@adhdev/daemon-core 0.9.82-rc.301 → 0.9.82-rc.303

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@adhdev/daemon-core",
-  "version": "0.9.82-rc.301",
+  "version": "0.9.82-rc.303",
   "description": "ADHDev daemon core — CDP, IDE detection, providers, command execution",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",
@@ -46,7 +46,7 @@
   "author": "vilmire",
   "license": "AGPL-3.0-or-later",
   "dependencies": {
-    "@adhdev/mesh-shared": "0.9.82-rc.301",
+    "@adhdev/mesh-shared": "0.9.82-rc.303",
     "@adhdev/session-host-core": "*",
     "@agentclientprotocol/sdk": "^0.16.1",
     "ajv": "^8.20.0",

package/src/cli-adapters/pty-transport.ts

@@ -1,5 +1,6 @@
 import * as os from 'os';
 import { ensureNodePtySpawnHelperPermissions } from './spawn-env.js';
+import { resolveWin32Executable } from './resolve-executable.js';
 
 let cachedPty: any | null | undefined;
 
@@ -119,7 +120,7 @@ export class NodePtyTransportFactory implements PtyTransportFactory {
         cwd = os.homedir();
       }
     }
-    const handle = pty.spawn(command, args, {
+    const handle = pty.spawn(resolveWin32Executable(command), args, {
       name: 'xterm-256color',
       cols: options.cols,
       rows: options.rows,

package/src/cli-adapters/resolve-executable.ts

@@ -0,0 +1,45 @@
+import { execFileSync } from 'child_process';
+import { existsSync } from 'fs';
+import * as path from 'path';
+
+// Extensions ConPTY/CreateProcess can launch directly (not .cmd/.bat, which
+// need a cmd.exe wrapper).
+const DIRECT_EXEC_EXT = new Set(['.exe', '.com']);
+
+/**
+ * Resolve a launch command to an absolute executable path on Windows.
+ *
+ * node-pty's ConPTY backend resolves a bare/relative command against the
+ * *calling process's* `Path` env var and — critically — does NOT apply PATHEXT.
+ * So a provider command like `claude` never matches `claude.exe` and the native
+ * layer throws `File not found:` (empty), which crashes the daemon. We resolve
+ * it to an absolute `.exe` here (in the daemon process, which has the full PATH)
+ * before the command ever reaches node-pty.
+ *
+ * No-op on non-Windows and when the command is already an existing absolute path
+ * or cannot be resolved (caller keeps the original behaviour).
+ */
+export function resolveWin32Executable(command: string): string {
+  if (process.platform !== 'win32') return command;
+  const trimmed = (command || '').trim();
+  if (!trimmed) return command;
+
+  // Already an absolute path that exists — keep it.
+  if (path.isAbsolute(trimmed) && existsSync(trimmed)) return trimmed;
+
+  try {
+    const out = execFileSync('where', [trimmed], {
+      encoding: 'utf8',
+      windowsHide: true,
+    }).trim();
+    if (out) {
+      const matches = out.split(/\r?\n/).map((s) => s.trim()).filter(Boolean);
+      // Prefer a directly-launchable executable (.exe/.com) over .cmd/.bat shims.
+      const direct = matches.find((m) => DIRECT_EXEC_EXT.has(path.extname(m).toLowerCase()));
+      return direct || matches[0] || command;
+    }
+  } catch {
+    // `where` not found / non-zero exit — fall through to original command.
+  }
+  return command;
+}

package/src/cli-adapters/session-host-transport.ts

@@ -6,6 +6,7 @@ import {
     type SessionHostRecord,
 } from '@adhdev/session-host-core';
 import { LOG } from '../logging/logger.js';
+import { resolveWin32Executable } from './resolve-executable.js';
 import type { PtyRuntimeMetadata, PtyRuntimeTransport, PtySpawnOptions, PtyTransportFactory } from './pty-transport.js';
 
 interface SessionHostPtyTransportFactoryOptions {
@@ -435,7 +436,7 @@ export class SessionHostPtyTransportFactory implements PtyTransportFactory {
     spawn(command: string, args: string[], spawnOptions: PtySpawnOptions): PtyRuntimeTransport {
         return new SessionHostRuntimeTransport({
             ...this.options,
-            command,
+            command: resolveWin32Executable(command),
             args,
             spawnOptions,
         });

package/src/commands/upgrade-helper.ts

@@ -169,6 +169,34 @@ export function buildPinnedGlobalInstallCommand(options: {
   };
 }
 
