Packages

LFS ch8. Book offers 3 variants (BIOS, 64-bit UEFI, 32-bit UEFI) -- BIOS chosen since that's what our QEMU install target boots by default; UEFI can be added as a separate package later if needed. This installs grub's utilities (grub-install, grub-mkrescue, etc), not the actual boot sector -- that's a later ISO-assembly step. Skips the optional test suite.

Not in BLFS's core list for this minimal-XFCE build (deliberately skipped originally since nothing in scope looked like it needed it). Real, significant bug found live during the second speed-optimization pass: at-spi2-core's at-spi-bus-launcher hard-requires the 'org.gnome.desktop.interface' and 'org.gnome.desktop.a11y.applications' GSettings schemas (confirmed by reading bus/at-spi-bus-launcher's own get_schema() in the real at-spi2-core source tree) -- without them, g_settings_schema_source_lookup() finds nothing and the launcher calls g_error() (a deliberate abort()), producing the 'Cannot get the default GSettingsSchemaSource - is the gsettings-desktop-schemas package installed?' crash seen repeatedly in dmesg/journalctl throughout this project's session history, always previously dismissed as a harmless side warning. It is NOT harmless: every GTK/AT-SPI-aware application that tries to connect to org.a11y.Bus at window-realization time blocks for the FULL systemd D-Bus service-activation timeout (service_start_timeout=120000ms, i.e. up to two real minutes) before its window finally appears, since the crashed launcher never successfully registers the bus name and each new activation request immediately crashes again rather than the caller reacting to the crash quickly. Diagnosed for real: launched the Phase 1.5 installer via the desktop icon, watched it appear to 'hang' with zero visible window for over a minute, confirmed via `ps` that the underlying process chain (xfce4-terminal -> su -> the installer) was genuinely alive and NOT stuck, then waited out the full delay and captured a screenshot of the installer window finally appearing mid-progress, already partitioning/formatting/copying -- proving the process had been running the whole time, just invisible. This is very likely the single largest real contributor to this project's 'apps feel slow to open' complaint that motivated the whole preload/governor pass -- far larger than anything squashfs/preload/cpufreq tuning could fix, since it affects EVERY GTK app's window-open latency, not just cold-cache first-reads. Plain meson build, only real dependency beyond glib2 (already built) is glib-mkenums (part of glib2). introspection explicitly disabled (-Dintrospection=false) since nothing in this project's stack needs GIR data for these schemas specifically, matching the project's general bias toward skipping g-ir-scanner where not strictly needed. || Bumped 48.0 -> 51.0 (2026-09-18, part of the GNOME 51 push). Real sha256 verified against GNOME's own published sha256sum file.

GSound -- small GObject wrapper around libcanberra for playing system sounds. cinnamon-settings (the whole Cinnamon Settings GUI) hard-requires the GSound-1.0 typelib (cinnamon-settings/bin/util.py does gi.require_version('GSound','1.0')) -- without it every Cinnamon Settings panel, including Display, fails to launch with 'Namespace GSound not available'. Plain meson. Cinnamon 6.4 desktop, ship_in_iso=false.

GStreamer 1.24.12 -- CORE only (libgstreamer-1.0, gst-launch, the registry). No gst-plugins-base/good/bad. Budgie Desktop 10.10's meson has a hard top-level dependency('gstreamer-1.0') (used for the sound-effect / media-key path). meson; tests/examples/docs/benchmarks off, introspection on, ptp-helper perms none (no setcap in the build sandbox). ship_in_iso=false.

BLFS x/gtk2.html -- last real BLFS page is from BLFS 11.2 (GTK2 was dropped from the current/svn book entirely as obsolete, no newer page exists). Added as a real, previously-unbuilt legacy toolkit specifically because packages/audacity's own Conan-managed wxWidgets build hard-targets GTK2 for this Audacity version (confirmed directly: 'Audacity is still built against GTK2 internally for various reasons' -- a real, current upstream limitation, not a misconfiguration on this project's part; a system-wxWidgets/GTK3 workaround exists upstream but has documented real friction). Every real required dependency (ATK -- folded into at-spi2-core upstream, same as documented elsewhere in this project -- gdk-pixbuf, Pango) was already built here for the GTK3/XFCE stack; only GTK2 itself was missing. Skips the optional test suite and the DocBook-utils/gnome-themes-extra/gobject-introspection/GTK-Doc optional extras -- this is purely a build-time dependency satisfying wxWidgets' Conan recipe, not a real desktop toolkit anything else in this project uses. Real build failure hit and fixed: GTK+2's own code (last released ~2020) predates GCC 14's much stricter default C standard and hard-errors on things like incompatible pointer assignments (a real 'error: assignment to GObject* from incompatible pointer type GtkBuildable*' in gtkcomboboxtext.c) -- the same old-code-vs-modern-GCC-defaults failure class already hit repeatedly elsewhere in this project (NetSurf's calloc-transposed-args, Zip needing -std=gnu89). Fixed with CFLAGS="-Wno-error=incompatible-pointer-types -Wno-error=implicit-function-declaration" rather than patching every individual call site across this large, old codebase -- downgrades these back to warnings (their original severity when this code was written), not silencing anything GTK2 itself considered fatal.

