Release v0.51.269 — Release IK (stage-b2 — sidebar perf + search scope + Windows ctl)

[nesquena-hermes]

Release v0.51.269 — Release IK (stage-b2)

Low-risk Phase-1 batch of 3 isolated contributor PRs (disjoint files: sessions.js / routes.py / ctl.sh). Full suite green, Codex SAFE TO SHIP, Opus ship.

Fixed

• fix: scope session search to active profile #3646 (@Hinotoi-agent) — scope session search to the active profile (mirrors existing list endpoints; ?all_profiles=1 returns the aggregate).
• fix(tests): tree-kill ctl.sh stop on Windows (#3669) #3670 (@rodboev) — Windows tree-kill for ctl.sh stop (taskkill //F //T) + Windows/POSIX path normalization; POSIX behavior unchanged.

Performance

• perf(ui): single-pass sidebar session row partitioning #3658 (@pamnard) — single-pass sidebar session row partitioning (replaces a 5-stage .filter() chain per render); behavior-identical.

Gates

• Full sequential suite: 7804 passed, 9 skipped, 3 xpassed, 0 failed (231s)
• Pre-Opus gate: merge-markers clean, node -c + ast.parse + bash -n clean, ruff CLEAN, ESLint runtime CLEAN, browser smoke CLEAN
• Codex regression gate: SAFE TO SHIP (ran isolated POSIX ctl.sh stop check confirming Linux unchanged)
• Opus advisor: ship (verified perf(ui): single-pass sidebar session row partitioning #3658 partition diff-equivalence)

Closes #3658, #3646, #3670

[greptile-apps]

Greptile Summary

Three isolated contributor PRs merged as a single release batch: session search is now scoped to the active profile (matching existing list endpoints), ctl.sh stop gains Windows tree-kill support via taskkill //F //T, and the sidebar session list is refactored from a 5-stage .filter() chain into a single-pass partition helper.

• fix: scope session search to active profile #3646 (routes.py): _handle_sessions_search filters all_sessions() down to the active profile before any title or content matching; ?all_profiles=1 restores the aggregate. Response now includes all_profiles and active_profile metadata fields on both the empty-query and full-search paths.
• fix(tests): tree-kill ctl.sh stop on Windows (#3669) #3670 (ctl.sh): New _is_windows_bash, _windows_pid_for_bash_pid, and _stop_webui_pid helpers route the stop signal through taskkill //F //T on Windows and fall back to POSIX kill otherwise; _is_owned_webui_pid gains slash-normalised and cygpath-converted path variants to handle mixed Windows/POSIX path forms.
• perf(ui): single-pass sidebar session row partitioning #3658 (sessions.js): _partitionSidebarSessionRows replaces the chained .filter() calls with two sequential passes over allMatched, preserving filter semantics and archivedCount exactly.

Confidence Score: 4/5

Safe to merge; the Python and JavaScript changes are behavior-equivalent and well-covered by new tests. The ctl.sh Windows path is the only area warranting a second look.

The routes.py profile-scoping and sessions.js single-pass refactor are clean, well-tested, and match the existing codebase patterns exactly. The ctl.sh Windows changes introduce more complexity: _stop_webui_pid ignores the signal parameter on Windows (always force-kills, no graceful TERM window), and the wmic-based command-line detection is deprecated on Windows 10 21H1+ and absent on many Windows 11 installs — though the ps -f fallback should cover most cases. Neither issue is a hard breakage, but the Windows stop path relies on several assumptions about Git Bash's ps -l column layout.

ctl.sh — particularly the _stop_webui_pid signal handling and _proc_args wmic/ps-f fallback chain on Windows.

Important Files Changed

Filename	Overview
api/routes.py	Adds profile scoping to _handle_sessions_search by filtering sessions to the active profile before any search, mirroring the existing list-endpoint pattern. Both empty-query and full-search paths are updated; all_profiles=1 opt-in is preserved.
ctl.sh	Adds Windows (Git Bash/MSYS) support for stop: new _is_windows_bash, _windows_bash_path, _windows_pid_for_bash_pid, and _stop_webui_pid helpers; _is_owned_webui_pid extended with slash-normalised and cygpath-converted path variants. POSIX path unchanged. Two P2 concerns: signal parameter ignored on Windows (always force-kills), and wmic deprecated on modern Windows builds.
static/sessions.js	Extracts five sequential .filter() passes into _sidebarRowHasVisibleMessages and _partitionSidebarSessionRows, iterating allMatched exactly twice. All filter conditions, archivedCount semantics, and profileFiltered population are equivalent to the original chain.
tests/test_sessions_search_profile_scope.py	New test file covering empty-query scope, title search isolation, content search isolation, and all_profiles=1 aggregate opt-in. Fixtures patch all_sessions, get_session, get_active_profile_name, and j cleanly.
tests/test_ctl_script.py	Windows compatibility helpers added (bash_path, bash_pid, windows_pid, process_exists, start_fake_launchd_process). Existing tests updated to use path-normalised assertions and _kill_tree cleanup. launchd test skipped on Windows.
tests/test_sidebar_session_partition.py	New regression test validating the single-pass partition helper exists and contains all expected filter gates. Source-text assertions are precise.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[stop_cmd called] --> B{PID file exists & process alive?}
    B -- No --> C[_clear_stale_pid / return]
    B -- Yes --> D[_stop_webui_pid pid TERM]
    D --> E{_is_windows_bash && taskkill available?}
    E -- Yes --> F[_windows_pid_for_bash_pid]
    F --> G{winpid numeric?}
    G -- Yes --> H["taskkill //F //T //PID winpid"]
    H --> I[return]
    G -- No --> J[POSIX kill pid]
    E -- No --> J
    I --> K{_is_alive loop x50 / 5s}
    K -- dead --> L[rm PID+STATE / Stopped]
    K -- still alive --> M[_stop_webui_pid pid KILL]
    M --> E
    M --> N[rm PID+STATE]

Loading

_{Reviews (1): Last reviewed commit: "docs(changelog): v0.51.269 — Release IK ..." | Re-trigger Greptile}

[greptile-apps]

signal parameter silently ignored on Windows — always force-kills

_stop_webui_pid accepts signal (TERM or KILL) but the Windows branch unconditionally invokes taskkill //F //T on both calls. The caller in stop_cmd still prints "Process did not exit after SIGTERM; sending SIGKILL" and then calls this function a second time, but on Windows the first call already force-terminates the tree; the escalation path never has a chance to run and the log message is misleading. If the WebUI needs a graceful-shutdown window (e.g. flushing a write buffer), that window doesn't exist on Windows. Consider omitting the //F flag on the TERM call and reserving the force flag for the KILL call, or documenting that Windows always gets a hard kill.

[greptile-apps]

wmic is deprecated and absent on many modern Windows installs

wmic was deprecated in Windows 10 21H1 (May 2021) and is absent by default in many Windows 11 builds. The command -v wmic >/dev/null 2>&1 guard degrades gracefully to the ps -f fallback, so this isn't a crash — but wmic will silently be missing on an increasing share of Windows targets. The ps -p "${pid}" -f path (columns starting at field 8) is the de-facto fallback; Get-WmiObject/Get-CimInstance via powershell.exe would be a forward-compatible alternative if the ps -f fallback turns out to be unreliable for detecting the process in practice.

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!