feat(gnome.gio): gio 到位,pango 的闸门开了 —— 以及一次被推翻的判断

mcpp.toml

@@ -107,6 +107,7 @@ members = [
     "tests/examples/glib",
     "tests/examples/gobject",
     "tests/examples/gmodule",
+    "tests/examples/gio",
     "tests/examples/libgbm",
     "tests/examples/libpng",
     "tests/examples/libwebp",

pkgs/g/gnome.gio.lua

@@ -0,0 +1,207 @@
+-- gnome.gio — part of GLib 2.82.5.
+--
+-- Files, streams, sockets, D-Bus, GApplication, GSettings, GResource — and
+-- `GListModel`, which is what pango's `PangoFontMap` implements.
+--
+-- Upstream ships glib as one source tree producing FOUR separate shared
+-- libraries, and this index follows that split: `gnome.glib`, `gnome.gobject`,
+-- `gnome.gmodule` and `gnome.gio` are four packages from one fork, because
+-- that is what a consumer links.
+--
+-- ─────────────────────────────────────────────────────────────────────────
+-- ⭐ THIS PACKAGE EXISTS BECAUSE AN EARLIER ARGUMENT WAS WRONG
+--
+-- gio was left out of this index once, with the reason: two of its six
+-- generators are `gdbus-codegen`, an 8,351-line Python program that turns
+-- D-Bus interface XML into GObject skeletons, and reimplementing it in
+-- `build.mcpp` is not proportionate. That premise was measured and correct.
+-- The conclusion did not follow, because it conflated two different things:
+--
+--     reproducing the generator     ≠     obtaining its output
+--
+-- The build never needed the generator. It needed 15,392 lines of C that are a
+-- PURE FUNCTION of five XML files and the codegen version — no target, no
+-- host, no locale enters it. So `mcpp/tools/gengdbus.sh` produces them once,
+-- they live in `mcpp/generated/`, and a CI job REGENERATES AND DIFFS them on a
+-- runner with a different Python than the one that produced them.
+--
+-- That last part is the whole argument. Committed output has exactly one
+-- failure mode a build never sees: it can stop matching its input, and it
+-- still compiles. A diff is what makes "checked in" different from "stale".
+--
+-- The same arrangement `freedesktop.wayland-protocols-*` already used for its
+-- 195 generated files — where it was FORCED, since a package whose product is
+-- headers cannot generate at build time. Here it is not forced; it is simply
+-- the cheap way to get code a rewrite would otherwise have to produce.
+--
+-- ─────────────────────────────────────────────────────────────────────────
+-- THE OTHER FIVE GENERATORS ARE C++, AS EVERYWHERE ELSE IN THIS FORK
+--
+--     gio-visibility.h        gen_visibility("GIO", …)
+--     gnetworking.h           one substitution
+--     gioenumtypes.{h,c}      glib-mkenums over 82 enums
+--
+-- ⚠️ 82 ENUMS AGAINST gobject's FOUR, and what changes at that scale is not
+-- the loop but the ANNOTATIONS — with a trap in each direction:
+--
+--   gio writes `/*< private >*/`, `/*< public >*/` and `/*< protected >*/`
+--   144 times. Those are GTK-DOC annotations for STRUCT MEMBERS and mean
+--   nothing to mkenums. A scanner that read every `/*< … >*/` would silently
+--   drop enumerators.
+--
+--   `GConverterFlags` has NO `/*< flags >*/` despite the name, so upstream
+--   registers it with `g_enum_register_static`. A scanner that guessed from
+--   the type name would disagree with upstream's ABI.
+--
+-- The nicks are public API — `g_flags_get_value_by_nick` reads them — and gio
+-- overrides seventeen of them. Derived, `G_CONVERTER_NO_FLAGS` would be
+-- "no-flags"; upstream says "none".
+--
+-- `gconstructor_as_data.h` is upstream's sixth generator and is NOT here: it
+-- embeds `glib/gconstructor.h` as a C string for `glib-compile-resources`, a
+-- TOOL. See below.
