merge: sync upstream cjpais/Handy v0.9.5 - #4
Open
github-actions[bot] wants to merge 50 commits into
Open
Conversation
* Refactor Sidebar settings structure * re arrange to my liking --------- Co-authored-by: CJ Pais <cj@cjpais.com>
…is#369) * feat: add double-click support for tray icon * windows left click open window --------- Co-authored-by: CJ Pais <cj@cjpais.com>
Paste-and-restore only ever saved plain text, so a clipboard holding a screenshot came back as an empty string — the copied image was silently destroyed by dictating. Save the image too, but only probe for one when there is no text to restore, so the common text path costs exactly what it did before and behaves identically. When the clipboard held nothing at all, clear it rather than leaving the transcription behind. Deliberately scoped to text and images via the clipboard plugin already in the tree: no new dependency, no new setting. HTML formatting and file lists still aren't preserved — they need arboard directly, and there are no reports of anyone hitting those. Fixes cjpais#921 Co-authored-by: CJ Pais <cj@cjpais.com> Co-authored-by: Egor Sokolov <236158718+egsok@users.noreply.github.com> Co-authored-by: Nazmus Sayad <87106526+NazmusSayad@users.noreply.github.com>
* fix: place Windows overlay on secondary monitors * fixes which work on my machine --------- Co-authored-by: CJ Pais <cj@cjpais.com>
* logging and retry for hf * mirror backup for dl * back to parallel dl but backoff * reorg some of the changes
…h-TW) (cjpais#1798) * fix(i18n): resolve zh-Hant-* system locales to Traditional Chinese (zh-TW) On macOS with system language set to Traditional Chinese (Taiwan), the OS reports the locale with a script subtag (zh-Hant-TW). Both language resolution paths missed it and fell back to the bare "zh" prefix, i.e. Simplified Chinese: - frontend: getSupportedLanguage() in src/i18n/index.ts (webview UI) - backend: get_tray_translations() in src-tauri/src/tray_i18n.rs (tray menu) Add a script-aware fallback to both: a zh locale whose second subtag is "hant" resolves to zh-TW before the language-only prefix match. zh-Hans-* still resolves to zh via the existing prefix path; saved zh-TW preferences still exact-match first; other locales are unchanged. Fixes cjpais#1794 * fix pre-existing bug and add tests * improve Chinese locale fallbacks * simplify --------- Co-authored-by: CJ Pais <cj@cjpais.com>
* secure input warning and fallback * translations * change icon to be better * better secure input dialog * harden secure input fallback lifecycle
* dont fire bindings when recording * regen bindings + format
…ais#1813) The `db` value computed in AudioVisualiser::feed() is not true dBFS: it is a per-bin average magnitude divided by the FFT window size, which for speech-shaped input lands roughly 20 dB below the band's actual dBFS. DB_MIN/DB_MAX were calibrated as if it were dBFS, so the window sat about 20 dB too high: - src-tauri/src/audio_toolkit/audio/visualizer.rs (level buckets) - src/overlay/RecordingOverlay.tsx (draws 3 + v^0.7 * 15 px bars) Normal dictation therefore mapped to ~0.1, which the overlay draws as a 4 px bar against a 3 px floor and an 18 px maximum — roughly one pixel of travel, which reads as a frozen waveform. Recalibrate to -68/-30, measured against captured audio on a built-in mic at 48 kHz (dictation ~-32 dBFS median, room tone ~-48 dBFS) and scored on rendered pixels: speech gets ~5 px of travel while room tone stays pinned at the floor. Not lowered to -70, where a noisy room starts making the idle waveform twitch. Fixes cjpais#1694 Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
cjpais#1668) * docs: replace Parakeet section with Parakeet Unified EN (huggingface-cli install) * corrections --------- Co-authored-by: andrzej.nescior <andrzej.nescior@nielsen.com> Co-authored-by: CJ Pais <cj@cjpais.com>
* add more reliable paste * translations * minor fixes
* drop sigusr1 on linux Co-authored-by: Murilo Vasconcelos <muriloime@gmail.com> * consolidate signal setup into signal_handle.rs --------- Co-authored-by: Murilo Vasconcelos <muriloime@gmail.com>
