sessionNumberColumnSelect(db, 'tokens_cache_read')} AS tokens_cache_read,
                 ${sessionNumberColumnSelect(db, 'cost')} AS cost,
                 ${messageCountSubquery} AS message_count
          FROM session s
          WHERE s.id = ?
          LIMIT 1`
}

function extractModelId(modelJson: string | null): string | null {
  if (!modelJson) {
    return null
  }
  try {
    const parsed = JSON.parse(modelJson) as unknown
    const record =
      parsed && typeof parsed === 'object' && !Array.isArray(parsed)
        ? (parsed as Record<string, unknown>)
        : null
    if (!record) {
      return null
    }
    // Why: OpenCode 1.17.x stores model as {"id":"glm-5.2","providerID":"..."}.
    // Older schemas used {"modelID":"..."}; accept both.
    return (
      (typeof record.id === 'string' && record.id.trim()) ||
      (typeof record.modelID === 'string' && record.modelID.trim()) ||
      null
    )
  } catch {
    return null
  }
}

function mapPreviewRole(role: string | null): AiVaultSessionPreviewMessage['role'] {
  if (role === 'user' || role === 'assistant' || role === 'system' || role === 'tool') {
    return role
  }
  return 'unknown'
}

function extractPartText(partData: string): string | null {
  try {
    const parsed = JSON.parse(partData) as unknown
    const record =
      parsed && typeof parsed === 'object' && !Array.isArray(parsed)
        ? (parsed as Record<string, unknown>)
        : null
    if (!record) {
      return null
    }
    if (typeof record.text === 'string') {
      return record.text
    }
    return null
  } catch {
    return null
  }
}

function readFirstUserPromptFromOpenCodeDb(db: SyncDatabase, sessionId: string): string | null {
  if (
    !canCountOpenCodeMessages(db) ||
    !tableExists(db, 'part') ||
    !columnExists(db, 'message', 'id') ||
    !columnExists(db, 'part', 'message_id') ||
    !columnExists(db, 'part', 'time_created') ||
    !columnExists(db, 'part', 'data')
  ) {
    return null
  }

  try {
    // Why: pin to the single earliest user message that actually has text parts,
    // then take all of its parts. Ordering parts across every user message would
    // pad a short first prompt with later turns and truncate a long one.
    const rows = db
      .prepare(
        `SELECT p.data AS part_data
         FROM part p
         WHERE p.message_id = (
                 SELECT m.id
                 FROM message m
                 JOIN part fp ON fp.message_id = m.id
                 WHERE m.session_id = ?
                   AND json_extract(m.data, '$.role') = 'user'
                   AND json_extract(fp.data, '$.type') = 'text'
                 ORDER BY m.time_created ASC, m.id ASC
                 LIMIT 1
               )
           AND json_extract(p.data, '$.type') = 'text'
         ORDER BY p.time_created ASC, p.rowid ASC
         LIMIT ${FIRST_USER_PROMPT_PART_LIMIT}`
      )
      .all(sessionId) as { part_data: string }[]

    const parts: string[] = []
    for (const row of rows) {
      const text = extractPartText(row.part_data)
      if (text) {
        parts.push(text)
      }
    }
    if (parts.length === 0) {
      return null
    }
    return normalizeFullFirstUserPromptText(parts.join('\n'))
  } catch {
    return null
  }
}

function buildPreviewQuery(db: SyncDatabase): string | null {
  if (
    !canCountOpenCodeMessages(db) ||
    !tableExists(db, 'part') ||
    !columnExists(db, 'message', 'id') ||
    !columnExists(db, 'part', 'message_id') ||
    !columnExists(db, 'part', 'time_created') ||
    !columnExists(db, 'part', 'data')
  ) {
    return null
  }
  return `SELECT json_extract(m.data, '$.role') AS role,
                 p.data AS part_data,
                 p.time_created,
                 json_extract(m.data, '$.summary.title') AS summary_title,
                 json_extract(m.data, '$.summary.body') AS summary_body
          FROM (SELECT id, data FROM message
                WHERE session_id = ?
                ORDER BY time_created DESC, id DESC
                LIMIT ${OPENCODE_SQLITE_PREVIEW_MESSAGE_WINDOW}) m
          JOIN part p ON p.message_id = m.id
          WHERE json_extract(m.data, '$.role') IN ('user','assistant')
            AND json_extract(p.data, '$.type') = 'text'
          ORDER BY p.time_created DESC
          LIMIT ?`
}

