fix(模块): 托管方不是负责方 —— 改名 cairo/displayinfo,并补齐两个不完备的包装体

改名:freedesktop.cairo -> cairo,freedesktop.displayinfo -> displayinfo。
索引 namespace 是货架标签,从不进模块名。cairo 是它自己的项目,freedesktop.org
托管 git 但不拥有那套接口;libdisplay-info 解析的是 VESA 的 EDID/DisplayID。同
一条规则下 freedesktop.egl 导出的是 khronos.egl。

⭐ 而且两个包装体都不完备:

    cairo         470 -> 697   缺 cairo_t,以及全部 192 个枚举量
    displayinfo   206 -> 501   缺全部 295 个枚举量

cairo_t 的原因很具体:正则 cairo_[a-z0-9_]+_t 要求 cairo_ 与 _t 之间至少一个
字符,于是 `typedef struct _cairo cairo_t;` —— 这个库里用得最多的类型 —— 恰好
是它唯一匹配不到的名字。

为什么一直没人发现,这才是要记的:
  - cairo:唯一 import 那个模块的测试同时写了 #include <cairo.h>,头文件把模块
    缺的都补上了。一个从没被要求独立站住的 import,不是被测试了,是被装饰了。
  - displayinfo:没有任何测试点过一个枚举量。而且只在 GCC 上加一个也发现不了 ——
    导出 enum 会让枚举量在 GCC 可见,clang 拒绝同一个文件。

它是从外面被发现的:gnome.pangocairo 不能混用两条消费路线,只能问模块要
cairo_t,结果什么都没拿到。pango 那一版因此自己扫了 cairo.h;现在 cairo 修好,
那个变通已经删掉,普通的 export import cairo; 就够了。

发布形态:这两个包用 release 资产而不是 tag 归档,所以走 -mcpp3 新资产,tag 不
动、零窗口。pango 重切 tag(sha 0882a260…,CN 容器 tag 1.56.1-5)。

⚠️ 验证时撞上了本地 store 按 (name, version) 索引的固定代价:已解开过 1.18.2 的
机器拿不到新资产,报 `failed to read compiled module`,清掉 store 条目才对。已
写进 §23.4。

cairo 的测试拆成两个文件,一条路线一个,module.cpp 一个头都不包;displayinfo 的
测试比较两个枚举量 —— 导出清单再塌陷就是编译错误。

五个示例 × 两条工具链全绿。

Author: Sunrisepeak

.agents/docs/2026-08-30-graphics-stack-coverage-design.md

@@ -1,6 +1,6 @@
 # mcpp 图形栈:从「能跑通」到「能开发」的覆盖面设计
 
-Date: 2026-08-30 · 前置:[`2026-08-30-gbm-cross-repo-closed-loop-plan.md`](2026-08-30-gbm-cross-repo-closed-loop-plan.md) §19/§20 · **状态:已实现并闭环验证(v1.9,§11 总账 / §14 fork 规范 / §18 八角度复核 / §19 合成器闸门与 GObject 栈,含沙箱实测 · §20 gio 与一次被推翻的判断 · §21 namespace 是契约 · §22 pango:文本排版接上了)**
+Date: 2026-08-30 · 前置:[`2026-08-30-gbm-cross-repo-closed-loop-plan.md`](2026-08-30-gbm-cross-repo-closed-loop-plan.md) §19/§20 · **状态:已实现并闭环验证(v2.0,§11 总账 / §14 fork 规范 / §18 八角度复核 / §19 合成器闸门与 GObject 栈,含沙箱实测 · §20 gio 与一次被推翻的判断 · §21 namespace 是契约 · §22 pango:文本排版接上了 · §23 模块名与包装体完备性)**
 
 ## 0. 这份文档解决什么
 
@@ -2174,3 +2174,72 @@ error: dependency 'gnome.gobject' is requested as both a version dep
 | 端到端 | 1,492 个像素 |
 