The Linux dependency lists were missing libraries that several bundled crates require, so a clean Fedora build failed in three separate rounds: - evdev-sys needs libevdev (libevdev-devel / libevdev-dev / libevdev) - whisper-rs-sys (bindgen) needs libclang (clang + *-devel) - whisper-rs-sys (ggml-vulkan) needs the glslc shader compiler Fedora was missing all three; Ubuntu was missing libevdev + libclang; Arch was missing all three (shaderc provides glslc). Add them so a fresh Linux checkout builds without hunting down each failure. Co-authored-by: CJ Pais <cj@cjpais.com>
… (cjpais#1753) * fix: link portable update dialog straight to the correct installer Portable installs can't self-update in place (no installer, and Windows won't let a running exe replace itself), so the update dialog directs users to download the NSIS setup.exe manually. It previously opened the generic releases page, forcing the user to hand-pick one of ~27 assets (issue cjpais#1750: "you need to find it"). Resolve the exact installer for the running architecture from the update version + `plugin-os` arch() and deep-link that asset directly. Falls back to the releases page when the version can't be determined. Also reword the dialog to note the Data/ folder is preserved on reinstall. URL construction is a pure helper with a standalone assert check (no unit-test runner in this repo; runnable via bun). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * add proper translations and clean up the pr --------- Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com> Co-authored-by: CJ Pais <cj@cjpais.com>
…1838) * fix: recover microphone stream after the capture worker dies run_consumer is driven entirely by the sample channel, so when cpal tears the stream down mid-session (device unplugged, USB/Bluetooth dropout) sample_rx.recv() returns Err and the worker thread exits. cmd_tx and worker_handle stay populated, so both AudioRecorder::open() and AudioRecordingManager::start_microphone_stream() still report the stream as open: recording captures nothing, stop() fails on the closed channel, and the app stays wedged until the on-demand close timeout resets it. Detect the exited worker and rebuild the stream instead of handing back a dead recorder. Fixes cjpais#1743 * fix for always on as well --------- Co-authored-by: CJ Pais <cj@cjpais.com>
* feat(nix): add Cachix binary cache to CI Replace magic-nix-cache (GitHub-local only) with cachix-action so build artifacts are pushed to a public binary cache. Nix users can then pull pre-built binaries instead of compiling from source (~25 min). Requires repo owner to: 1. Create the cache: cachix create handy 2. Add CACHIX_AUTH_TOKEN to GitHub repo secrets * style: fix prettier formatting in nix-check.yml * fix(nix): use correct cache name 'handy-computer' * chore(ci): pin Cachix action v17 --------- Co-authored-by: CJ Pais <cj@cjpais.com>
…s_recording (cjpais#1716) * fix(audio): move blocking cpal work off the main thread + lock-free is_recording Synchronous #[tauri::command] handlers run inline on the webview/main run loop, and the audio manager guards its state with a std Mutex held across blocking CoreAudio syscalls (cpal stream start/stop, device enumeration). A worker holding that mutex across a slow device open/close (Bluetooth/USB mic) serializes the main thread, freezing the UI (spinning beachball). Fix A: is_recording() reads a lock-free Arc<AtomicBool> mirror of the "state in {Recording, Stopping}" membership, flipped at the state transitions, instead of locking `state`. The hot-path UI poll can no longer deadlock against a worker holding `state`. Fix B: the four cpal-running commands (update_microphone_mode, get_available_microphones, set_selected_microphone, get_available_output_devices) become async and run their blocking cpal work via tokio::task::spawn_blocking. Tauri's invoke is identical for sync/async commands, so there is no frontend/binding change. Live verification (Bluetooth/USB mic recording + device change mid-use) is left to manual testing; concurrency/hardware behavior is not unit-testable. * add helper function --------- Co-authored-by: CJ Pais <cj@cjpais.com>
