rebind the Claude pane on an in-place workspace switch (T-269)

Separate clide windows are isolated (own process, per-root IPC socket, per-repo deterministic session id), so parallel repos in separate windows were already fine. But switching the workspace in place (Open Project / Open Folder) only emitted ProjectOpened — nothing rebound the Claude session, so the primary pane kept the PREVIOUS repo's conversation. Two compounding causes, fixed in layers: - ClaudeSessionOrchestrator.spawn() was idempotent on the literal key 'primary' without checking cwd, so it handed the old repo's session to the new repo. It now reuses a cached session only when its cwd matches the spec; a mismatch tears the stale one down and spawns fresh. - The primary ClaudePane is built once behind a GlobalKey and spawns once, so it never re-resolved. It now listens for ProjectOpened and rebinds: close its orchestrator entry, drop the cached session id + repo root, and respawn against the now-active repo. Secondaries don't self-rebind. - ClaudeSessionHost drops the old repo's secondary/fork tabs on a switch, so a switched workspace starts like a fresh launch (lone primary). - The extension closes any remaining sessions whose cwd != the new root, catching team/non-pane sessions no pane owns. Tested at the orchestrator: cwd-aware idempotency (reuse on cwd match, teardown + respawn on mismatch). Pane/host widget coverage is intentionally deferred — claude_pane.dart has no widget-test harness yet and pulling it into coverage piecemeal would drop the gate; tracked separately. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-07 12:37:53 +02:00
parent 892240b3a6
commit f813fd4d5a
8 changed files with 176 additions and 5 deletions
@@ -764,3 +764,5 @@ Refs: D-78 (interaction zone / display-only conversation widgets). Built on T-16
 SEQUENCING UPDATE (after T-266 refinement): the shared container/holder primitive now lives in T-266 and is CONSUMED by T-264 (nested agent run), so T-266 is no longer "last" — its primitive lands before/with T-264. T-264 is now blocked by T-266. Revised order: T-262 / T-263 (fold success result, fold agent prompt) → T-266 (shared holder/container primitive + activity-card restyle) → T-264 (nest the whole agent run on that primitive) → T-265 (relabel sidechain prose) can land anytime alongside.', NULL, '2026-06-07 08:49:16', '2026-06-07 08:49:16', '2026-06-07 08:49:16', NULL, '7795b980e57c7bf096b495b3f61c8289', 1) ON CONFLICT(hash) DO NOTHING;
 INSERT INTO ticket_history (ticket_id, field, old_value, new_value, changed_by, changed_at, created_at, updated_at, deleted_at, hash, canonical_version) VALUES ('T-255', 'status', 'in_progress', 'done', NULL, '2026-06-07 08:51:08', '2026-06-07 08:51:08', '2026-06-07 08:51:08', NULL, '68cf2ebb539ca57c1c8165acecf3c111', 1) ON CONFLICT(hash) DO NOTHING;
 INSERT INTO ticket_history (ticket_id, field, old_value, new_value, changed_by, changed_at, created_at, updated_at, deleted_at, hash, canonical_version) VALUES ('T-268', 'status', 'backlog', 'done', NULL, '2026-06-07 09:36:59', '2026-06-07 09:36:59', '2026-06-07 09:36:59', NULL, 'dcce34ebf831a8836fe31bcad6bab92c', 1) ON CONFLICT(hash) DO NOTHING;
+INSERT INTO ticket_history (ticket_id, field, old_value, new_value, changed_by, changed_at, created_at, updated_at, deleted_at, hash, canonical_version) VALUES ('T-269', 'status', 'backlog', 'done', NULL, '2026-06-07 09:51:32', '2026-06-07 09:51:32', '2026-06-07 09:51:32', NULL, '3fec60e8de272979ad791c313eac76f8', 1) ON CONFLICT(hash) DO NOTHING;
+INSERT INTO ticket_history (ticket_id, field, old_value, new_value, changed_by, changed_at, created_at, updated_at, deleted_at, hash, canonical_version) VALUES ('T-48', 'status', 'backlog', 'ready', NULL, '2026-06-07 10:03:36', '2026-06-07 10:03:36', '2026-06-07 10:03:36', NULL, '643716825d71326f6066fb04cbb67077', 1) ON CONFLICT(hash) DO NOTHING;
@@ -1597,3 +1597,27 @@ Fix: /clear in the PRIMARY pane should empty the deterministic session in place
 INSERT INTO tickets (id, type, parent_id, title, description, status, priority, assigned_to, team, decision_ref, created_at, updated_at, deleted_at, hash, canonical_version) VALUES ('T-268', 'bug', NULL, 'primary /clear must clear the deterministic session in place, not orphan to a fresh id', 'The primary Claude pane is anchored to a deterministic session id derived from the repo path (primarySessionId), so it resumes the same transcript across restarts (D-77/T-146). But clide-owned /clear (T-156) respawns on a fresh RANDOM id (freshSessionId). Next clide launch recomputes the deterministic id, finds its old transcript on disk, and resumes the PRE-clear conversation — the cleared session is orphaned. Continuity appears broken.

