feat(eui-neo): add 0.6.1

.agents/docs/2026-10-14-bump-eui-neo-0.6.1-plan.md

@@ -0,0 +1,132 @@
+# Design doc: bump `compat.eui-neo` to 0.6.1
+
+Date: 2026-10-14
+
+Follow-up to `.agents/docs/2026-09-26-bump-eui-neo-0.6.0-plan.md` and commit `2863bb1` (0.6.0).
+Upstream released 0.6.1; this bumps the index from 0.6.0 to 0.6.1.
+
+## Source and version
+
+| | |
+|---|---|
+| Upstream | `https://github.com/sudoevolve/EUI-NEO` |
+| Version | `0.6.1` (latest release) |
+| Tarball | `archive/refs/tags/v0.6.1.tar.gz` |
+| sha256 | `cd0bd6bd437177db097da146fc20f4b8b01c70d0db561a6f8e26baa26dbbbf5f` (computed twice, stable) |
+| Wrap dir | `EUI-NEO-0.6.1/` — absorbed by the existing `install()` hook (`normalise_layout`) |
+| CN mirror | **pending** — no `mcpp-res` write access on this machine; plain-string url fallback per SOP |
+| License | Apache-2.0 (unchanged) |
+
+## What changed upstream (0.6.0 → 0.6.1)
+
+Release notes: `docs/releases/v0.6.1.md` (window lifecycle + self-drawn windows, Linux input,
+component interaction, cross-backend build stability).
+
+- **`CORE_SOURCES` and `VULKAN_SOURCES` are byte-identical to 0.6.0** (extracted and diffed from
+  both `CMakeLists.txt`). The lib's source list, deps, features and flags carry over unchanged.
+- `core/window/x11_moveresize.h` is **new but header-only**: EWMH `_NET_WM_MOVERESIZE`
+  move/resize delegation. libX11 is dlopen'd at runtime — the header states verbatim that "no
+  build configuration gains a link-time X11 dependency", and `-ldl` is already on the linux
+  ldflags (glad). It DOES add a compile-time `<X11/Xlib.h>` on linux:
+  `window_backend.cpp` includes the header **unconditionally** (line 3) and the header guards
+  its X11 block only on `__linux__ && !__ANDROID__`. Resolution path: the transitive
+  `compat.glfw → compat.x11` dep chain (dep `include_dirs` propagate to consumers — the same
+  propagation the glib `private_include_dirs` pairing exists to suppress). Verified by the
+  build, not assumed — see below.
+- `ime_bridge.c` grows a third arm, `#elif defined(__linux__) &&
+  defined(EUI_GLFW_X11_IME_POSITION)`, the X11 IME candidate-follows-caret fix. The arm calls
+  `glfwSetX11InputMethodCursorPos`, a function upstream's `scripts/patch_glfw_x11_ime.cmake`
+  **adds to its bundled glfw** (`3rd/dependencies.cmake` now rewrites the bundled target's
+  sources at configure time). The index builds `compat.glfw`, unpatched upstream 3.4 — the
+  define is never set here, the arm is inert, and our linux IME keeps the 0.6.0 `#else` stub.
+  Enabling the fix would mean patching `compat.glfw`, out of scope for a version bump.
+- `3rd/dependencies.cmake` also tightens `find_package(glfw3 3.4)` — applies to upstream's
+  bundled-vs-system glfw choice only; we pin `compat.glfw` 3.4.
+- The rest is header/app-level — with ONE exception, found by testing rather
+  than reading: in the SDL2 backend, `beginWindowMove` / `beginWindowResize`
+  call `sendX11MoveResizeForWindow` **unguarded**, but the SDL2 branch's
+  declaration sits inside `#if __linux__ && … && defined(SDL_VIDEO_DRIVER_X11)`
+  (window_backend.cpp:223-298) — unlike the GLFW branch's copy (:608), which
+  upstream hoisted out of the guard with its own `#else return false`. Three
+  builds fail, all measured:
+  - linux + `compat.sdl2` — its config is dummy+offscreen only, so the macro is
+    absent (local eui-neo-sdl2 run): `'sendX11MoveResizeForWindow' was not
+    declared` at :571.
+  - **macOS and Windows** (PR #531 CI legs): `use of undeclared identifier` at
+    :567/:571 — `__linux__` is false there, so NO macro can wake the
+    declaration; it does not exist at all. Upstream's own CI builds the SDL2
+    backend on Linux only and never sees this. The other SDL2 X11 helper,
+    `x11ContentScale`, is safe — its call site is inside the same triple guard
+    as its body.
