pantheon-opencode 1.5.0-beta.6 → 1.5.0-beta.8

package/src/pantheon/file-persistence-salvage.ts

@@ -0,0 +1,64 @@
+/**
+ * Best-effort salvage of complete top-level JSON objects from a corrupt
+ * `state.json`. The file is normally a JSON array; a torn write can leave a
+ * truncated tail (or interleaved NUL bytes). We scan for balanced `{…}`
+ * objects, parsing each independently, so every intact record survives.
+ *
+ * Extracted from file-persistence.ts to keep that module within the 300-line
+ * ceiling. Exported for tests.
+ */
+import type { BackgroundJobRecord } from './background-job-board.ts'
+
+export function salvageRecords(content: string): BackgroundJobRecord[] {
+  const records: BackgroundJobRecord[] = []
+  let depth = 0
+  let start = -1
+  let inString = false
+  let escaped = false
+
+  for (let i = 0; i < content.length; i++) {
+    const ch = content[i]
+
+    if (inString) {
+      if (escaped) {
+        escaped = false
+      } else if (ch === '\\') {
+        escaped = true
+      } else if (ch === '"') {
+        inString = false
+      }
+      continue
+    }
+
+    if (ch === '"') {
+      inString = true
+      continue
+    }
+    if (ch === '{') {
+      if (depth === 0) start = i
+      depth++
+      continue
+    }
+    if (ch === '}') {
+      if (depth > 0) depth--
+      if (depth === 0 && start >= 0) {
+        const candidate = content.slice(start, i + 1)
+        try {
+          const parsed: unknown = JSON.parse(candidate)
+          if (
+            parsed !== null &&
+            typeof parsed === 'object' &&
+            typeof (parsed as BackgroundJobRecord).taskID === 'string'
+          ) {
+            records.push(parsed as BackgroundJobRecord)
+          }
+        } catch {
+          // Incomplete/partial object — skip it, keep scanning.
+        }
+        start = -1
+      }
+    }
+  }
+
+  return records
+}

package/src/pantheon/file-persistence.ts

@@ -1,91 +1,238 @@
 /**
  * File-based persistence adapter for the BackgroundJobBoard.
  *
- * Stores all jobs in a single `state.json` file using atomic writes
- * (write to `state.json.tmp` then rename) to prevent partial writes
- * from crashes.
+ * Stores all jobs in a single `state.json` file. Writes are crash-safe:
+ * the payload is written to a UNIQUE temp file, fsynced, then renamed over
+ * the destination. Writes to the same path are serialized through an
+ * in-process queue so two logical writers (plugin × native) cannot
+ * interleave and corrupt the JSON or race the rename.
+ *
+ * Reads are corruption-tolerant: a malformed `state.json` is moved aside to
+ * `<state>.corrupt-<ts>`, a clear log is emitted, and whatever complete
+ * records can be salvaged from the array are returned — never a throw, so
+ * callers (delete/prune/recover) cannot get stuck in a retry loop.
  *
  * @module file-persistence
  */
 
-import { mkdir, readFile, rename, writeFile } from 'node:fs/promises'
-import { dirname } from 'node:path'
+import { mkdir, open, readFile, rename, unlink } from 'node:fs/promises'
+import { dirname, resolve } from 'node:path'
 
 import type { BackgroundJobRecord, PersistenceAdapter } from './background-job-board.ts'
+import { salvageRecords } from './file-persistence-salvage.ts'
+import { createPantheonLogger } from './logger.ts'
+
+export { salvageRecords } from './file-persistence-salvage.ts'
+
+const log = createPantheonLogger({ module: 'FilePersistence' })
 
 /** Default path for the state file. */
 const DEFAULT_STATE_PATH = '.pantheon/board/state.json'
 