 Fix: /clear in the PRIMARY pane should empty the deterministic session in place — delete its <id>.jsonl (and sidecar <id>/ dir) so ''claude --session-id'' re-creates it empty, then respawn on the SAME deterministic id. Secondary panes are throwaway and keep the fresh-random behavior. Factor the transcript path/dir construction (duplicated in claude_pane._spawn and _resumeFlow) into session_naming helpers (claudeProjectDir/claudeTranscriptPath/clearSessionTranscript) so the clear logic is DRY and unit-testable.', 'done', 'high', NULL, NULL, 'D-77', '2026-06-07 09:35:03', '2026-06-07 09:36:59', NULL, '824de390dc6da465575d998291976377', 1) ON CONFLICT(id) DO UPDATE SET type=excluded.type, parent_id=excluded.parent_id, title=excluded.title, description=excluded.description, status=excluded.status, priority=excluded.priority, assigned_to=excluded.assigned_to, team=excluded.team, decision_ref=excluded.decision_ref, updated_at=excluded.updated_at, deleted_at=excluded.deleted_at, hash=excluded.hash, canonical_version=excluded.canonical_version WHERE excluded.updated_at > tickets.updated_at OR (excluded.updated_at = tickets.updated_at AND excluded.hash > tickets.hash);
+INSERT INTO tickets (id, type, parent_id, title, description, status, priority, assigned_to, team, decision_ref, created_at, updated_at, deleted_at, hash, canonical_version) VALUES ('T-269', 'bug', NULL, 'primary Claude pane keeps the previous repo''s session after an in-place workspace switch', 'Separate clide WINDOWS are isolated (separate process, per-root IPC socket, per-repo deterministic session id), so parallel repos in separate windows are fine. The bug is the in-place workspace switch (Open Project / Open Folder -> kernel.project.open, app.dart:456/474): it only emits ProjectOpened and updates settings (project.dart:147) — nothing tears down Claude sessions.
+
+Two compounding causes:
+1. ClaudeSessionOrchestrator.spawn() is idempotent on the literal key ''primary'' WITHOUT checking cwd (session_orchestrator.dart:192-194) — it returns the old repo''s session for the new repo.
+2. ClaudePane spawns once behind a _spawned guard, under a GlobalKey host (claude_pane.dart:140, extension.dart:33/52) — so after an in-place switch it never re-resolves; it keeps the previous repo''s session/conversation.
+
+Fix (T-269):
+- Orchestrator safety net: spawn() must reuse an existing session only when its cwd matches spec.cwd; a cwd mismatch is a stale cross-workspace binding -> close + spawn fresh.
+- ClaudePane (primary) listens for ProjectOpened and rebinds: close its orch entry, null out _sessionId/_repoRoot, respawn for the new repo. Secondaries do not self-rebind.
+- ClaudeSessionHost listens for ProjectOpened and drops secondary/fork tabs (the old repo''s), keeping just the primary — a switched workspace starts like a fresh launch.
+- Extension tears down all sessions whose cwd != new root on ProjectOpened (catches team/non-pane sessions).', 'backlog', 'high', NULL, NULL, 'D-77', '2026-06-07 09:44:17', '2026-06-07 09:44:17', NULL, 'e2da4b11d8f64cfeddeab64f6407004a', 1) ON CONFLICT(id) DO UPDATE SET type=excluded.type, parent_id=excluded.parent_id, title=excluded.title, description=excluded.description, status=excluded.status, priority=excluded.priority, assigned_to=excluded.assigned_to, team=excluded.team, decision_ref=excluded.decision_ref, updated_at=excluded.updated_at, deleted_at=excluded.deleted_at, hash=excluded.hash, canonical_version=excluded.canonical_version WHERE excluded.updated_at > tickets.updated_at OR (excluded.updated_at = tickets.updated_at AND excluded.hash > tickets.hash);
+INSERT INTO tickets (id, type, parent_id, title, description, status, priority, assigned_to, team, decision_ref, created_at, updated_at, deleted_at, hash, canonical_version) VALUES ('T-269', 'bug', NULL, 'primary Claude pane keeps the previous repo''s session after an in-place workspace switch', 'Separate clide WINDOWS are isolated (separate process, per-root IPC socket, per-repo deterministic session id), so parallel repos in separate windows are fine. The bug is the in-place workspace switch (Open Project / Open Folder -> kernel.project.open, app.dart:456/474): it only emits ProjectOpened and updates settings (project.dart:147) — nothing tears down Claude sessions.