+/**
+ * Build an env for the `npm install` child whose PATH is prefixed with the
+ * directory of the node binary currently running this helper.
+ *
+ * npm runs lifecycle scripts (e.g. adhdev's `preinstall` Node-version guard) by
+ * spawning a bare `node`, which resolves from PATH — NOT from the node that runs
+ * npm. On Windows a machine can have several node installs (e.g. a standalone
+ * `C:\Program Files\nodejs` ahead of an nvm-managed node on PATH). Without this,
+ * the guard sees the wrong (unsupported) node version and aborts the upgrade,
+ * even though npm/adhdev actually run under a supported node. Pinning the
+ * running node's dir to the front of PATH makes lifecycle scripts use the same
+ * node as the install itself.
+ */
+function buildInstallEnvWithNodeOnPath(baseEnv: NodeJS.ProcessEnv = process.env): NodeJS.ProcessEnv {
+  // The Node-version guard this works around only fires on Windows, so scope the
+  // PATH rewrite to win32 — POSIX keeps its env untouched.
+  if (process.platform !== 'win32') return { ...baseEnv };
+  const nodeBinDir = path.dirname(process.execPath);
+  if (!nodeBinDir) return { ...baseEnv };
+  const env: NodeJS.ProcessEnv = { ...baseEnv };
+  // Windows env keys are case-insensitive and conventionally spelled `Path`;
+  // prepend to the existing key (whatever its case) to avoid creating a dupe.
+  const pathKey = Object.keys(env).find((k) => k.toLowerCase() === 'path') || 'PATH';
+  const current = env[pathKey] || '';
+  env[pathKey] = current ? `${nodeBinDir};${current}` : nodeBinDir;
+  return env;
+}
+
 export function getNpmExecOptions(platform: NodeJS.Platform = process.platform): NpmExecOptions {
   if (platform === 'win32') {
     return { shell: false, windowsHide: true };
@@ -375,6 +403,7 @@ async function runDaemonUpgradeHelper(payload: DaemonUpgradeHelperPayload): Prom
       encoding: 'utf8',
       stdio: 'pipe',
       maxBuffer: 20 * 1024 * 1024,
+      env: buildInstallEnvWithNodeOnPath(),
       ...installCommand.execOptions,
     },
   );

package/src/git/git-executor.ts

@@ -10,6 +10,18 @@ const execFileAsync = promisify(execFile);
 const DEFAULT_TIMEOUT_MS = 5_000;
 const DEFAULT_MAX_BUFFER = 1024 * 1024;
 
+/**
+ * Timeout for status-collection git commands (status/log/submodule/stash/fetch).
+ * The default 5s is fine for porcelain status, but on Windows the `git` subprocess
+ * spawn itself is pathologically slow (measured: `submodule status` ~3.5s, cold
+ * `log -1` ~4.2s) and overlaps with refreshUpstream fetches — a single command
+ * routinely exceeds 5s, which previously collapsed the whole status to all-null and
+ * dropped the node from the mesh graph. Give the collection path a much larger
+ * budget so a slow-but-healthy repo never reads as "not a git repo". Windows gets a
+ * larger budget than POSIX because the spawn cost is OS-specific, not repo-specific.
+ */
+export const GIT_STATUS_TIMEOUT_MS = process.platform === 'win32' ? 30_000 : 20_000;
+
 export interface GitExecutorOptions {
   timeoutMs?: number;
   maxBuffer?: number;

package/src/git/git-status.ts

@@ -1,9 +1,31 @@
 import type { DaemonBuildBehind, GitRepoStatus, GitSubmoduleStatus, GitUpstreamFreshness } from './git-types.js';
-import { GitCommandError, resolveGitRepository, runGit } from './git-executor.js';
+import { GIT_STATUS_TIMEOUT_MS, GitCommandError, resolveGitRepository, runGit } from './git-executor.js';
 import { getDaemonBuildInfo, type DaemonBuildInfo } from '../build-info.js';
 
 type ResolvedGitRepo = { workspace: string; repoRoot: string | null; isGitRepo: boolean };
 