+/** Max rename attempts after an ENOENT before giving up (dir/tmp recreated each try). */
+const RENAME_MAX_ATTEMPTS = 3
+
+/** Minimal file handle surface used to fsync the temp file before rename. */
+export interface PersistenceFileHandle {
+  writeFile(data: string, encoding: 'utf-8'): Promise<void>
+  sync(): Promise<void>
+  close(): Promise<void>
+}
+
+/**
+ * Filesystem operations used by the adapter. Injectable so tests can fault
+ * inject a failing `rename` (ENOENT) deterministically.
+ */
+export interface PersistenceFsOps {
+  mkdir(dir: string): Promise<void>
+  readFile(path: string, encoding: 'utf-8'): Promise<string>
+  rename(from: string, to: string): Promise<void>
+  openForWrite(path: string): Promise<PersistenceFileHandle>
+  unlink(path: string): Promise<void>
+}
+
+/** Real Node.js filesystem implementation (the default). */
+const defaultFsOps: PersistenceFsOps = {
+  mkdir: async (dir) => {
+    await mkdir(dir, { recursive: true })
+  },
+  readFile: (path, encoding) => readFile(path, encoding),
+  rename: (from, to) => rename(from, to),
+  openForWrite: (path) => open(path, 'w'),
+  unlink: (path) => unlink(path),
+}
+
+function isEnoent(err: unknown): boolean {
+  return err instanceof Error && 'code' in err && (err as NodeJS.ErrnoException).code === 'ENOENT'
+}
+
+/**
+ * In-process write queues keyed by absolute state path. Shared across
+ * adapter INSTANCES so a plugin adapter and a native adapter pointing at the
+ * same file serialize their read-modify-write cycles.
+ */
+const writeQueues = new Map<string, Promise<unknown>>()
+
+function enqueueWrite<T>(key: string, task: () => Promise<T>): Promise<T> {
+  const previous = writeQueues.get(key) ?? Promise.resolve()
+  const run = previous.then(task, task)
+  writeQueues.set(
+    key,
+    run.then(
+      () => undefined,
+      () => undefined,
+    ),
+  )
+  return run
+}
+
 /**
  * File-based persistence adapter.
  *
- * All jobs are stored in a single JSON array in `statePath`.
- * Writes are atomic: content is written to a `.tmp` file, then renamed.
- *
- * The board's in-memory `this.jobs` Map is the source of truth —
- * this file is a crash-recovery snapshot. `deleteJob()` removes the
- * record from the snapshot so pruned/reconciled jobs don't resurface
- * after a restart.
+ * All jobs are stored in a single JSON array in `statePath`. Writes are
+ * atomic and serialized (unique tmp + fsync + rename, with ENOENT retry).
+ * Corrupt state is backed up and salvaged instead of throwing.
  */
 export class FilePersistenceAdapter implements PersistenceAdapter {
   private readonly statePath: string
   private readonly dir: string
+  private readonly ops: PersistenceFsOps
+  private readonly lockKey: string
+  private tmpCounter = 0
 
   /**
    * @param statePath  Path to the state JSON file (default: `.pantheon/board/state.json`)
+   * @param ops        Filesystem operations (defaults to real Node.js fs; injectable for tests)
    */
-  constructor(statePath?: string) {
+  constructor(statePath?: string, ops: PersistenceFsOps = defaultFsOps) {
     this.statePath = statePath ?? DEFAULT_STATE_PATH
     this.dir = dirname(this.statePath)
+    this.ops = ops
+    this.lockKey = resolve(this.statePath)
   }
 
   /** Ensure the parent directory exists. */
   private async ensureDir(): Promise<void> {
-    await mkdir(this.dir, { recursive: true })
+    await this.ops.mkdir(this.dir)
   }
 
-  /** Path to the atomic-write temp file. */
-  private get tmpPath(): string {
-    return this.statePath + '.tmp'
+  /** Unique temp path per write — prevents two writers racing the same tmp. */
+  private makeTmpPath(): string {
+    this.tmpCounter += 1
+    const unique = `${process.pid}.${this.tmpCounter}.${Math.random().toString(36).slice(2, 8)}`
+    return `${this.statePath}.${unique}.tmp`
   }
 