+
+Two compounding causes:
+1. ClaudeSessionOrchestrator.spawn() is idempotent on the literal key ''primary'' WITHOUT checking cwd (session_orchestrator.dart:192-194) — it returns the old repo''s session for the new repo.
+2. ClaudePane spawns once behind a _spawned guard, under a GlobalKey host (claude_pane.dart:140, extension.dart:33/52) — so after an in-place switch it never re-resolves; it keeps the previous repo''s session/conversation.
+
+Fix (T-269):
+- Orchestrator safety net: spawn() must reuse an existing session only when its cwd matches spec.cwd; a cwd mismatch is a stale cross-workspace binding -> close + spawn fresh.
+- ClaudePane (primary) listens for ProjectOpened and rebinds: close its orch entry, null out _sessionId/_repoRoot, respawn for the new repo. Secondaries do not self-rebind.
+- ClaudeSessionHost listens for ProjectOpened and drops secondary/fork tabs (the old repo''s), keeping just the primary — a switched workspace starts like a fresh launch.
+- Extension tears down all sessions whose cwd != new root on ProjectOpened (catches team/non-pane sessions).', 'done', 'high', NULL, NULL, 'D-77', '2026-06-07 09:44:17', '2026-06-07 09:51:32', NULL, '9b92ed68f5c352d5461c01d7e98c0b50', 1) ON CONFLICT(id) DO UPDATE SET type=excluded.type, parent_id=excluded.parent_id, title=excluded.title, description=excluded.description, status=excluded.status, priority=excluded.priority, assigned_to=excluded.assigned_to, team=excluded.team, decision_ref=excluded.decision_ref, updated_at=excluded.updated_at, deleted_at=excluded.deleted_at, hash=excluded.hash, canonical_version=excluded.canonical_version WHERE excluded.updated_at > tickets.updated_at OR (excluded.updated_at = tickets.updated_at AND excluded.hash > tickets.hash);
+INSERT INTO tickets (id, type, parent_id, title, description, status, priority, assigned_to, team, decision_ref, created_at, updated_at, deleted_at, hash, canonical_version) VALUES ('T-48', 'story', NULL, 'application menu bar (Zed-style)', 'Add a Zed-style application menu integrated into the hat bar: File, Edit, Selection, View, Help/About. Rendered as custom Flutter widgets (no native menu bar — we own the chrome per D-7). Menus expose the same commands registered in the command palette with their keybindings shown inline. Submenus for View should include panel toggles, zoom, focus mode. About opens a modal with version, license, and links.', 'ready', 'medium', NULL, NULL, NULL, '2026-04-23 20:30:09', '2026-06-07 10:03:36', NULL, 'e09873c62cfab523b2b4048d9681c9ae', 1) ON CONFLICT(id) DO UPDATE SET type=excluded.type, parent_id=excluded.parent_id, title=excluded.title, description=excluded.description, status=excluded.status, priority=excluded.priority, assigned_to=excluded.assigned_to, team=excluded.team, decision_ref=excluded.decision_ref, updated_at=excluded.updated_at, deleted_at=excluded.deleted_at, hash=excluded.hash, canonical_version=excluded.canonical_version WHERE excluded.updated_at > tickets.updated_at OR (excluded.updated_at = tickets.updated_at AND excluded.hash > tickets.hash);
+INSERT INTO tickets (id, type, parent_id, title, description, status, priority, assigned_to, team, decision_ref, created_at, updated_at, deleted_at, hash, canonical_version) VALUES ('T-270', 'story', NULL, 'make sidebar filter boxes addressable via MessageBus (clide ui filter)', NULL, 'backlog', 'medium', NULL, NULL, 'D-6', '2026-06-07 10:23:13', '2026-06-07 10:23:13', NULL, '0d79acd088da9d0b240ee5473c3a21c4', 1) ON CONFLICT(id) DO UPDATE SET type=excluded.type, parent_id=excluded.parent_id, title=excluded.title, description=excluded.description, status=excluded.status, priority=excluded.priority, assigned_to=excluded.assigned_to, team=excluded.team, decision_ref=excluded.decision_ref, updated_at=excluded.updated_at, deleted_at=excluded.deleted_at, hash=excluded.hash, canonical_version=excluded.canonical_version WHERE excluded.updated_at > tickets.updated_at OR (excluded.updated_at = tickets.updated_at AND excluded.hash > tickets.hash);
@@ -154,6 +154,11 @@ heading, and (b) bumping `pubspec.yaml` `version:` in the same commit.