+--
+-- ─────────────────────────────────────────────────────────────────────────
+-- ⚠️ WHAT gio NEEDS BESIDES gio/*.c — AND WHY EACH IS CHECKED BY NAME
+--
+-- gio is assembled from four kinds of input. A missing `gio/*.c` is an
+-- undefined reference, which the linker catches. The other three DEGRADE
+-- SILENTLY, so the example test asserts each one by name rather than trusting
+-- that it built:
+--
+--     gio/xdgmime/        absent → g_content_type_guess answers
+--                         "application/octet-stream" for everything
+--     gio/inotify/        absent → the file monitor quietly falls back to
+--                         polling; the test reads the backend's TYPE NAME,
+--                         which is the only place the two differ
+--     subprojects/gvdb/   absent → GResource and GSettings' schema reader are
+--                         never reachable, and nothing says so
+--     mcpp/generated/     stale → it still compiles; the portal calls fail
+--                         only inside a Flatpak sandbox
+--
+-- `-DXDG_PREFIX=_gio_xdg` is load-bearing rather than hygiene: xdgmime's
+-- header macro-renames every `xdg_mime_*` symbol, and `gcontenttype.c` calls
+-- the unprefixed spelling. Both sides must see the same value or the calls and
+-- the definitions are different symbols.
+--
+-- ─────────────────────────────────────────────────────────────────────────
+-- NO TOOLS, AND NO EXTERNAL DEPENDENCY BUT zlib
+--
+-- The ten tools — `gio-tool`, `gdbus-tool`, `glib-compile-schemas`,
+-- `glib-compile-resources`, `gsettings`, … — are NOT built. Each provides
+-- `main`, and a dependency that provides one collides with the consumer's;
+-- that is the rule every package in this index follows.
+--
+-- Upstream's meson names exactly one external dependency for the library
+-- beyond glib itself: `libelf`, used by `gresource-tool.c` — an
+-- `executable()`, not the library. Measured: no library source mentions elf at
+-- all. There is therefore NO libelf feature, because there is nothing for it
+-- to switch.
+--
+-- `libmount`, `selinux` and `sysprof` are `auto` upstream and absent here.
+-- Each is `#ifdef`-tested, so each is ABSENT rather than 0, and gio degrades
+-- the way upstream intends: GUnixMountMonitor falls back to reading
+-- /proc/self/mountinfo, and the SELinux and tracing attributes are not
+-- offered. Defining any of them as 0 would say "yes" — see `gnome.glib` for
+-- the nine macros that lesson came from.
+--
+-- ─────────────────────────────────────────────────────────────────────────
+-- ⚠️ NAME WHAT YOU USE; DO NOT NAME glib
+--
+--     [dependencies]
+--     gnome.gio     = "2.82.5"
+--     gnome.gobject = "2.82.5"
+--     # gnome.glib and gnome.gmodule arrive transitively
+--
+-- gio declares glib, gobject and gmodule as workspace PATH dependencies — they
+-- are built from one tree — and mcpp rejects a package requested both ways:
+--
+--     error: dependency 'gnome.glib' is requested as both a version dep
+--            (by 'your-package') and a path dep (by 'gnome.gio')
+--
+-- ─────────────────────────────────────────────────────────────────────────
+-- ⚠️ NO extern "C" WRAPPER
+--
+-- gio decorates every header with G_BEGIN_DECLS, so a wrapper is redundant —
+-- and harmful: gio.h reaches <stdlib.h>, which libc++ routes through <cstdlib>,
+-- which defines TEMPLATES. Inside an extern "C" block that is
+-- `templates must have C++ linkage`, dozens of times, against a header the
+-- consumer never named. libstdc++ does not route them that way, so gcc is
+-- green and clang is a wall.
+--
+-- ❌ "THERE IS NO MODULE" USED TO BE HERE, AND IT WAS THE WRONG CONCLUSION.