   /**
    * Read all jobs from the state file.
-   * Returns `null` if the file doesn't exist (first run / no jobs yet).
+   *
+   * Return contract distinguishes three failure modes:
+   * - absent file (ENOENT) → `[]` (first run / no jobs yet);
+   * - corrupt content → quarantine + `salvageRecords()` (best-effort recovery);
+   * - non-ENOENT IO error (e.g. EIO) → `null`, signalling the read failed.
+   *
+   * Callers doing read-modify-write MUST abort on `null` so a failed read
+   * cannot overwrite the on-disk state with a partial snapshot.
    */
   private async readState(): Promise<BackgroundJobRecord[] | null> {
+    let content: string
     try {
-      const content = await readFile(this.statePath, 'utf-8')
-      return JSON.parse(content) as BackgroundJobRecord[]
+      content = await this.ops.readFile(this.statePath, 'utf-8')
     } catch (err: unknown) {
-      if (
-        err instanceof Error &&
-        'code' in err &&
-        (err as NodeJS.ErrnoException).code === 'ENOENT'
-      ) {
-        return null
+      if (isEnoent(err)) return []
+      log.error(`[FilePersistence] Failed to read ${this.statePath}:`, err)
+      return null
+    }
+
+    try {
+      const parsed: unknown = JSON.parse(content)
+      if (!Array.isArray(parsed)) {
+        throw new Error('state.json is not a JSON array')
       }
-      throw err
+      return parsed as BackgroundJobRecord[]
+    } catch (err: unknown) {
+      await this.quarantineCorrupt(err)
+      const salvaged = salvageRecords(content)
+      log.warn(
+        `[FilePersistence] Corrupt ${this.statePath} — recovered ${salvaged.length} record(s)`,
+      )
+      return salvaged
+    }
+  }
+
+  /** Move a corrupt state.json aside so the next write starts from a clean file. */
+  private async quarantineCorrupt(cause: unknown): Promise<void> {
+    const backupPath = `${this.statePath}.corrupt-${Date.now()}`
+    try {
+      await this.ops.rename(this.statePath, backupPath)
+      log.error(`[FilePersistence] Corrupt ${this.statePath} backed up to ${backupPath}:`, cause)
+    } catch (err: unknown) {
+      log.error(
+        `[FilePersistence] Could not back up corrupt ${this.statePath} — continuing with salvaged state:`,
+        err,
+      )
     }
   }
 
-  /** Write the full state array atomically (tmp + rename). */
+  /** Write the full state array: unique tmp + fsync + rename (with ENOENT retry). */
   private async writeState(records: BackgroundJobRecord[]): Promise<void> {
-    await this.ensureDir()
     const content = JSON.stringify(records, null, 2)
-    await writeFile(this.tmpPath, content, 'utf-8')
-    await rename(this.tmpPath, this.statePath)
+    const tmpPath = this.makeTmpPath()
+    let lastError: unknown
+
+    for (let attempt = 1; attempt <= RENAME_MAX_ATTEMPTS; attempt++) {
+      try {
+        await this.ensureDir()
+        const handle = await this.ops.openForWrite(tmpPath)
+        try {
+          await handle.writeFile(content, 'utf-8')
+          await handle.sync()
+        } finally {
+          await handle.close()
+        }
+        await this.ops.rename(tmpPath, this.statePath)
+        return
+      } catch (err: unknown) {
+        lastError = err
+        if (!isEnoent(err)) break
+        log.warn(
+          `[FilePersistence] rename ENOENT for ${this.statePath} ` +
+            `(attempt ${attempt}/${RENAME_MAX_ATTEMPTS}) — recreating dir and retrying`,
+        )
+        await this.safeUnlink(tmpPath)
+      }
+    }
+
+    await this.safeUnlink(tmpPath)
+    throw lastError instanceof Error ? lastError : new Error(String(lastError))
   }
 