+- `apps/vcd_viewer` removed, new upstream tests/docs — none reach the lib or this index.
+
+## Descriptor changes (`pkgs/e/compat.eui-neo.lua`)
+
+- `xpm.<linux|macosx|windows>["0.6.1"]`: plain-string GLOBAL url (no CN mirror), sha256 above.
+- **One functional change**, in `generated_files["mcpp_generated/mcpp_eui_backends.h"]`,
+  split by platform, active only when `MCPP_FEATURE_SDL2` is on (the header is already
+  force-included into every one of this package's TUs):
+  - **linux**: `#define SDL_VIDEO_DRIVER_X11 1` wakes the guarded declaration. It can
+    never fire at runtime — compat.sdl2 has no X11 driver, so no SDL window is ever an
+    X11 window, `SDL_GetWindowWMInfo` never reports `SDL_SYSWM_X11`, and every delegation
+    returns false (upstream's "unavailable → report failure" contract). The
+    `<SDL_syswm.h>` / `<X11/Xresource.h>` it pulls resolve through the same transitive
+    `compat.glfw → compat.x11` chain the GLFW leg already uses.
+  - **macOS / Windows**: `__linux__` is false, so no macro can wake the block — instead a
+    `core::window` stub `sendX11MoveResizeForWindow(void*, int)` returning false, guarded
+    by `defined(__cplusplus)` so force-included C TUs (`tray_bridge.c`, `native_bridge.c`)
+    do not hit `unknown type name 'namespace'`. That is
+    the fallback upstream's GLFW branch (:608) and its own release notes both describe
+    ("on non-X11 platforms the delegation returns false and callers keep their
+    programmatic fallback"). `Handle` is `void*` in the SDL2 branch (`window_types.h`),
+    so the signature matches the unguarded call sites exactly, and the real linux-only
+    definition cannot exist on this branch of the `#if`, so there is no redefinition.
+  - Both halves stay package-private; neither may become an interface define — consumers
+    would inherit the lie on every platform (on macOS, SDL_syswm.h would then demand X11
+    headers consumers don't have). The stub branch is additionally syntax-checked against
+    the exact call-site pattern in both C and C++ modes.
+- No source list, dep, feature-shape or flag changes otherwise.
+
+## Test members
+
+All seven `tests/examples/eui-neo*` members bumped to `0.6.1` (from 0.6.0). No test source
+changes: the 0.6.1 surface this workspace exercises (backends, app-mains, tray, markdown,
+network, window) is unchanged in shape; the new API (transparent framebuffer, delegated
+move/resize, onStart/appId) is window-lifecycle surface that needs a compositor and is not
+assertable headless.
+
+## Verification
+
+Locally, mcpp 2026.10.1.2 (same as CI `MCPP_VERSION`), Linux x86_64, `MCPP_INDEX_MIRROR=GLOBAL`:
+
+- Index lint: `check_mirror_urls`, `check_package_name`, `check_reserved_namespace`,
+  `check_cross_package_refs`, `check_platform_version_parity`, `check_duplicate_versions` —
+  all clean; `mcpp xpkg parse` accepts the descriptor.
+- First pass: six of seven members green, **eui-neo-sdl2 failed to compile** —
+  `'sendX11MoveResizeForWindow' was not declared in this scope`
+  (window_backend.cpp:571), the upstream guard asymmetry described above. That is what
+  produced the `SDL_VIDEO_DRIVER_X11` shim; it is measured, not hypothetical.
+- Second pass, with the shim: all seven members `mcpp test` green: eui-neo (headless),
+  eui-neo-window, eui-neo-app-main, eui-neo-markdown, eui-neo-sdl2 (sdl2+network),
+  eui-neo-vulkan, eui-neo-tray.
+- **PR #531 CI then failed the macOS and Windows workspace legs**:
+  - First CI revision: `use of undeclared identifier 'sendX11MoveResizeForWindow'`
+    at :567/:571. Same root cause, worse shape: off linux the declaration does not
+    exist at all, so the linux macro shim did not apply.
+  - Second CI revision: the `core::window` stub was introduced, but because
+    `mcpp_eui_backends.h` is force-included into all package TUs via `cflags`, C TUs
+    (`tray_bridge.c`, `native_bridge.c`) failed with `error: unknown type name 'namespace'`.
+  - Third revision: guarded the stub with `defined(__cplusplus)`, ensuring C TUs
+    bypass the C++ namespace block while C++ TUs retain the fallback declaration.