* fix: preserve HTTP transport error causes * fix: keep HTTP diagnostics free of sensitive data
* This fixes rendering corruption (cropping) after monitor reconfiguration This prevents rendering corruption/cropping after monitor reconfiguration (display off/on, disconnect/reconnect, or DPI changes while hidden) by reapplying the overlay's logical size (OVERLAY_WIDTH/OVERLAY_HEIGHT). Refactor overlay listeners and visibility handling Move listener setup out of an inner async function and track unlisten handles so cleanup runs from the effect. Add an isVisibleRef (kept in sync via an effect) and early-return in the mic-level handler to avoid processing level updates when the overlay is hidden. Reset levels and overlay state when receiving hide-overlay. These changes prevent stale-closure issues, avoid unnecessary mic processing while hidden, and ensure listeners are properly torn down. * fix for 0.9 version I didn't have any problems with that. Fix now. * fix(overlay): render nothing while hidden --------- Co-authored-by: CJ Pais <cj@cjpais.com>
…is#1284) (cjpais#1308) On systems where debloaters or privacy tools (O&O ShutUp10, etc.) set the UWP master key (HKCU\...\ConsentStore\microphone) to "deny", the app would get stuck on the mic permission onboarding page even though desktop app access (NonPackaged key) was correctly set to "allow". Change the priority logic so that desktop_app_access (NonPackaged) takes precedence over app_access (UWP master) when determining overall mic permission status, since Handy is a desktop app and only needs the NonPackaged scope. Co-authored-by: CJ Pais <cj@cjpais.com>
…jpais#1254) * fix: add input channel selection for multi-channel audio interfaces Multi-channel audio interfaces (e.g. Focusrite Scarlett, MOTU M4) expose loopback/output channels alongside mic inputs. Handy averaged all channels to mono, causing music and other audio to bleed into transcription. Add an "Input Channel" dropdown to settings that only appears when the selected microphone has more than one channel. Default is "Average all channels" which preserves the original behavior. Users with multi-channel interfaces can select their mic's specific channel. Fixes cjpais#591 * fix: preserve recorder when changing input channel * add translations --------- Co-authored-by: CJ Pais <cj@cjpais.com>
…s#1487) * fix(paste): release the modifier keys after pasting with wtype * format --------- Co-authored-by: CJ Pais <cj@cjpais.com>
…s of models (cjpais#1815) * feat: include streaming and translation filter buttons for easy access of models * style: standardize model button sizes and add divider * format + translations --------- Co-authored-by: vscg <vscg@vscg.com> Co-authored-by: CJ Pais <cj@cjpais.com>
* remove .so from appimage * narrow exclusions
…de (cjpais#1873) In always-on mode the capture stream stays open continuously so recording can start with no latency. While idle (not recording) each incoming chunk still entered the level-meter path and the resampler, but handle_frame returns early when not recording, so the resampled output was discarded and the level meter has no idle consumer. Skip both while idle to stop doing unnecessary work whose output is thrown away. Cmd::Start resets the visualizer and the resampler, so both resume cleanly when recording begins.
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
20 participants
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.
Upstream sync:
v0.9.4→v0.9.5自動生成 PR (
upstream-sync.yml)。upstream リリースノート
A bug fix release!
What's Changed
New Contributors
Full Changelog: cjpais/Handy@v0.9.4...v0.9.5
upstream commits
機械解決済み
upstream workflows 差分 (未取込)
残存コンフリクト (手動解決必要)
ビルド検証 (macos-14)
skippedfailure手動対応チェックリスト
src-tauri/src/managers/model.rsのセマンティックコンフリクト解決src-tauri/src/managers/transcription.rsのセマンティックコンフリクト解決 (CJK token rate / MoonshineParams 維持)~/.cargo/registry/src/index.crates-io-*/transcribe-rs-X.Y.Z/とsrc-tauri/transcribe-rs-patch/の diff)bun run tauri dev起動確認