BLFS 'GTK-3'. man=false (no rst2man/docutils built, same reasoning as gdk-pixbuf/glib2). broadway_backend=false (HTML5 backend, out of the minimal-XFCE scope). x11_backend=true always. introspection=true (rebuilt for the Phase 1.5 GTK3+PyGObject installer -- was false originally; needs glib2 and gobject-introspection to both already be built/rebuilt first). Skips the optional dbus-run-session test suite. || 2026-09-28: wayland_backend flipped false -> true, part of the Budgie 10.10 Wayland chain (its own gtk-layer-shell dependency needs a real GDK Wayland backend to create layer-shell surfaces on labwc). The original false setting was load-bearing at the time ("meson's auto-detect wasn't enough to avoid a wayland-client hard error") because this project had no wayland-client/wayland-protocols built yet -- both now exist (built for the GNOME/KDE Wayland work earlier this same day), so the hard error no longer applies. This is additive, not a replacement -- x11_backend stays true, so every existing GTK3 app keeps running exactly as before under X11; only apps that explicitly request the Wayland backend (or run under a session with no X11 DISPLAY at all, i.e. only WAYLAND_DISPLAY) would ever pick the new backend. `ship_in_iso` is unset (defaults true) for this package -- unlike the desktop-specific packages built this same session, a rebuild of gtk3 merges straight onto the live $LFS/`.101`'s own `/`, so real regression-testing of ordinary GTK3/X11 apps after this rebuild matters more here than for any other package touched today.

GTK 4 -- toolkit for GNOME Shell's control-center bits, Nautilus, libadwaita. Built X11+Wayland, no vulkan/gstreamer/colord for a lean build. ship_in_iso=false -- GNOME Shell is a Packages-page / Package Manager install, kept off linuxdoors.iso (XFCE + MATE are the shipped desktops). Built on .101 against the existing rustc/llvm/p11-kit/librsvg/gi/glib2/pango/cairo stack; only the GNOME-specific libraries are new packages. Bumped 4.18.6 -> 4.22.5 (2026-09-18, part of the GNOME 51 push) -- libadwaita 1.8.8 (the version actually used, NOT 1.10 -- see packages/libadwaita's own notes) needs GTK4 >= 4.19.4, and GTK 4.22.x's own real glib requirement is only >= 2.84, comfortably satisfied by this project's existing glib2 2.88.3. Deliberately did NOT jump to GTK 4.24.0, which needs glib >= 2.89.3 -- just above what we have -- since glib2 is ship_in_iso:true and depended on by 66 other packages; bumping it would have been a much bigger, riskier, base-system-wide change for no real benefit here. Real sha256 verified against GNOME's own published sha256sum file.

2026-09-28: new package, part of the Budgie 10.10 Wayland chain -- Budgie's own panel uses this real library (`gtk-layer-shell-0 >= 0.8.0` per its meson.build) to create proper wlr-layer-shell panel/dock surfaces on labwc (or any other wlroots-based compositor). Real, minimal dependency list (gtk+-3.0, wayland-client/server/scanner/protocols) -- needs this project's gtk3 rebuilt with `wayland_backend=true` (see that package's own notes for the same-day rebuild) since this library calls GDK's Wayland-specific window-creation APIs internally; it cannot function against an X11-only GDK. `introspection=true`/`vapi=true` (both real meson defaults) need gobject-introspection (already built, Phase 1.5 installer work) and vapigen (part of this project's own vala package, already built for Budgie's gee-0.8 dependency).

C++ bindings for GTK 3 -- provides the gtkmm-3.0 pkg-config module mate-system-monitor is written against. meson; --libdir=lib; -Dbuild-demos=false -Dbuild-tests=false. ship_in_iso:false.

