glibd

Build times (auto-updated)

• x86_64: 1m57s (2026-09-06 17:48 UTC, serial)

---

Category: maintained/glibd (adopted from official Arch repos – flagged by buildmaster’s “Missing state file” checker as a candidate to move to the AUR; see https://buildmaster.archlinux32.org/checker/summary.html).

2026-09-05: adopted from official repos

• Dropped from: extra-x86_64
• Method used: B (no AUR repo existed, cloned straight from gitlab packaging repo)
• Reason dropped upstream: unknown, needs manual look
• Conflict notes: n/a (Method B, no merge performed)

2026-09-05/06: three unrelated build failures, all fixed (pkgrel 5 -> 6)

Same ldc/meson toolchain as [[gir-to-d]] and [[mustache-d]], plus two problems unique to this package’s own .gir split and its meson.build.

• Fix 1 – ldc/LDFLAGS: same as gir-to-d/mustache-d. unset LDFLAGS in build() instead of the -flto=auto->-flto=full sed.
• Fix 2 – GLib/Gio “-unix” .gir split: same upstream restructuring as [[qt5-nemo-qml-plugin-configuration]]’s cause, but affecting this package directly instead of a dependency. Both GLib-2.0.gir and Gio-2.0.gir spun off separate GLibUnix-2.0.gir/GioUnix-2.0.gir files upstream, stripping the redundant unix_/Unix prefix from names in the process (namespace already implies it: g_unix_error_quark -> GIR name error_quark under namespace GLibUnix, etc.).
  • fix-glib-unix-split.patch: added file: GLibUnix-2.0.gir to the existing wrap: glib block in APILookupGLib.txt, renamed the 9 move: unix_* UnixUtils ... targets to drop the now-gone prefix (unix_fd_add -> fd_add etc.) – safe since these all stayed inside the same glib package, no naming collision risk.
  • fix-gio-unix-split.patch: same idea but Gio’s split is much bigger and DOES collide – GioUnix-2.0.gir declares its own InputStream/OutputStream classes (really GUnixInputStream/ GUnixOutputStream, subclasses of Gio.InputStream/OutputStream, same namespace-strip), which would clash with Gio-2.0.gir’s own same-named classes if merged into one wrap: gio package. Fix: a brand new wrap: gio_unix package (own APILookupGioUnix.txt, own D module namespace) instead of extending the existing one. Migrated DesktopAppInfo’s hand-written createFromFilename() override there verbatim (the class moved to GioUnix-2.0.gir entirely – Gio-2.0.gir now only has doc-comment links to it, no real class). Dropped the old UnixMountEntry/move-based binding for mount functions entirely – MountEntry/MountPoint are now real GIR records with real <method> elements in GioUnix-2.0.gir, so gir-to-d auto-generates proper methods with no customization needed at all.
  • First attempt used wrap: gio_unix… no wait, used wrap: gio-unix (hyphen) – broke immediately once compilation actually started: D doesn’t allow hyphens in module names (module gio-unix.c.functions; is a syntax error). Renamed to wrap: gio_unix (underscore). Lesson: gir-to-d’s wrap package name becomes a literal D module/package identifier – must be a valid D identifier, not just a valid directory name.
• Fix 3 – real meson bug, bare relative build-tree source paths: after fixes 1-2, hit ninja: error: 'X', needed by 'Y', missing and no known rule to make it – but for a DIFFERENT file every time depending on list order (proved via direct testing: filtering out the failing file just shifted the error to the next one in the list; reversing the whole list moved it to the new first entry). Root cause, traced directly into this system’s meson (/usr/lib/python3.14/site-packages/mesonbuild/interpreter/interpreter.py, source_strings_to_files()): a bare relative string source whose resolved location is under the build directory takes the File.from_built_file(subdir='', fname=<whole relative path>) path (not split into subdir/fname), so File.absolute_path() / relative_name() later compute build_root + '/' + fname where fname already starts with build/ (since meson.build passed run_command(girtod, ..., '--print-files', 'relative,' + source_root), and this project’s build dir sits directly inside its source dir as <srcdir>/build/, making relative-to-source_root and relative-to-build_root coincidentally look like the same shape of string) – doubling to a path that doesn’t exist. An absolute path takes a different branch (path_has_root -> true -> File.from_built_relative(), which DOES os.path.split() correctly first) and sidesteps the bug entirely. Fix: --print-files full instead of --print-files relative,SOURCE_ROOT in the run_command() girtod invocation (fix-generated-sources-abspath.patch) – one-line change, no meson.build restructuring needed, and it also transparently fixed gtkd/*.d (the static, non-GIR-derived files copy:’d into the same build dir), which had hit the identical bug and no longer needs any special-casing.
  • This is a real meson defect (confirmed by reading its own source, not a guess) – worth reporting upstream at some point, but out of scope here. Ruled out extensively before finding this: reordering sources, wrapping in files() (whole-list and per-element, doesn’t matter – files() on a build-tree-shaped string hits the exact same code path, no protection), a separate static_library() warm-up target sharing the list (bug is per-target, not project-wide), and a properly-declared custom_target() output placed first in the same list (didn’t matter either – the bug triggers on the first entry that RESOLVES to a build-tree path lacking a real producing target, regardless of what precedes it).
• Built and published (2.4.3-6, x86_64). qt5-quickcontrols-nemo’s own build (a separate package, see its own memory file) depends on this fix indirectly for nemo-qml-plugin-configuration’s Qt6->mlite chain, but glibd itself has no relation to that Qt5/Qt6 story – pure coincidence both needed the same GLib/Gio Unix .gir-split treatment around the same time.

2026-09-06: wired in the upstream test suite (pkgrel 6 -> 7)

tests/gobject/meson.build declares three real functional tests (gobject_dclosure_test/DClosureTest.d, gobject_objectg_test/ ObjectGTest.d with G_DEBUG=fatal-warnings, gobject_value_test/ ValueTest.d) exercising the generated GObject/GLib bindings themselves, not just compilation – these existed upstream but were never run by this PKGBUILD before. Added a check() calling meson test --print-errorlogs. Confirmed passing: Ok: 3 / Fail: 0. Packaging-only change, no source patch needed, so pkgrel bump not pkgver per project convention. Built, tested, and published (2.4.3-7, x86_64).