+-- The API being macro-heavy is true and does not make a module pointless: this
+-- one hands a consumer 2,850 declarations, and gio's function API — GFile,
+-- GSocket, GVariant, GListModel — needs no macros at all. See above.
+--
+-- ─────────────────────────────────────────────────────────────────────────
+-- LINUX ONLY
+--
+-- This is the Unix build: the Windows and Cocoa backends are excluded because
+-- they are ALTERNATIVES to the Unix ones, not companions — both halves define
+-- the volume monitor, the app-info backend and the settings backend. The
+-- generated `glibconfig.h` fixes `G_OS_UNIX` besides, and the upstream tree
+-- carries two `COPYING` symlinks a Windows extraction would not survive.
+-- ─────────────────────────────────────────────────────────────────────────
+-- ⭐ TWO WAYS TO CONSUME IT, AND YOU PICK ONE
+--
+--     import gnome.gio;          -- the module route
+--     #include <gio/gio.h>       -- the header route
+--
+-- The namespace is the contract in this index: `compat.xxx` means headers, an
+-- owner namespace means the package exposes `import`. This module exports
+-- 2,850 names. It re-exports all three siblings, because
+-- `gio.h` reaches all three and because a consumer CANNOT name them itself —
+-- they are workspace path dependencies.
+--
+-- ⚠️ THE TWO ROUTES DO NOT COMPOSE. A TU that imports the module AND textually
+-- includes a glib header reaches <time.h> twice — once through the module's
+-- global fragment, once directly — and the same `struct tm` from the same file
+-- becomes two entities:
+--
+--     error: conflicting declaration 'struct tm'
+--     note: previous declaration as 'struct tm'   (of module gnome.gio)
+--
+-- WHICH ROUTE IS DECIDED BY MACROS. A module cannot carry them, and glib's are
+-- half its API — 1,337 `#define` against 1,312 declarations for glib, 1,679
+-- against 1,753 for gio. `G_DEFINE_TYPE`, `G_OBJECT`, `g_signal_connect` and
+-- every `G_TYPE_*` are macros, so code that defines a GObject subclass takes
+-- the HEADER route. Code that uses the function API — most of gio — takes the
+-- MODULE route and includes nothing at all.
+--
+-- The wrapper is GENERATED from upstream's public headers, so a name upstream
+-- adds or removes cannot be silently missed. That is not a preference at this
+-- size: four separate silent misses were found by consumers rather than by the
+-- build — a brace inside a char literal that swallowed the rest of a file,
+-- `G_DECLARE_INTERFACE` (which expands to the names rather than spelling
+-- them), `<glibconfig.h>` being spelled without a `glib/` prefix, and glib's
+-- habit of parenthesising a name to defend it from macro expansion.
+--
+package = {
+    spec        = "1",
+    namespace   = "gnome",
+    name        = "gio",
+    description = "GIO 2.82.5 — files, streams, sockets, D-Bus, GSettings and GListModel",
+    licenses    = {"LGPL-2.1-or-later"},
+    repo        = "https://github.com/mcpplibs/glib",
+    type        = "package",
+
+    xpm = {
+        linux = {
+            ["2.82.5"] = {
+                url = {
+                    GLOBAL = "https://github.com/mcpplibs/glib/archive/refs/tags/2.82.5.tar.gz",
+                    -- ⚠️ The container tag is `2.82.5-4`, not `2.82.5`. gitcode
+                    -- refuses to REPLACE an asset of the same name in an
+                    -- existing release, so each corrected tarball needs a new
+                    -- container tag while the PACKAGE version stays upstream's.
+                    -- Verified byte-identical to the GLOBAL tag archive.
+                    CN     = "https://gitcode.com/mcpp-res/glib/releases/download/2.82.5-4/glib-2.82.5.tar.gz",
+                },
+                sha256 = "38a175bcd8899f376b6540e2e3ef7450169205fc56bdb909cfb086a0438553e5",
+            },
+        },
+    },
+
+    mcpp = "*/mcpp/gio/mcpp.toml",
+}