Release Core 2.14.0 and Desktop 0.5.3 - #28
Merged
Merged
Conversation
## Why Desktop sleep can outlast the browser login deadline and leave the renderer locked without a usable recovery token. ## What changed - Create a process-lifetime login through the private Desktop launcher. - Preserve its session-only cookie across reload, renewal and recovery. - Keep ordinary browser expiry, Agent authorization and older Core fallback. ## Testing Pass 164 Core backend cases, 95 Windows Desktop unit tests and isolated backend smoke on Windows and WSL. Cover simulated long idle, browser expiry, Agent gates, cookie policy and rejection after backend restart.
## What changed - Check destination conflicts before waiting for a busy SSH source, and revalidate the source after the user chooses an action. - Keep copy progress, conflict choices and cancellation visible in short Files windows. - Show the actual download path with a native folder reveal action. ## Testing Cover conflict ordering, progress during a 4 MiB transfer, source changes, cancellation and short-window layout. Windows Electron downloads through WSL Core pass the focused Files smoke. The full Desktop smoke stops at the existing window-focus assertion.
## Why Connected SSH tabs expose only the managed Agent Tunnel preset, leaving users without general local and remote TCP forwarding controls. ## What changed - Add temporary browser-owned local and remote forwards with status and Stop. - Share bounded duplex relays and reverse-port dispatch with Agent Tunnel. - Preserve Agent API permissions and reject non-loopback remote clients. ## Testing Exercise both directions through real OpenSSH and three jump hosts, large half-close and duplex transfers, independent Agent Tunnel cancellation, viewer ownership, late setup and Core API permission checks. Verify browser controls and the existing Core headless smoke collection.
## Why Settings retains a previously selected SSH profile across connections and only exposes the final route target. Browsers without Document PiP cannot open a floating terminal or Files window. ## What changed - Bind new-session drafts to structured metadata from the current tab. - Separate name and order management from atomic full-route editing. - Reject changed connection preparation before saving or connecting. - Reuse an ordinary popup when Document PiP is unavailable, restoring the terminal on close and handling browser navigation cleanup restrictions. ## Testing Headless checks, SSH profile/key and route regressions, native PiP regressions, and real Chromium/WebKit popup input, Files and restore checks pass. Linux WebKit may require native close after reloading an opaque-origin blank popup; macOS Safari still needs manual qualification.
## Why A single-node SSH entry requires an extra editor layer even though its fields fit within Settings. Direct and jump nodes need consistent key and host identity controls without duplicating their persistence behavior. ## What changed - Show Direct node fields inline and save name, endpoint and key together. - Share node controls and revision-checked graph/key persistence with routes. - Carry incomplete drafts and temporary keys into Add jump node while keeping the inline draft on cancellation and preserving other node references. - Keep multi-node details in the second layer and retain inline edits on reorder. ## Testing SSH context, shared-node preservation, key persistence, incomplete route promotion, cancel and cross-window save conflicts pass in isolated Chromium. Existing route, profile/key transfer and PiP regressions pass. Narrow-window controls, JavaScript syntax, Python compilation and static-site checks pass.
## Why tmux cannot infer StandTerm RGB support from COLORTERM alone and quantizes truecolor output when its client capabilities omit RGB. ## What changed - Answer bounded XTVERSION and XTGETTCAP queries with truthful capabilities. - Isolate automatic replies from human input and deduplicate live viewers. - Add a private POSIX terminfo overlay while preserving explicit overrides. - Document the explicit RGB fallback required by remote tmux 3.4. ## Testing Cover protocol bounds, authorization, replay, multiple viewers and overrides. Verify actual RGB cells through Chromium with local and SSH tmux 3.4 clients, including existing sessions, plus the complete 168-test backend smoke suite.
## Why Treating main as an omitted terminal can route explicitly targeted input to another handoff terminal. Continuous output can extend capture settling indefinitely after the initial output wait. ## What changed - Preserve explicit terminal and token selections while retaining latest-handoff defaults. - Bound capture by wait_ms and retain completed writes and captured output on timeout. - Document the contracts and add a staged UI copy review and translation table draft. ## Testing Backend smoke checks pass across 169 cases, including multi-terminal isolation, CLI mismatch rejection without writes, continuous and late output, and capture timeouts. All 56 CLI/helper smoke cases pass, including explicit overrides and omitted-target defaults. The 19-row copy table passes key and placeholder checks.
## What changed - Clarify Agent authorization, permission and token wording while preserving protocol values and access behavior. - Generate reviewed English and Traditional Chinese messages from a translation exchange table with English fallback. - Apply the saved local language on the next page opening and include required resources in Core bundles. ## Testing Agent browser coverage includes both languages, token lifecycle, unchanged active sessions after saving language, connection info and approval flows. Translation validation, safe interpolation, bilingual dialog layouts and Core bundle selection pass.
## What changed - Clarify browser authorization and session recovery with bilingual static, pending and error text while retaining the English login fallback. - Distinguish input approval, file-copy approval, rejection, pause and transfer cancellation without changing their protocol bindings. - Keep uncertain copy results visible in both details and the transfer queue, preserving diagnostic codes and exact paths. ## Testing The full browser and backend suites cover access-token rejection, device recovery, stale proposals, exact file-copy plans and cancellation barriers. Focused bilingual checks after the final layout correction verify visible warnings, safe dismissal and narrow windows. Catalog validation and English fallback checks pass.
