voculinux/snap/snapcraft.yaml

# Vocalinux Snap packaging
#
# First ship: strict confinement, default whisper.cpp engine path, GTK tray UI,
# mic capture, model download, and text injection (RemoteDesktop portal, wtype, xdotool).
# Informed by issue #48 and packaging/flatpak lessons on feat/flatpak-packaging.
#
# Build (from repo root, with snapcraft + LXD or Multipass):
#   snapcraft pack
# See docs/INSTALL.md for pack and store upload steps.

name: vocalinux
base: core24  # Ubuntu 24.04 / Python 3.12 (project requires-python >=3.11)
# Fallback version; live version is set from src/vocalinux/version.py via
# adopt-info + craftctl set in the vocalinux part (required for craftctl).
version: "0.17.0"
adopt-info: vocalinux
summary: Free offline voice dictation for Linux
description: |
  Vocalinux is free voice typing for Linux. Hold a shortcut or toggle dictation, and text lands in almost any app on X11 or Wayland.

  Speech recognition runs on your machine after you download a model. No Voca account. No cloud transcription on the default path.

  This snap ships the whisper.cpp engine plus VOSK. The optional OpenAI Whisper / PyTorch stack is not bundled (too large for the snap).

  Features:
    - System tray control and GTK settings
    - Offline dictation (models download on first use)
    - Microphone capture and short feedback tones
    - Text injection via the Wayland RemoteDesktop portal, wtype, or xdotool

  Sandbox notes:
    - Global hotkeys need: sudo snap connect vocalinux:raw-input
      and sudo snap connect vocalinux:hardware-observe
    - Native Wayland typing uses the RemoteDesktop portal: the compositor
      asks once on first dictation. The snap ships no uinput plug and no
      ydotool; the Store denied the super-privileged grant.
    - KDE Plasma: KWin's portal can mistype letter case (KDE bug 489021);
      set "Virtual Keyboard = IBus Wayland" so the IBus path delivers text.
    - Microphone may need: sudo snap connect vocalinux:audio-record
      The microphone list comes from the sound server, so pick a device
      in Settings rather than relying on the system default.
    - whisper.cpp runs on the CPU in this snap; it has no GPU backend.
      For Vulkan acceleration install the AppImage or use install.sh.
    - Config and models live under ~/snap/vocalinux/

grade: stable
confinement: strict
license: AGPL-3.0-only
icon: snap/gui/vocalinux.png
contact: https://github.com/VocaHQ/vocalinux/issues
issues: https://github.com/VocaHQ/vocalinux/issues
website: https://vocalinux.com
source-code: https://github.com/VocaHQ/vocalinux
donation: https://github.com/sponsors/jatinkrmalik

apps:
  vocalinux:
    command: bin/vocalinux
    extensions: [gnome]
    plugs:
      # gnome extension already grants desktop, desktop-legacy, gsettings,
      # opengl, wayland, x11, unity7 (tray / indicators).
      - audio-record # microphone (may require manual connect)
      - audio-playback # start/stop feedback tones (PipeWire-pulse / PulseAudio)
      - home # optional access outside snap HOME (shared files)
      - network # download speech models on first run
      - network-bind # optional remote_api engine
      # Global hotkeys via python-evdev (/dev/input/event*).
      # Super-privileged: not auto-connected; user must:
      #   sudo snap connect vocalinux:raw-input
      - raw-input
      # Needed for /proc/bus/input/devices; not auto-connected:
      #   sudo snap connect vocalinux:hardware-observe
      - hardware-observe
      # No uinput plug; the RemoteDesktop portal types on Wayland instead.
      # The Store denied the super-privileged grant
      # (https://forum.snapcraft.io/t/53528), and ydotool is not staged
      # because it cannot reach /dev/uinput under confinement.
    common-id: com.vocalinux.Vocalinux
    desktop: usr/share/applications/vocalinux.desktop
    environment:
      # Prefer bundled CLI helpers (xdotool, xsel, xclip, wl-copy).
      # Do NOT prepend $SNAP/usr/lib over the gnome extension platform path:
      # that loads a second GLib/GTK and segfaults on `from gi.repository import Gtk`.
      PATH: $SNAP/usr/bin:$SNAP/bin:$PATH
      # PortAudio opens ALSA first. Point at snap-local config that routes to Pulse.
      ALSA_CONFIG_PATH: $SNAP/etc/asound.conf
      # core24 gnome-46-2404: desktop-launch rebuilds gdk-pixbuf-loaders.cache
      # by running query-loaders, which picks the SVG loader from the first
      # LD_LIBRARY_PATH entry ending in /$CRAFT_ARCH_TRIPLET. Stage-packages
      # (via libavcodec and friends) put a gdk-pixbuf tree without
      # libpixbufloader_svg.so at $SNAP/usr/lib/$CRAFT_ARCH_TRIPLET, ahead of
      # gnome-platform. About page and other GdkPixbuf SVG loads then fail
      # with "Image type svg is not supported".
      # https://forum.snapcraft.io/t/missing-svg-support-for-gtk3-in-core24/48851
      LD_LIBRARY_PATH: $SNAP/gnome-platform/usr/lib/$CRAFT_ARCH_TRIPLET:$LD_LIBRARY_PATH

