gir-to-d

Build times (auto-updated)

• x86_64: 20s (2026-09-06 07:25 UTC, serial)

---

Category: maintained/gir-to-d (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: gitlab commit: upgpkg: 0.23.4-3: Dropped gcc-libs dependency
• Conflict notes: n/a (Method B, no merge performed)

2026-09-05: build failure, two distinct causes, both fixed (pkgrel 3 -> 4)

• Cause 1 – ldc rejects GCC-style LDFLAGS. meson forwards the global hardening LDFLAGS (-Wl,-O1 -Wl,--sort-common ...) verbatim to the D compiler during its sanity check; this system’s ldc (1.43.0) doesn’t accept -Wl, passthrough at all (ldc: Unknown command line argument '-Wl,-O1', suggests --O1 instead), so meson setup fails before compiling a single line: “ERROR: Compiler ldc cannot compile programs.” Verified directly: ldc -Wl,-O1 t.d fails the same way; a synthetic meson project reproduces/fixes with LDFLAGS present/unset respectively. Fix: unset LDFLAGS in build(), replacing the previous -flto=auto->-flto=full sed (which is now moot – LTO was only ever requested via a flag folded into LDFLAGS). Costs FULL RELRO on the resulting binary (namcap now warns about it) – acceptable, not worth chasing further for one D binary.
• Cause 2 – real dmd/ldc compiler regression, surfaced only once cause 1 was fixed and actual D compilation started: /usr/include/dlang/ldc/core/internal/newaa.d(110): Error: variable ...Entry!(string, inout(Node)*).Entry.value - only parameters or stack-based variables can be inout, from gir-to-d’s own source/gtd/LinkedHasMap.d (Map = LinkedHashMap, used throughout GirPackage.d/GirStruct.d). Root cause: LinkedHashMap stores private Node*[Key] data; and exposes several inout-qualified accessors (get, opIndex, opBinaryRight “in”, opEquals, keys, values) over it; instantiating druntime’s internal AA machinery for an inout-qualified value type hits a real compiler restriction. Researched via GitHub API search (no gh CLI on this host): dlang/dmd#22510 (“dup of inout aa broke with 2.112”) is the same error signature, a genuine regression introduced in DMD 2.112, fixed by PR #22516 (merged 2026-02-05) and backported into 2.112.1 (2026-03-20) – but that fix was scoped narrowly to .dup() calls specifically. LinkedHashMap never calls .dup(); it hits the same underlying “inout can’t be a struct field” restriction through plain inout accessor methods instead, a broader pattern the #22510 fix doesn’t cover. No open upstream issue found for this broader case, and gir-to-d’s own git HEAD is exactly the 0.23.4 tag (no newer commits at all) – nothing to pull in from either side.
• Fix: fix-inout-linkedhashmap.patch (new, applied in prepare() after the existing add-main-context-pusher.patch) strips inout from those six accessors/their locals. Safe because Map/ LinkedHashMap is never used as const/immutable anywhere in gir-to-d – it’s an application, not a library, so losing const-correctness on a type that’s always used mutably costs nothing real. Verified: patch applies cleanly against a fresh tarball extract (patch --strip=1, matches the existing patch’s invocation style), the modified file compiles standalone, and a full build produces a working girtod binary (--help runs correctly).
• Both fixes bundled under pkgrel 4 (pkgrel 3, LDFLAGS-only, was never actually published – only pkgrel 4 with both fixes together was built+published), matching [feedback_pkgrel_vs_pkgver].