3 Commits
Author SHA1 Message Date
e7ac9f09fd Release: mobile drawer + terminal-backend cache fixes (#6921 #6933 #6671 #6175) (#6990)
* fix: include terminal backend identity in agent cache signature (#5937) (#6175)

* fix(mobile): preserve extension panels in drawer sync

* fix: close mobile sidebar drawer when opening main-view create forms

The add-space/new-profile/new-skill/new-cron forms render in the main
view, which on phones is covered by the full-screen sidebar drawer
(z-index 200). Opening one from inside the drawer only focused the
input (keyboard popped up) while the form stayed hidden underneath.

- openWorkspaceCreate / openProfileCreate / openSkillCreate /
  openCronCreate now close the drawer after rendering the form,
  mirroring their *_Detail siblings.
- move the fromRailClick drawer re-open in switchPanel() into the
  synchronous section: it used to run after await loadXxxPanel(),
  racing (and re-opening over) a drawer close issued from inside the
  drawer during the panel load window.

Verified in Playwright (390x844): drawer closes and stays closed after
plus-click, form visible with focus on the path input; desktop (1280x800)
unchanged; 83 mobile + 26 workspace tests pass.

* fix: close mobile sidebar after project quick-create plus

The per-project '+' button (_attachProjectQuickCreateButton) created the
new project conversation but never closed the mobile sidebar drawer, so on
phones the tap looked like a no-op: the session was created, just hidden
underneath the full-screen drawer (z-index 200).

Close the drawer in the success path after the sidebar repaint, mirroring
$('btnNewChat').onclick in boot.js and the #5409 close in
_openSidebarSession. Failure path keeps the drawer open so the error toast
stays visible for retry.

Adds static regression tests pinning the guarded close call and its
position after the repaint.

* test: pin quick-create drawer close to the success path (try block)

Greptile review note: the string-presence and ordering assertions alone
would still pass if the guarded closeMobileSidebar call were moved into
the later catch branch, silently reintroducing the mobile drawer bug.

Add an explicit assertion that the close sits before the success-path
catch(err), so a regression that moves it into the failure branch is
caught.

* docs(changelog): stamp batch1-p1 (#6921 #6933 #6671 #6175)

---------

Co-authored-by: webtecnica <75556242+webtecnica@users.noreply.github.com>
Co-authored-by: Rodrigo Gomes da Silva <rodrigo.smscom@gmail.com>
Co-authored-by: silent-reader-cn <silent-reader-cn@users.noreply.github.com>
Co-authored-by: nesquena-hermes <nesquena-hermes@users.noreply.github.com>
2026-08-13 16:41:32 +00:00
nesquena-hermesandreinocheong 4b390e115c Release v0.51.310 — Release JZ (#3760 — long-press project chips to manage on touch) (#3767)
* feat(sidebar): long-press project chips to open the context menu on touch (#3760)

Project filter chips could only be deleted/renamed via the right-click context
menu (oncontextmenu), which has no touch equivalent — so mobile/tablet users had
no way to delete a project from the sidebar; the list grew forever.

Adds a 500ms long-press gesture mirroring the existing session-item long-press
pattern: touchstart schedules the menu, touchmove cancels on >10px drift,
touchend suppresses the synthetic click when the long-press fired, touchcancel
cleans up. `.project-chip.long-pressing` gives accent + slight-scale feedback;
`touch-action:manipulation` + `user-select:none` + `-webkit-touch-callout:none`
prevent the native callout/selection from competing.

Maintainer fix on top of the contributor PR (multi-touch correctness, flagged in
review): touchstart now clears any in-flight `_lpTimer` before scheduling a new
one (a second finger / stray touchstart previously orphaned the prior timer,
which then fired unsuppressed ~500ms later and popped the menu after the gesture
was cancelled), and the timer body bails if `_lpHandled` is already set so a
stale fire is a no-op — matching the session-item belt-and-suspenders. Also
dropped a stale, unrelated issue reference from the original comment.

Co-authored-by: reinocheong <[email protected]>

* docs(changelog): stamp v0.51.310 — Release JZ (stage-3760 long-press project chips)

---------

Co-authored-by: nesquena-hermes <[email protected]>
2026-06-06 21:36:24 -07:00
d625bac6d4 v0.50.219: project chip context menu + input auto-sizing (#1087)
* fix(projects): opaque context menu + auto-sizing rename/create input

Two project chip UI bugs reported in project-ui-bugs.md:

1. Right-click context menu was transparent and the session list bled
   through it. Root cause: _showProjectContextMenu set
   background: var(--panel), but --panel is not defined anywhere in
   style.css, so the menu fell back to transparent. Fix: use
   var(--surface) -- the same opaque variable used by
   .session-action-menu and other floating popovers.

2. The rename and new-project input field was hard-coded to 100px
   regardless of the project name being edited (a 3-letter name got
   the same field size as a 20-letter name). Fix: drop width:100px
   from .project-create-input, replace with
   min-width:40px / max-width:180px / width:auto. Add a
   _resizeProjectInput() helper that measures the current value with
   a hidden span and sets pixel width inside those bounds. Wired into
   both _startProjectRename (called once on focus, again on every
   input event) and _startProjectCreate (same pattern).

Tests: 9 new static-source tests in tests/test_project_chip_ui.py
that pin (a) var(--panel) is undefined in style.css so the fallback
trap doesn't return; (b) menu uses var(--surface); (c) the fixed
width:100px is gone and min/max bounds are present; (d) the
_resizeProjectInput helper is defined and called from both flows.

Full suite: 2419 passed, 47 skipped, 0 PR-related failures.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix(projects): use getComputedStyle in _resizeProjectInput sizer span

Switch the hidden sizer span from hardcoded font-size:10px / font-family:inherit
to reading the live values from getComputedStyle(inp). This keeps the sizer
calibrated if the CSS rule ever changes, rather than silently drifting.

Also update test_resize_helper_uses_hidden_span to assert getComputedStyle
is used rather than the old literal font-size check.

Suggested by Opus independent review of #1086.

* docs: v0.50.219 release notes, test count 2467, roadmap update

---------

Co-authored-by: Nathan Esquenazi <nesquena@gmail.com>
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-authored-by: nesquena-hermes <nesquena-hermes@users.noreply.github.com>
2026-04-25 23:28:29 -07:00