BLFS. GTK3 source-code text widget -- required by Mousepad (syntax highlighting) and pluma (see packages/pluma). GObject Introspection was originally skipped (gir=false), consistent with the rest of the stack at the time -- but this predates gobject-introspection existing in this project at all (same silent-skip class as gdk-pixbuf/harfbuzz/at-spi2-core during the Phase 1.5 installer work). Rebuilt with gir=true after pluma's own plugin-engine C source turned out to unconditionally #include <girepository.h> regardless of pluma's own --enable-introspection flag (an upstream inconsistency: libpeas's .pc doesn't propagate girepository's CFLAGS, but pluma's plugin loader glue code needs the header anyway) -- and separately, pluma's own libpluma-*.gir generation needs GtkSource-4.gir to exist as an include. Real build bug: gir=true alone then failed on a SEPARATE, independent `vapi` option (default true, needs `vapigen`/Vala, never built in this project -- same class as dconf's own vapi=false fix). Fixed with `-D vapi=false`.

GVfs userspace virtual filesystem -- built minimal (computer:// / trash:// / network:// backends only; every network/device/cloud backend and both volume monitors disabled) so MATE's caja desktop 'Computer' icon opens instead of erroring 'Operation not supported'. LinuxDoors has no udisks2, so computer:// shows the filesystem root + fstab user-mountable entries via glib's GUnixMountMonitor -- no removable-media auto-mount, but the icon works. Part of the MATE package set (ship_in_iso:false). meson build; --libdir=lib for the .101 multiarch-default gotcha. After apply, gio-querymodules /usr/lib/gio/modules must be run so libgvfsdbus.so is registered (the pkg-manager backend now does update-mime-database but NOT gio-querymodules -- worth adding). SMB/Windows-share browsing (network://) additionally needs samba + gvfs -Dsmb=true, not built here.

LFS ch8, final native build. Skips the optional make check.

BLFS. -D graphite2=enabled dropped since we don't build graphite2 (not needed for our minimal Pango/GTK3 use); harfbuzz falls back to its built-in shaper without it. introspection=enabled added later for the Phase 1.5 GTK3+PyGObject installer -- pango's own introspection rebuild needs HarfBuzz-0.0.gir, which this package never produced originally (built before gobject-introspection existed, same silent auto-detect skip as gdk-pixbuf). Skips the optional test suite.

BLFS. The required fallback icon theme. adwaita-icon-theme is skipped -- it needs librsvg, which is Rust-only since 2.40, and we're avoiding a Rust toolchain (same call made for gdk-pixbuf's glycin loaders). XFCE falls back to its own default icons instead of Adwaita's.

BLFS general/hunspell.html -- spell-check library. Sonnet (KDE Frameworks) refuses to build with zero backend plugins; hunspell is the standard one, and Kdenlive's title editor uses spell-check. ship_in_iso=false (Kdenlive stack = packages only). Plain autotools, --disable-static. Dictionaries are runtime data, not needed for the build.

BLFS general-libraries. Hardware ID databases -- required by libdisplay-info, needed by libxfce4windowing. No test suite.

LFS ch8. Just two data files (services, protocols) copied into /etc, no build. Bumped 20260723 -> 20260805 to track LFS 13.1's package set (sha256 from a real downloaded tarball).

IBus 1.5.31 -- input-method bus. libibus + IBus-1.0 typelib (gnome-shell hard-imports it). Built without the GTK immodules / setup UI / emoji+unicode dicts. ship_in_iso=false, built on .101.

Build-time-only Perl tool (icon-name-mapping) required by mate-icon-theme's configure. Needs XML::Simple, which LinuxDoors lacks -- build.sh cpanm's it into a scratch PERL5LIB (pure-perl, uses the already-present XML::Parser backend). Not shipped in the ISO.

BLFS general/icu.html. Recommended by both Node.js (--with-intl=system-icu) and Firefox. Tarball expands to a top-level 'icu' directory, source lives in icu/source.

BLFS general/imath.html -- Imath (half/vector/matrix math from OpenEXR). Required by Kdenlive 26.08 (find_package(Imath REQUIRED)) for its OpenTimelineIO import/export. ship_in_iso=false (Kdenlive stack = packages only). Plain CMake, --libdir=lib, python bindings off.

LFS ch8. Skips the optional make check.

BLFS general/inih.html (tiny .INI file parser, produces libinih.so + libINIReader.so). Second real dependency exiv2's CMakeLists.txt hard-requires despite BLFS listing it as merely 'Recommended' -- same gotcha as brotli (see packages/brotli/metadata.json), caught by a real 'inih library not found' CMake configure failure, not guessed.