 **文本排版那条线接上了。**
+
+---
+
+## 23. 模块名与包装体完备性:两个已发布包的修正(2026-08-31)
+
+### 23.1 托管方不是负责方 —— 模块名改了两个
+
+| 包(索引键) | 旧模块名 | 新模块名 |
+|---|---|---|
+| `freedesktop.cairo` | `freedesktop.cairo` | **`cairo`** |
+| `freedesktop.libdisplay-info` | `freedesktop.displayinfo` | **`displayinfo`** |
+
+**索引 namespace 是货架标签,从不进模块名。** cairo 是它自己的项目,
+freedesktop.org 托管 git 但不拥有那套接口;libdisplay-info 解析的是 **VESA** 的
+EDID/DisplayID,`freedesktop.*` 这个模块前缀留给 freedesktop 自己的规范(像
+`freedesktop.wayland.client`)。同一条规则下 `freedesktop.egl` 导出的是
+`khronos.egl`。
+
+改名会让老消费者**编译报错并点名模块** —— 这是最该发现问题的地方。
+
+### 23.2 ⭐ 两个包装体都不完备,而且都是「没人问过」
+
+| | 导出 | 缺了什么 |
+|---|---|---|
+| `cairo` | 470 → **697** | `cairo_t`,以及**全部 192 个枚举量** |
+| `displayinfo` | 206 → **501** | **全部 295 个枚举量** |
+
+`cairo_t` 的原因很具体:正则是 `cairo_[a-z0-9_]+_t`,要求 `cairo_` 与 `_t` 之间
+至少有一个字符 —— 于是
+```c
+typedef struct _cairo cairo_t;      /* cairo.h:135 */
+```
+**这个库里用得最多的类型,恰好是它唯一匹配不到的名字。**
+
+**为什么一直没人发现,这才是要记的:**
+
+- cairo:唯一 import 那个模块的测试**同时**写了 `#include <cairo.h>`,头文件把
+  模块缺的都补上了。**一个从没被要求独立站住的 import,不是被测试了,是被装饰了。**
+- displayinfo:**没有任何测试点过一个枚举量**。而且只在 GCC 上加一个也发现不了 ——
+  导出 enum 会让枚举量在 GCC 可见,clang 拒绝同一个文件(§21.5)。
+
+### 23.3 它是从**外面**被发现的
+
+`gnome.pangocairo` **不能混用两条消费路线**(`pangocairo.h` 经 glib 到
+`<stdio.h>`),所以它只能问模块要 `cairo_t` —— 结果什么都没拿到。
+
+于是 pango 那一版带了个变通:**自己扫 `cairo.h`**。现在 cairo 修好了,
+**那个变通已经删掉**,普通的 `export import cairo;` 就够了 —— 本来就该够。
+
+> 一个包的包装体缺不缺,自己的测试可能永远问不出来;**是别的包在用它的时候问出来的。**
+
+### 23.4 发布形态:发资产,不重切 tag
+
+这两个包用的是 **release 资产**(`cairo-1.18.2-mcpp2.tar.gz`)而不是 tag 归档,
+所以修正走 `-mcpp3` 新资产:**tag 不动,零窗口**,而且资产名自带修订号。
+(对比 glib/pango 用 tag 归档,只能原地重切 + 换 sha。)
+
+⚠️ 本地 store 按 `(name, version)` 索引,所以已经解开过 `1.18.2` 的机器**拿不到
+新资产** —— 本次验证时就撞上了(`failed to read compiled module`),清掉 store
+条目才对。这是「包版本与上游对齐」这条约定的固定代价,值得写在这里。
+
+### 23.5 结果
+
+| | |
+|---|---|
+| `cairo` | 697 个导出;测试拆成两个文件,一条路线一个,module.cpp **一个头都不包** |
+| `displayinfo` | 501 个导出;测试比较两个枚举量,导出清单再塌陷就是编译错误 |
+| `gnome.pangocairo` | 变通删除,`export import cairo;` |
+| 验证 | 五个示例 × 两条工具链全绿 |