-  async saveJob(record: BackgroundJobRecord): Promise<void> {
-    const records = (await this.readState()) ?? []
-    const idx = records.findIndex((r) => r.taskID === record.taskID)
-    if (idx >= 0) {
-      records[idx] = record
-    } else {
-      records.push(record)
+  private async safeUnlink(path: string): Promise<void> {
+    try {
+      await this.ops.unlink(path)
+    } catch {
+      // Best-effort temp cleanup — absent file is fine.
     }
-    await this.writeState(records)
+  }
+
+  async saveJob(record: BackgroundJobRecord): Promise<void> {
+    await enqueueWrite(this.lockKey, async () => {
+      const records = await this.readState()
+      if (records === null) {
+        log.error(
+          `[FilePersistence] Aborting saveJob(${record.taskID}): could not read ` +
+            `${this.statePath} — refusing to overwrite state with a partial snapshot`,
+        )
+        return
+      }
+      const idx = records.findIndex((r) => r.taskID === record.taskID)
+      if (idx >= 0) {
+        records[idx] = record
+      } else {
+        records.push(record)
+      }
+      await this.writeState(records)
+    })
   }
 
   async loadAllJobs(): Promise<BackgroundJobRecord[]> {
@@ -93,11 +240,13 @@ export class FilePersistenceAdapter implements PersistenceAdapter {
   }
 
   async deleteJob(taskID: string): Promise<void> {
-    const records = await this.readState()
-    if (!records) return
-    const idx = records.findIndex((r) => r.taskID === taskID)
-    if (idx < 0) return
-    records.splice(idx, 1)
-    await this.writeState(records)
+    await enqueueWrite(this.lockKey, async () => {
+      const records = await this.readState()
+      if (!records) return
+      const idx = records.findIndex((r) => r.taskID === taskID)
+      if (idx < 0) return
+      records.splice(idx, 1)
+      await this.writeState(records)
+    })
   }
 }

package/src/pantheon/v2-events.ts

@@ -104,12 +104,20 @@ export function createV2EventDispatcher(deps: V2EventDeps): V2EventDispatcher {
             } else {
               // Native task(background=true) child: register on the board
               // atomically (skip if V1 idle hook or re-dispatch already claimed it).
-              await deps.board.registerLaunchIfAbsent({
-                taskID: info.id,
-                parentSessionID: info.parentID,
-                agent: 'native',
-                description: `native child ${info.id}`,
-              })
+              try {
+                await deps.board.registerLaunchIfAbsent({
+                  taskID: info.id,
+                  parentSessionID: info.parentID,
+                  agent: 'native',
+                  description: `native child ${info.id}`,
+                })
+              } catch (err) {
+                // The per-agent concurrency limit can refuse the launch. Do not
+                // abort the rest of the handler: this child is simply not
+                // mirrored on the board, while finalize/idle dispatch for every
+                // other event must still run.
+                log.warn(`[Pantheon V2] native child ${info.id} not mirrored on board:`, err)
+              }
             }
           }
         }

package/src/plugins/tui/dist/tui.js

@@ -164,7 +164,7 @@ async function detectVersion(api) {
 			if (ver) return ver;
 		}
 	} catch {}
-	return "1.5.0-beta.6";
+	return "1.5.0-beta.8";
 }
 /**
 * usage-bar — AI subscription usage gauge for the opencode TUI.

package/src/plugins/tui/package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "pantheon-tui",
-  "version": "1.5.0-beta.6",
+  "version": "1.5.0-beta.8",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "pantheon-tui",
-      "version": "1.5.0-beta.6",
+      "version": "1.5.0-beta.8",
       "license": "MIT",
       "dependencies": {
         "@opencode-ai/plugin": ">=1.14.0",

package/src/plugins/tui/package.json

@@ -1,7 +1,7 @@
 {
   "name": "pantheon-tui",
   "type": "module",
-  "version": "1.5.0-beta.6",
+  "version": "1.5.0-beta.8",
   "description": "Pantheon TUI plugin — sidebar (version, preset, sessions, real-time delegations) + AI subscription usage gauges incl. OpenCode Go/Zen dollar limits (vendored, MIT)",
   "license": "MIT",
   "exports": {