 ### Fixed

+- Switching the workspace in place (Open Project/Folder) now rebinds the Claude
+  pane to the new repo's session instead of keeping the previous repo's
+  conversation, and drops the old repo's secondary tabs. Separate windows were
+  already isolated — this only affected reusing one window for another repo.
+  (T-269)
 - `/clear` in the primary Claude pane now clears that session **in place** —
  it empties the pane's deterministic, restart-stable session instead of
  starting a throwaway random one. Previously a cleared primary was orphaned:
@@ -67,6 +67,7 @@ class ClaudePane extends StatefulWidget {

 class _ClaudePaneState extends State<ClaudePane> {
  StreamSubscription<SessionStatus>? _statusSub;
+  StreamSubscription<ProjectOpened>? _projectSub;
  ConversationController? _conversation;
  StreamJsonSession? _session;
  SessionStatus _status = const SessionStatus();
@@ -145,12 +146,19 @@ class _ClaudePaneState extends State<ClaudePane> {
      unawaited(activeClaudeConfig?.ensureProbe());
      // Reflect skills/config changes in the status line (T-154).
      activeClaudeConfig?.addListener(_onConfigChanged);
+      // Rebind to the new repo's session when the workspace is switched in
+      // place (Open Project/Folder). This pane is built once behind a
+      // GlobalKey and spawns once, so without this it would keep the previous
+      // repo's session after a switch (T-269).
+      _projectSub = ClideKernel.of(context).events.on<ProjectOpened>().listen(_onProjectChanged);
    }
  }

  @override
  void dispose() {
    activeClaudeConfig?.removeListener(_onConfigChanged);
+    _projectSub?.cancel();
+    _projectSub = null;
    _statusSub?.cancel();
    _statusSub = null;
    // The orchestrator owns the session, so disposing this pane does NOT kill
@@ -185,6 +193,37 @@ class _ClaudePaneState extends State<ClaudePane> {
    return _spawn();
  }