+/**
+ * Last successfully-collected status per workspace, used to survive a transient git
+ * failure (timeout, slow Windows spawn under load, a momentary lock) WITHOUT dropping
+ * the node out of the mesh graph. A genuine "not a git repository" answer is NOT a
+ * transient failure — it never populates this cache and always reports isGitRepo:false.
+ */
+const lastKnownGoodStatus = new Map<string, GitRepoStatus>();
+
+/** Test seam: clear the last-known-good status cache between cases. */
+export function __resetGitStatusCacheForTests(): void {
+  lastKnownGoodStatus.clear();
+}
+
+/**
+ * git failure reasons that are transient/environmental rather than a real statement
+ * that the workspace is not a repo. On these we prefer the last-known-good status so a
+ * single slow git call cannot make a healthy node vanish from the graph.
+ */
+function isTransientGitFailure(error: GitCommandError): boolean {
+  return error.reason === 'timeout' || error.reason === 'git_command_failed';
+}
+
 export interface GitStatusOptions {
   timeoutMs?: number;
   /** When true, include submodule status in the result. Defaults to true. */
@@ -35,9 +57,50 @@ export async function getGitRepoStatus(
 ): Promise<GitRepoStatus> {
   const lastCheckedAt = Date.now();
   const includeSubmodules = options.includeSubmodules !== false;
+  // Status collection fans out into several git subprocesses (status, head, stash,
+  // submodule, optionally fetch). On Windows the per-spawn cost alone can exceed the
+  // 5s default, so unless the caller pinned a timeout, give the whole collection path
+  // the larger status budget. A caller that explicitly sets timeoutMs (e.g. a test
+  // injecting 1ms to exercise the transient-failure path) is respected.
+  const effectiveOptions: GitStatusOptions =
+    options.timeoutMs === undefined ? { ...options, timeoutMs: GIT_STATUS_TIMEOUT_MS } : options;
 
   try {
-    const repo = await resolveGitRepository(workspace, options);
+    const repo = await resolveGitRepository(workspace, effectiveOptions);
+    const status = await collectGitRepoStatus(repo, includeSubmodules, lastCheckedAt, effectiveOptions);
+    lastKnownGoodStatus.set(workspace, status);
+    return status;
+  } catch (error) {
+    const gitError = error instanceof GitCommandError
+      ? error
+      : new GitCommandError('git_command_failed', 'Failed to read Git status', { cause: error });
+
+    // A transient/environmental failure (timeout, slow-spawn-under-load) must NOT make
+    // a healthy node lose its repo identity and drop out of the mesh graph. Prefer the
+    // last status we successfully collected for this workspace, re-stamped as stale.
+    if (isTransientGitFailure(gitError)) {
+      const cached = lastKnownGoodStatus.get(workspace);
+      if (cached) {
+        return {
+          ...cached,
+          lastCheckedAt,
+          upstreamStatus: 'unavailable',
+          error: gitError.stderr || gitError.message,
+          reason: gitError.reason,
+        };
+      }
+    }
+
+    return emptyStatus(workspace, lastCheckedAt, gitError);
+  }
+}
+
+async function collectGitRepoStatus(
+  repo: ResolvedGitRepo,
+  includeSubmodules: boolean,
+  lastCheckedAt: number,
+  options: GitStatusOptions,
+): Promise<GitRepoStatus> {
     let parsed = await readPorcelainStatus(repo, options);
     let upstreamProbe: GitUpstreamProbe = getInitialUpstreamProbe(parsed);
 
@@ -89,16 +152,6 @@ export async function getGitRepoStatus(
       submodules,
       ...(daemonBuildBehind ? { daemonBuildBehind } : {}),
     };
-  } catch (error) {
-    if (error instanceof GitCommandError) {
-      return emptyStatus(workspace, lastCheckedAt, error);
-    }
-    return emptyStatus(
-      workspace,
-      lastCheckedAt,
-      new GitCommandError('git_command_failed', 'Failed to read Git status', { cause: error }),
-    );
-  }
 }
 
 /**
@@ -507,7 +560,12 @@ async function getSubmoduleStatuses(
   if (!repo.repoRoot) return [];
 
   try {
-    const result = await runGit(repo, ['submodule', 'status', '--recursive'], options);
+    // No `--recursive`: this superproject's submodules (oss, adhdev-providers) are
+    // leaf repos with no nested submodules, so `--recursive` doubles the (already
+    // slow on Windows) submodule-status spawn time for zero additional rows. If a
+    // nested submodule is ever introduced, restore --recursive WITH its own longer
+    // per-command timeout rather than reverting this wholesale.
+    const result = await runGit(repo, ['submodule', 'status'], options);
     const submodules = parseSubmoduleStatusOutput(result.stdout, repo.repoRoot, options.submoduleIgnorePaths);
     await Promise.all(submodules.map(submodule => enrichSubmoduleWorktreeStatus(repo, submodule, options)));
     return submodules;