## Why Desktop menus and toolbar text remain English even when Core uses another language. Setup and Core source selection also require the shell to own its language independently of the selected Core. ## What changed - Save the Desktop language per profile and apply it on the next launch. - Localize custom menus, toolbar labels and Agent help through 50 reviewed bilingual messages, preserving command IDs and per-tab authorization guidance. - Ship the shell catalog independently and check both catalog outputs for staleness. - Keep remaining workflow text in English and disclose partial coverage. ## Testing All 110 Desktop unit tests and seven catalog regression tests pass. Both locales pass the real toolbar DOM checks with mocked native IPC, including narrow layout, ARIA labels, SVG preservation and failure feedback without replay. Staged language assets match source and builder inclusion rules. Native OS menus, dialogs and installers still require platform acceptance.
## Why Browser access confirmations and diagnostic windows stay English after changing the Desktop language. The log status label also implies a live writability check although it reflects recorded write outcomes. ## What changed - Add 76 reviewed bilingual messages for browser access, diagnostics, About and external-browser confirmations. - Keep authorization actions, consent indices, sensitive clipboard payloads and sanitized error feedback unchanged. - Escape translated diagnostic text while retaining scriptless rendering and literal event JSON, identifiers, paths and version metadata. - Describe log write outcomes without promising current filesystem writability. ## Testing All 115 Desktop unit tests and seven catalog regression tests pass; both catalogs are current. Headless Chromium checks pass in both languages for diagnostic headings, literal paths/events, escaped markup, empty state and narrow layout. Native OS dialogs, menus and installer acceptance remain separate.
## Why Recording finalization can fail while close or quit is pending. Losing the file outcome across startup or notification awaits permits premature exit and hides the retained file locations. ## What changed - Cancel the current close or quit on failed or unconfirmed recording saves and show the original error with available partial and destination paths. - Preserve structured job results across concurrent operations and keep notification failure from changing publication or exit decisions. - Localize Capture settings, status and dialogs in the Desktop table while retaining typed actions, raw diagnostics and later explicit exit. - Keep paused recording controls visible in the compact toolbar. ## Testing All 133 Desktop unit tests and seven catalog tests pass. Both generated catalogs are current. English and Traditional Chinese toolbar DOM checks pass at 640px with mocked native IPC. Native OS dialogs, encoder smoke and installer acceptance remain separate.
## Why Environment preparation and Core recovery can run before the Core UI is available. Their English-only dialogs do not follow the Desktop language selected for the current launch. ## What changed - Localize setup consent, progress, cancellation, typed error explanations, per-mode cleanup, and Core source management through the review table. - Keep normal dialogs on the launch language and read the relevant mode profile for installer preparation and cleanup. - Preserve numeric actions, process ownership, raw diagnostics, retained files and restart consequences; remove an inaccurate shared Git promise. - Leave installer-wide summaries and remaining shell notices explicitly scheduled for a separate batch. ## Testing All 145 Desktop unit tests and seven catalog regression tests pass. Both catalogs are current. Six real setup DOM cases cover both languages and all platform display branches, with native processes mocked. Focused checks also cover conflicting installer profile preferences and the final copy changes. Native GUI and packaged acceptance remain separate.
## Why Remaining native notices ignore the Desktop language, and several messages infer more than their structured operation results establish. ## What changed - Localize port decisions, download results, installer summaries and native paste confirmation through the reviewed Desktop table. - Select the common effective installer profile language or fall back to English without writing a shared preference. - Describe retained result counts, partial maintenance changes and uncertain download or paste completion without suggesting automatic replay. - Preserve numeric decisions, typed states, clipboard guards, verification timing and installer ownership. ## Testing All 165 Desktop unit tests and seven catalog regression tests pass. Both catalogs are current. Bilingual fixtures cover partial uninstall, owner loss, mixed locale preferences, port persistence timing, pathless downloads and lost paste acknowledgments. Native GUI and packaged acceptance remain separate.
## Why Browser and mocked dialog tests cannot establish Windows font/layout or prove that a built installer contains every bootstrap helper. ## What changed - Add an isolated Windows Electron rendering probe for both Desktop locales, native menu labels, compact toolbar states and setup progress. - Let the setup fixture target an arbitrary build stage using an exact URL. - Compare all six packaged Python helpers against the matching stage. - Record the candidate installer and source hashes, automated acceptance and remaining native dialog, installer lifecycle and macOS limits. ## Testing Windows Node passes all 165 Desktop tests. Native localization captures pass for both locales. The fresh extracted installer payload passes inspection, and its Windows and WSL backends pass packaged smoke with isolated profiles. The installed StandTerm instance is not replaced; installation, upgrade and uninstall are not exercised.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Core 2.14.0 and Desktop 0.5.3 bring the accumulated SSH, Agent, interface and Desktop improvements to the public release. Windows x64 and macOS arm64 installers share release source
0c8e7f44409bf0d6a32d8c24efce5cba407742df; the final commit only updates public download links and the known-issue notice.What changed
TERM=xterm-256color.Testing
Desktop packages remain evaluation pre-releases. Windows is unsigned; macOS uses ad-hoc signing without notarization. No private repository history, internal handovers or local runtime files are included.