pkgs/f/freedesktop.cairo.lua

@@ -47,6 +47,36 @@
 -- `cairo_paint` working, and gives every PATH garbage fixed-point coordinates:
 -- `cairo_rectangle(4,4,16,16)` came out with extents -8.03e+06 … 4.37e+06 and
 -- `cairo_fill` changed zero pixels. Measured.
+--
+-- ─────────────────────────────────────────────────────────────────────────
+-- ⚠️ THE MODULE IS `cairo`, NOT `freedesktop.cairo`
+--
+--     import cairo;
+--
+-- cairo is its own project; freedesktop.org hosts the git and does not own the
+-- interface. **The index namespace is a shelf label and never enters the module
+-- name** — the same reason `freedesktop.egl` exports `khronos.egl`.
+--
+-- This changed in the mcpp3 asset. A consumer on the old name gets a compile
+-- error naming the module, which is the right place to find out.
+--
+-- ⭐ AND THE WRAPPER WAS INCOMPLETE: 470 names, missing `cairo_t` and all 192
+-- enumerators. It was found from OUTSIDE, by `gnome.pangocairo`, which cannot
+-- mix the two consumption routes and therefore had to ask the module for
+-- `cairo_t` and got nothing.
+--
+--     cairo_t          `typedef struct _cairo cairo_t;` — the typedef regex
+--                      needed a character between `cairo_` and `_t`, so the
+--                      most-used type in the library was the one it could not
+--                      match.
+--     192 enumerators  every CAIRO_FORMAT_*, CAIRO_STATUS_*, CAIRO_OPERATOR_*.
+--                      The scan never entered a `typedef enum` body.
+--
+-- ⚠️ WHY NOBODY NOTICED, and the lesson worth keeping: the only test that
+-- imported the module ALSO wrote `#include <cairo.h>`, so the header supplied
+-- whatever the module lacked. **An import that is never asked to stand on its
+-- own is not tested, it is decorated.** The fork now has two test files, one
+-- per route, and exports 697 names.
 package = {
     spec        = "1",
     namespace   = "freedesktop",
@@ -60,10 +90,10 @@ package = {
         linux = {
             ["1.18.2"] = {
                 url = {
-                    GLOBAL = "https://github.com/mcpplibs/cairo/releases/download/v1.18.2/cairo-1.18.2-mcpp2.tar.gz",
-                    CN     = "https://gitcode.com/mcpp-res/cairo/releases/download/1.18.2/cairo-1.18.2-mcpp2.tar.gz",
+                    GLOBAL = "https://github.com/mcpplibs/cairo/releases/download/v1.18.2/cairo-1.18.2-mcpp3.tar.gz",
+                    CN     = "https://gitcode.com/mcpp-res/cairo/releases/download/1.18.2/cairo-1.18.2-mcpp3.tar.gz",
                 },
-                sha256 = "202352a38d0847628c55cd0f9c67e85ac78667ad76bc88691ab47606d1104ef7",
+                sha256 = "5a82cfb52365f40b7ecc679d7d67c3c34459aef4a69b4b75c2c0c964d74ec6e1",
             },
         },
     },