+  /// Rebind to the active workspace when the project is switched in place
+  /// (T-269). Only the primary rebinds — secondaries/forks belong to the old
+  /// repo and are dropped by the host. A no-op when the path is unchanged or
+  /// the session hasn't resolved its repo yet.
+  void _onProjectChanged(ProjectOpened e) {
+    if (!mounted || !widget.isPrimary) return;
+    if (_repoRoot == null || e.path == _repoRoot) return;
+    unawaited(_rebindToActiveProject());
+  }
+
+  /// Tear down the current session and respawn against the now-active
+  /// workspace: drop the cached session id and repo root so [_spawn]
+  /// re-resolves both for the new repo (T-269).
+  Future<void> _rebindToActiveProject() async {
+    _statusSub?.cancel();
+    _statusSub = null;
+    await activeSessionOrchestrator?.close(_orchId); // kills the old repo's session
+    _conversation = null;
+    _session = null;
+    _sessionId = null;
+    _repoRoot = null;
+    if (mounted) {
+      setState(() {
+        _status = const SessionStatus();
+        _error = null;
+        _statusLine = 'starting…';
+      });
+    }
+    await _spawn();
+  }
+
  Future<void> _spawn() async {
    if (!mounted) return;
    final ipc = _ipc();
@@ -1,3 +1,6 @@
+import 'dart:async';
+
+import 'package:clide/kernel/kernel.dart';
 import 'package:clide/widgets/widgets.dart';
 import 'package:flutter/widgets.dart';

@@ -21,6 +24,9 @@ class ClaudeSessionHostState extends State<ClaudeSessionHost> {
  late final MultitabController<_Session> _controller;
  int _nextSecondary = 1;

+  StreamSubscription<ProjectOpened>? _projectSub;
+  String? _projectRoot;
+
  @override
  void initState() {
    super.initState();
@@ -39,8 +45,34 @@ class ClaudeSessionHostState extends State<ClaudeSessionHost> {
    );
  }

+  @override
+  void didChangeDependencies() {
+    super.didChangeDependencies();
+    _projectSub ??= ClideKernel.of(context).events.on<ProjectOpened>().listen(_onProjectChanged);
+  }
+
+  /// Reset to a lone primary tab when the workspace is switched in place
+  /// (T-269): the old repo's secondaries/forks don't belong in the new
+  /// workspace. Removing them disposes their panes, which close their sessions
+  /// through the orchestrator. The primary tab stays and rebinds itself.
+  void _onProjectChanged(ProjectOpened e) {
+    final prev = _projectRoot;
+    _projectRoot = e.path;
+    if (prev == null || prev == e.path) return; // initial open / no change
+    if (!mounted) return;
+    final stale = _controller.entries.where((x) => x.id != _primaryId).map((x) => x.id).toList();
+    if (stale.isEmpty) return;
+    setState(() {
+      for (final id in stale) {
+        _controller.remove(id);
+      }
+      _nextSecondary = 1;
+    });
+  }
+
  @override
  void dispose() {
+    _projectSub?.cancel();
    _controller.dispose();
    super.dispose();
  }
@@ -36,6 +36,10 @@ class ClaudeExtension extends ClideExtension {
  ClaudeSessionOrchestrator? _orchestrator;
  final List<StreamSubscription<dynamic>> _subs = [];

+  /// Active workspace root, tracked so an in-place project switch can tear
+  /// down the previous repo's sessions (T-269).
+  String? _projectRoot;
+
  /// App-wide Claude environment (skills, commands, settings, permissions,
  /// slash list). Built and loaded at activation (D-76, T-151).
  ClaudeConfig? get config => _config;
@@ -324,12 +328,34 @@ class ClaudeExtension extends ClideExtension {
    _orchestrator = ClaudeSessionOrchestrator();
    activeSessionOrchestrator = _orchestrator;

+    // An in-place workspace switch (Open Project/Folder) must not leave the
+    // previous repo's sessions running — including team/non-pane sessions the
+    // panes don't own. Close every session that doesn't belong to the new
+    // root; panes rebind themselves to the new repo (T-269).
+    _subs.add(ctx.events.on<ProjectOpened>().listen(_onProjectChanged));
+
    // `clide image show <path>` (T-249): the dispatcher resolves + publishes an
    // 'image' message; we inject the matching card into the conversation the
    // user is looking at (the primary lead, else the first visible session).
    _subs.add(ctx.messages.subscribe(channel: imageShowChannel).listen(_onImageShow));
  }

