Route services.Session errors through the same classifier Connection
uses so RequestExtend/WaitExtend return a structured ClientError with a
clean localized short message instead of the raw daemon error. Extract
the shared errorClassifier into errors.go, and fall back to the gRPC
status code when no message substring matches, since the daemon now
forwards a clean desc without the English marker text.
Resolve conflicts and complete the profile display-name -> ID migration
across the daemon, CLI, UI services, tray, and React frontend.
- Regenerate client/proto daemon.pb.go from the merged proto so it carries
both branches' RPCs (ui-refactor: SubscribeStatus/RegisterUILog/ExtendAuthSession,
main: RenameProfile + id fields); keep the v6.33.1 generator header
- server.go: combine ui-refactor's profile-list-changed events with main's
id-bearing responses; publish profile-list-changed on rename
- Drop the deleted Fyne client_ui.go/profile.go and port main's profile-ID
changes into the refactored services/tray
- UI services, tray and React: send the profile ID as the daemon handle and
keep the display name for rendering only (activeProfileId vs activeProfile)
- Relax the profile-name input to match the daemon's sanitizeDisplayName
(spaces, emoji, any valid UTF-8); cap at 128
- Expose RenameProfile via a Profiles.Rename services wrapper (+ regenerated
bindings) for the frontend to wire up
- cmd/login.go: use the profile ID for GetProfileState
* Migrate to profile ids
* Migrate android profile manager
* Clean up
* Fix review
* Add ID type
* Fix test and runes in ShortID()
* Fix profile switch on up and android comments
* Revert android profile to string id
* Fix feedback
* Fix UI feedback
* Fix id assignment
* Add renaming of profiles
* Fix review
* Remove ui binary
* Fix getProfileConfigPath not validating id
* Change resolve handle order and fix server merge problems
* Fix mdm test
The _other.go fallback stubs in client/ui and client/ui/services matched
FreeBSD, dragging the Wails-importing services package into the FreeBSD
build and failing it (Wails v3 has no FreeBSD port). Add !freebsd (plus
!android/!ios/!js) to align them with the rest of these packages so the
whole UI is excluded on FreeBSD.
Add a no-op WailsUIReady RPC the UI probes once at startup. A reachable
daemon that returns Unimplemented predates this UI and is too old to
drive it; the probe distinguishes that from an unreachable daemon. On an
outdated daemon the UI fires a localized OS notification, since the main
window may not open to show an in-app error.
Proto regenerated with protoc 33.1 to keep the v6.33.1 header.
tray_theme_other.go references Tray but its build tag selected it on
FreeBSD, where tray.go (which defines Tray) is excluded via !freebsd.
Add the same !android && !ios && !freebsd && !js exclusions to the stub.
The daemon runs as SYSTEM and writes the bundle into C:\Windows\SystemTemp,
whose ACL denies the logged-in user, so a plain 'explorer /select' could not
open it. The Windows reveal now elevates via ShellExecuteW with the runas verb,
falling back to an unelevated reveal of the parent dir on decline.
The UI client migrated from Fyne to Wails v3 + React. Update the
Requirements and Build sections accordingly, fix the Windows installer
UI build command to use the task runner, bump the Go version to 1.25,
and drop the now-redundant client/ui/README.md.
The critical-event gate compared against a bare "critical" literal while
the matching string is produced from the proto SystemEvent_CRITICAL enum in
systemEventFromProto. Add services.SeverityCritical next to the Status*
constants so the two sides share one definition.
The critical-event gate used a case-sensitive == "critical" where every
other status/severity compare in the UI uses strings.EqualFold, so a
daemon-side reword to a different case would silently drop the bypass.
newXembedHost failure left the private session bus open, leaking an fd
per RegisterStatusNotifierItem on systems without an XEmbed tray. Add a
closeBus helper that logs on close failure and use it at every back-off
path instead of discarding the error.
Xlib's default protocol-error handler calls exit() on any async X error,
so a tray race (tray manager window dying between find and dock, XMoveWindow
on a popup the WM already destroyed) would take the whole UI process down.
Install process-global logging no-op handlers via XSetErrorHandler/
XSetIOErrorHandler after every XOpenDisplay. Since the handler slot is
process-global and GDK init can clobber it, additionally wrap the popup
code's raw Xlib calls on GDK's Display in gdk_x11_display_error_trap_push/
pop_ignored, which is independent of the global handler.
GetRestrictions called the daemon GetConfig with an empty
GetConfigRequest, which the daemon has rejected since multi-profile
support landed ("active profile name is empty") because it reads the
profile name from the request. Resolve the active profile via
GetActiveProfile first and pass its name/username, mirroring the
working Settings.GetConfig path. GetActiveProfile self-heals to the
default profile, so a valid name is always supplied; the
profile-switch-disabled MDM flag is orthogonal and unaffected.
WebKit (macOS WKWebView) omits the "Name: message" header from
Error.stack, so forwarding only the stack hid the real cause of
failures. Prepend the error name, message and optional cause before
the stack so daemon RPC errors surface in gui-client.log.
Eduard Gert