pkgs/f/freedesktop.libdisplay-info.lua

@@ -45,14 +45,35 @@
 -- THE MODULE IS GENERATED, not hand-written: 206 names read out of the public
 -- headers, so a version bump cannot silently drop one.
 --
---     import freedesktop.displayinfo;
+--     import displayinfo;
 --
 -- and that also removes a real burden — not one of the seven public headers has
 -- an `extern "C"` block, so a C++ TU that #includes them mangles every
 -- declaration and fails to link with `undefined reference to
 -- di_info_get_make(di_info const*)`. The module does that wrapping once, inside
 -- the module purview, so a consumer does not. (compat.libseat has the same
 -- upstream problem and no module, so there the consumer still wraps.)
+--
+-- ─────────────────────────────────────────────────────────────────────────
+-- ⚠️ THE MODULE IS `displayinfo`, NOT `freedesktop.displayinfo`
+--
+--     import displayinfo;
+--
+-- freedesktop.org HOSTS this library; the interfaces it parses are **VESA's**
+-- (EDID, DisplayID). `freedesktop.*` in a module name is for freedesktop's own
+-- specifications, the way `freedesktop.wayland.client` is. The index namespace
+-- is a shelf label and never enters the module name.
+--
+-- This changed in the mcpp3 asset; a consumer on the old name gets a compile
+-- error naming the module.
+--
+-- ⭐ AND THE WRAPPER EXPORTED 206 NAMES AND NOT ONE ENUMERATOR — the module
+-- carried `enum di_edid_…` but nothing inside it. 295 of them.
+--
+-- ⚠️ It went unnoticed because no test named an enumerator, and adding one on
+-- GCC alone would not have noticed either: exporting an enum makes its
+-- enumerators visible there, and clang rejects the same file. The same gap was
+-- in `freedesktop.cairo`. 501 names now, and the test compares two enumerators.
 package = {
     spec        = "1",
     namespace   = "freedesktop",
@@ -66,10 +87,10 @@ package = {
         linux = {
             ["0.2.0"] = {
                 url = {
-                    GLOBAL = "https://github.com/mcpplibs/libdisplay-info/releases/download/v0.2.0/libdisplay-info-0.2.0-mcpp2.tar.gz",
-                    CN     = "https://gitcode.com/mcpp-res/libdisplay-info/releases/download/0.2.0/libdisplay-info-0.2.0-mcpp2.tar.gz",
+                    GLOBAL = "https://github.com/mcpplibs/libdisplay-info/releases/download/v0.2.0/libdisplay-info-0.2.0-mcpp3.tar.gz",
+                    CN     = "https://gitcode.com/mcpp-res/libdisplay-info/releases/download/0.2.0/libdisplay-info-0.2.0-mcpp3.tar.gz",
                 },
-                sha256 = "8df9a8064146b2b38378bd8b146894d084ffda43b88b43c485b954da073a7617",
+                sha256 = "8b2238a7b275e7da831d96c4f73b607a522d03620e6d5e0f12c605ef04659c8b",
             },
         },
     },

pkgs/p/gnome.pango.lua

@@ -84,14 +84,14 @@ package = {
             ["1.56.1"] = {
                 url = {
                     GLOBAL = "https://github.com/mcpplibs/pango/archive/refs/tags/1.56.1.tar.gz",
-                    -- ⚠️ The container tag is `1.56.1-4`, not `1.56.1`. gitcode
+                    -- ⚠️ The container tag is `1.56.1-5`, not `1.56.1`. 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/pango/releases/download/1.56.1-4/pango-1.56.1.tar.gz",
+                    CN     = "https://gitcode.com/mcpp-res/pango/releases/download/1.56.1-5/pango-1.56.1.tar.gz",
                 },
-                sha256 = "b470a658e05ef0e14d779bc852371ce52dd11efa76d10782347be40f1d63476b",
+                sha256 = "0882a260346cbdb6b553177269a8acc083a2985bbe929b458406af98e90a03ad",
             },
         },
     },