+  /// Close every session that doesn't belong to the newly-active workspace
+  /// after an in-place project switch (T-269). The initial open (no previous
+  /// root) and a no-op re-open are skipped. Panes rebind to the new repo on
+  /// their own; this catches team/orphan sessions no pane owns.
+  void _onProjectChanged(ProjectOpened e) {
+    final prev = _projectRoot;
+    _projectRoot = e.path;
+    if (prev == null || prev == e.path) return;
+    final orch = _orchestrator;
+    if (orch == null) return;
+    final stale = orch.sessions.where((m) => m.cwd != e.path).map((m) => m.id).toList();
+    for (final id in stale) {
+      unawaited(orch.close(id));
+    }
+  }
+
  /// Inject an [ImageMessage] from a published `image` bus message (T-249).
  /// Dropped silently if no live conversation is available — the CLI already
  /// reported success at publish time, and a missing pane is transient.
@@ -187,11 +187,20 @@ class ClaudeSessionOrchestrator extends ChangeNotifier {

  ManagedSession? byId(String id) => _sessions[id];

-  /// Spawn and register a session. Idempotent on [SpawnSpec.id] — a repeat
-  /// call returns the existing session rather than starting a second process.
+  /// Spawn and register a session. Idempotent on [SpawnSpec.id] *within a
+  /// workspace* — a repeat call for the same [SpawnSpec.cwd] returns the
+  /// existing session rather than starting a second process (the fast path that
+  /// lets a hidden/kept-alive pane keep its session). A repeat call with the
+  /// SAME id but a DIFFERENT cwd means the workspace was switched in place
+  /// (T-269): the existing session belongs to the old repo, so it is torn down
+  /// and a fresh one spawned for the new repo — a pane must never inherit
+  /// another workspace's conversation.
  Future<ManagedSession> spawn(SpawnSpec spec) async {
    final existing = _sessions[spec.id];
-    if (existing != null) return existing;
+    if (existing != null) {
+      if (existing.cwd == spec.cwd) return existing;
+      await close(spec.id);
+    }

    // Team sessions host the clide-team MCP server and get a roster + role
    // injected into their system prompt (T-170). Register the member before
@@ -47,11 +47,11 @@ ClaudeSessionOrchestrator _orch(List<_FakeProc> created) {
  );
 }

-SpawnSpec _spec(String id, {bool resume = false, String? transcriptPath}) => SpawnSpec(
+SpawnSpec _spec(String id, {bool resume = false, String? transcriptPath, String cwd = '/repo'}) => SpawnSpec(
      id: id,
      role: id,
      sessionId: '$id-uuid',
-      cwd: '/repo',
+      cwd: cwd,
      resume: resume,
      transcriptPath: transcriptPath,
    );
@@ -140,6 +140,40 @@ void main() {
    });
  });

+  // ---- in-place workspace switch (T-269) ---------------------------------
+
+  group('spawn — cwd-aware idempotency (T-269)', () {
+    late List<_FakeProc> created;
+    late ClaudeSessionOrchestrator orch;
+
+    setUp(() {
+      created = [];
+      orch = _orch(created);
+    });
+
+    tearDown(() => orch.dispose());
+
+    test('same id + same cwd reuses the session (no second process)', () async {
+      final a = await orch.spawn(_spec('primary', cwd: '/repo-a'));
+      final b = await orch.spawn(_spec('primary', cwd: '/repo-a'));
+      expect(identical(a, b), isTrue);
+      expect(created, hasLength(1));
+    });
+
+    test('same id + different cwd tears down the old session and spawns fresh', () async {
+      // Workspace switched in place: the new repo must NOT inherit the old
+      // repo's 'primary' session.
+      final a = await orch.spawn(_spec('primary', cwd: '/repo-a'));
+      final b = await orch.spawn(_spec('primary', cwd: '/repo-b'));
+      expect(identical(a, b), isFalse);
+      expect(b.cwd, '/repo-b');
+      expect(created, hasLength(2));
+      expect(created.first.killed, isTrue, reason: 'old repo session is killed');
+      expect(orch.sessions, hasLength(1));
+      expect(orch.byId('primary')!.cwd, '/repo-b');
+    });
+  });
+
  // ---- /resume — bind to an existing session via orchestrator -------------

  group('/resume — resume an existing session via orchestrator', () {