# Make ALSA data/plugins visible at the paths libasound expects.
layout:
  /usr/share/alsa:
    bind: $SNAP/usr/share/alsa
  /usr/lib/$CRAFT_ARCH_TRIPLET/alsa-lib:
    bind: $SNAP/usr/lib/$CRAFT_ARCH_TRIPLET/alsa-lib

parts:
  vocalinux:
    plugin: python
    source: .
    source-type: local
    build-environment:
      - PIP_DISABLE_PIP_VERSION_CHECK: "1"
      # pywhispercpp's PyPI wheel ships a libggml-cpu compiled for AVX2+FMA,
      # which dies with SIGILL at model load on pre-Haswell CPUs (#957), so
      # the pinned sdist is built instead of installing that wheel.
      # GGML_NATIVE=OFF alone is not the portable baseline either: while
      # ggml's native default is on it turns every GGML_* SIMD option on, so
      # each one is pinned off for the plain x86-64 baseline.
      - PIP_NO_BINARY: pywhispercpp
      - CMAKE_ARGS: "-DGGML_NATIVE=OFF -DGGML_SSE42=OFF -DGGML_AVX=OFF -DGGML_AVX2=OFF -DGGML_BMI2=OFF -DGGML_FMA=OFF -DGGML_F16C=OFF"
    build-packages:
      - build-essential
      - cmake
      - git
      - pkg-config
      - portaudio19-dev
      - libgirepository1.0-dev
      - libcairo2-dev
      - python3-dev
      - libdbus-1-dev
    stage-packages:
      # Only extras not provided by the gnome-46-2404 platform content snap.
      # Staging GTK/GLib/AppIndicator here duplicates the platform and can crash GI.
      - libportaudio2
      - libasound2t64  # Ubuntu 24.04 / core24 name for libasound2
      - libasound2-data
      - libasound2-plugins # pcm/ctl pulse → host PulseAudio/PipeWire
      - pulseaudio-utils # paplay for start/stop UI sounds; pactl lists microphones
      - xdotool
      - wtype # virtual-keyboard protocol (GNOME often rejects this)
      - x11-utils # xprop for XWayland WM_CLASS
      - xsel
      - xclip
      - wl-clipboard
    python-packages:
      - pip
      - setuptools
      - wheel
    override-pull: |
      craftctl default
      version="$(python3 -c "import pathlib, re; t = pathlib.Path('${CRAFT_PART_SRC}/src/vocalinux/version.py').read_text(); m = re.search(r'__version__\\s*=\\s*[\\'\"]([^\\'\"]+)', t); assert m, 'version not found'; print(m.group(1))")"
      craftctl set version="$version"
      # PyGObject/GI provided by gnome extension.
      sed -i '/^[[:space:]]*"PyGObject/d' "${CRAFT_PART_SRC}/pyproject.toml"
    override-build: |
      craftctl default
      install -Dm644 "${CRAFT_PROJECT_DIR}/snap/gui/vocalinux.desktop" \
        "${CRAFT_PART_INSTALL}/usr/share/applications/vocalinux.desktop"
      install -Dm644 "${CRAFT_PROJECT_DIR}/resources/icons/scalable/vocalinux.svg" \
        "${CRAFT_PART_INSTALL}/usr/share/icons/hicolor/scalable/apps/vocalinux.svg"
      # meta/gui is what snapd uses for the store/desktop icon when present.
      install -Dm644 "${CRAFT_PROJECT_DIR}/resources/icons/scalable/vocalinux.svg" \
        "${CRAFT_PART_INSTALL}/meta/gui/icon.svg"
      install -Dm644 "${CRAFT_PROJECT_DIR}/snap/local/alsa/asound.conf" \
        "${CRAFT_PART_INSTALL}/etc/asound.conf"