pkgs/p/gnome.pangocairo.lua

@@ -22,12 +22,20 @@
 -- then segfaulted. The fork's test now reads the font map's FONT TYPE and
 -- requires `CAIRO_FONT_TYPE_FT`, which catches it immediately.
 --
--- ⚠️ AND freedesktop.cairo's MODULE IS NOT SUFFICIENT ON ITS OWN. Measured on
--- 1.18.2: 470 names, ZERO enumerators (no `CAIRO_FORMAT_ARGB32`, no
--- `CAIRO_FONT_TYPE_FT`) and no `cairo_t`. The index's own cairo example does
--- not notice, because it writes BOTH `#include <cairo.h>` and
--- `import freedesktop.cairo;`. pangocairo cannot mix the routes, so its module
--- scans cairo.h itself; that can go when cairo's wrapper is fixed.--
+-- ✅ AND THIS PACKAGE IS WHERE cairo's MODULE GAP WAS FOUND. On the
+-- `1.18.2-mcpp2` asset `freedesktop.cairo` exported 470 names, ZERO
+-- enumerators and no `cairo_t`. The index's own cairo example did not notice,
+-- because it wrote BOTH `#include <cairo.h>` and `import freedesktop.cairo;`
+-- — the header supplied what the module lacked, so that import had never been
+-- asked to stand on its own.
+--
+-- It surfaced here because pangocairo CANNOT mix the two routes
+-- (`pangocairo.h` reaches glib and therefore <stdio.h>), so it had to ask the
+-- module for `cairo_t` and got nothing. This package carried a workaround for
+-- one release — scanning cairo.h itself — and that workaround is GONE:
+-- `1.18.2-mcpp3` exports 697 names and the plain re-export is enough.
+--
+-- The module it re-exports is now `cairo`, not `freedesktop.cairo`.--
 -- ─────────────────────────────────────────────────────────────────────────
 -- ⭐ TWO WAYS TO CONSUME IT, AND YOU PICK ONE
 --
@@ -90,14 +98,14 @@ package = {
             ["1.56.1"] = {
                 url = {
                     GLOBAL = "https://github.com/mcpplibs/pango/archive/refs/tags/1.56.1.tar.gz",
-                    -- ⚠️ The container tag is `1.56.1-4`, not `1.56.1`. gitcode
+                    -- ⚠️ The container tag is `1.56.1-5`, not `1.56.1`. 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/pango/releases/download/1.56.1-4/pango-1.56.1.tar.gz",
+                    CN     = "https://gitcode.com/mcpp-res/pango/releases/download/1.56.1-5/pango-1.56.1.tar.gz",
                 },
-                sha256 = "b470a658e05ef0e14d779bc852371ce52dd11efa76d10782347be40f1d63476b",
+                sha256 = "0882a260346cbdb6b553177269a8acc083a2985bbe929b458406af98e90a03ad",
             },
         },
     },

pkgs/p/gnome.pangoft2.lua

@@ -68,14 +68,14 @@ package = {
             ["1.56.1"] = {
                 url = {
                     GLOBAL = "https://github.com/mcpplibs/pango/archive/refs/tags/1.56.1.tar.gz",
-                    -- ⚠️ The container tag is `1.56.1-4`, not `1.56.1`. gitcode
+                    -- ⚠️ The container tag is `1.56.1-5`, not `1.56.1`. 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/pango/releases/download/1.56.1-4/pango-1.56.1.tar.gz",
+                    CN     = "https://gitcode.com/mcpp-res/pango/releases/download/1.56.1-5/pango-1.56.1.tar.gz",
                 },
-                sha256 = "b470a658e05ef0e14d779bc852371ce52dd11efa76d10782347be40f1d63476b",
+                sha256 = "0882a260346cbdb6b553177269a8acc083a2985bbe929b458406af98e90a03ad",
             },
         },
     },

tests/examples/cairo/tests/cairo.cpp

@@ -1,4 +1,4 @@
-// freedesktop.cairo — draw something, read the pixels back, and check that the
+// cairo, via its HEADERS — draw something, read the pixels back, and check that the
 // feature split actually took effect.
 //
 // Cairo needs no display and no fonts on disk to draw: an image surface is
@@ -14,11 +14,26 @@
 #ifdef __linux__
 
 #include <cairo.h>
+// ⚠️ cairo.h does NOT reach the feature headers. This name used to arrive from
+// the module — which is precisely the confusion this split removes: on the
+// header route you include what you use.
+#include <cairo-ft.h>
 
 #include <cstdio>
 #include <cstring>
 
-import freedesktop.cairo;
+// ⚠️ THIS FILE IS THE HEADER ROUTE, AND IT NO LONGER IMPORTS.
+//
+// It used to do BOTH — `#include <cairo.h>` and `import freedesktop.cairo;` —
+// and that is why nobody noticed the module was incomplete: the header
+// supplied every name the module was missing, including `cairo_t` and all 192
+// enumerators. An import that is never asked to stand on its own is not
+// tested, it is decorated.
+//
+// `tests/module.cpp` next door takes the module route alone. Two files,
+// because on GCC the two routes cannot be combined in one TU anyway: reaching
+// a system header both textually and through a module's global fragment makes
+// the same type two entities.
 
 namespace {