/**
 * Parse a single OpenCode session from the SQLite database into an
 * `AiVaultSession`. Reads session metadata (title, cwd, model, tokens, cost)
 * and up to 5 preview messages by joining the `message` and `part` tables.
 * The database is opened read-only with `PRAGMA query_only = ON` as a
 * belt-and-suspenders guard against mutations.
 * @param args.dbPath - Absolute path to the opencode.db file.
 * @param args.sessionId - The session ID (primary key in the `session` table).
 * @param args.platform - The platform to use for resume command generation.
 * @returns The parsed `AiVaultSession`, or `null` if the session does not exist
 *   or the database lacks the required schema.
 */
export async function parseOpenCodeSqliteSession(args: {
  dbPath: string
  sessionId: string
  platform: NodeJS.Platform
}): Promise<AiVaultSession | null> {
  return readOpenCodeDatabase({
    dbPath: args.dbPath,
    read: (db) => readSession({ db, ...args })
  })
}

// Extracted so the open wrapper owns the handle's lifetime.
function readSession(args: {
  db: SyncDatabase
  dbPath: string
  sessionId: string
  platform: NodeJS.Platform
}): AiVaultSession | null {
  const { db, dbPath, sessionId, platform } = args
  if (!canReadOpenCodeSessions(db)) {
    return null
  }
  const row = db.prepare(buildSessionQuery(db)).get(sessionId) as SessionRow | undefined
  if (!row || row.id !== sessionId) {
    return null
  }

  const mtimeMs =
    typeof row.time_updated === 'number' && row.time_updated > 0
      ? row.time_updated
      : row.time_created
  // Why: discovery uses a synthetic db#session path only for parser routing.
  // The UI's log open/reveal actions need a real filesystem path.
  const accumulator = createAccumulator({
    agent: 'opencode',
    file: {
      path: dbPath,
      mtimeMs,
      modifiedAt: new Date(mtimeMs).toISOString()
    },
    sessionId
  })
  accumulator.title = normalizeTitleText(row.title ?? '')
  accumulator.cwd = row.directory
  accumulator.model = extractModelId(row.model_json)
  accumulator.totalTokens =
    (row.tokens_input ?? 0) + (row.tokens_output ?? 0) + (row.tokens_reasoning ?? 0)
  accumulator.messageCount = row.message_count ?? 0
  updateTimeline(accumulator, row.time_created)
  updateTimeline(accumulator, row.time_updated)

  const previewSql = buildPreviewQuery(db)
  if (previewSql) {
    // Why: SQL already dropped anything older than the newest-N window, so the
    // accumulator never shifts and cannot detect the truncation itself. Ask for
    // one extra row so an exactly-full window is not mistaken for a trimmed one.
    const probedRows = db
      .prepare(previewSql)
      .all(sessionId, OPENCODE_SQLITE_PREVIEW_LIMIT + 1) as PreviewRow[]
    if (probedRows.length > OPENCODE_SQLITE_PREVIEW_LIMIT) {
      accumulator.previewMessagesTruncated = true
    }
    // Newest-first, so the probe row to drop is the oldest one at the tail.
    const previewRows = probedRows.slice(0, OPENCODE_SQLITE_PREVIEW_LIMIT)
    // Why: query returns newest-first; push in chronological order so the
    // accumulator's ring buffer keeps the newest OPENCODE_SQLITE_PREVIEW_LIMIT
    // messages.
    for (let i = previewRows.length - 1; i >= 0; i--) {
      const previewRow = previewRows[i]
      if (!previewRow) {
        continue
      }
      const text = extractPartText(previewRow.part_data)
      if (!text) {
        continue
      }
      addPreviewMessage(accumulator, {
        role: mapPreviewRole(previewRow.role),
        text,
        timestamp: previewRow.time_created,
        // Preview window is newest-N; first-prompt is loaded separately below.
        seedFirstUserPrompt: false
      })
      if (previewRow.role === 'user' && !accumulator.title) {
        accumulator.title =
          normalizeTitleText(previewRow.summary_title ?? '') ||
          normalizeTitleText(previewRow.summary_body ?? '')
      }
    }
  }

  // Why: list preview only joins the newest messages. On-demand copy needs the
  // session's earliest real user text part, not a later turn still in the window.
  if (shouldCaptureFullFirstUserPrompt()) {
    accumulator.firstUserPrompt = readFirstUserPromptFromOpenCodeDb(db, sessionId)
  }

  return finalizeSession(accumulator, platform)
}