+- `compile_commands.json` (fresh, from the passing builds): the 0.6.1
+  `window_backend.cpp` TU carries `compat-x-x11/1.8.13/include` via the transitive
+  `compat.glfw → compat.x11` chain — `<X11/Xlib.h>` resolves with no direct dep added.
+  The shim reaches the sdl2 TU through the same force-include
+  (`-include mcpp_eui_backends.h`) every eui-neo TU already carries — the macro lives in
+  the header, so it never appears as a `-D` on the compile line; the evidence it is in
+  effect is the failed-then-passing eui-neo-sdl2 compile across the two runs.
+
+CI status: linux legs (default + llvm×4), lint, build, select, measure, timings,
+mirror-cn-reachable and graphics-side-effects passed on earlier revisions; macOS and
+Windows failed on the SDL2 member only due to the unguarded C TU namespace in the stub,
+which the `__cplusplus` guard (this revision) resolves.

pkgs/e/compat.eui-neo.lua

@@ -17,7 +17,7 @@
 -- them as `"3rd/stb_image.h"` — except miniaudio, which audio.cpp includes
 -- BARE; see the 0.6.0 note below.
 --
--- The build recipe below tracks upstream `CMakeLists.txt` (v0.6.0): CORE_SOURCES
+-- The build recipe below tracks upstream `CMakeLists.txt` (v0.6.1): CORE_SOURCES
 -- plus the OpenGL backend and, for the glfw window backend, `ime_bridge.c`.
 -- 0.5.5 grew a Shadertoy subsystem: render_backend.h and include/eui/types.h now
 -- include core/render/shadertoy.h unconditionally, and opengl_backend.cpp calls
@@ -80,8 +80,59 @@
 -- of tray_bridge.c (diff vs 0.5.9: comments + EOF newline only). `3rd/`,
 -- deps and the backend/app-main source lists are unchanged.
 --
+-- 0.6.1: CORE_SOURCES and VULKAN_SOURCES are byte-identical to 0.6.0, so
+-- the lib's source list, deps, features and flags carry over unchanged.
+-- The release is window lifecycle + component fixes, and two of its pieces
+-- deserve notes because they LOOK like descriptor work and are not:
+--   * `core/window/x11_moveresize.h` is NEW but header-only: EWMH
+--     _NET_WM_MOVERESIZE move/resize delegation, with libX11 dlopen'd at
+--     RUNTIME (no link-time X11 dependency — upstream says so verbatim in
+--     the header, and -ldl is already on the linux ldflags for glad).
+--     It DOES add a compile-time `<X11/Xlib.h>` on linux, because
+--     window_backend.cpp includes it UNCONDITIONALLY and the header guards
+--     its X11 block only on `__linux__ && !__ANDROID__`. That include
+--     resolves through the transitive compat.glfw → compat.x11 dep chain
+--     (verified by the 0.6.1 build of the eui-neo member — do not add a
+--     direct compat.x11 dep without measuring a failure first).
+--   * ime_bridge.c grows a third arm,
+--     `#elif defined(__linux__) && defined(EUI_GLFW_X11_IME_POSITION)`,
+--     implementing the X11 IME candidate-window-follows-caret fix. The arm
+--     calls glfwSetX11InputMethodCursorPos, a function upstream's
+--     `scripts/patch_glfw_x11_ime.cmake` ADDS TO ITS BUNDLED glfw — the
+--     index builds compat.glfw, which is unpatched upstream 3.4, so the
+--     define is never set here and the arm is inert; our linux IME keeps
+--     the `#else` stub exactly as 0.6.0 did. Enabling the fix would
+--     require patching compat.glfw, which is out of scope for a bump.
+--   * ⚠️ THE ONE REAL DESCRIPTOR CHANGE. In the SDL2 backend,
+--     window_backend.cpp calls sendX11MoveResizeForWindow from
+--     beginWindowMove/beginWindowResize UNGUARDED, but the SDL2 branch's
+--     declaration sits INSIDE `#if __linux__ && … &&
+--     defined(SDL_VIDEO_DRIVER_X11)` (:223-298) — unlike the GLFW branch's
+--     copy (:608), which upstream hoisted out of the guard with its own
+--     `#else return false`. Three builds fail (all measured): linux +
+--     compat.sdl2 (dummy+offscreen config, macro absent — local eui-neo-sdl2
+--     run), macOS and Windows (`__linux__` false, so NO macro can wake the
+--     declaration — both CI legs of PR #531; upstream's own CI builds the
+--     SDL2 backend on Linux only). The fix is one package-private shim in
+--     mcpp_eui_backends.h, split by platform: on linux define
+--     SDL_VIDEO_DRIVER_X11 for this package's own TUs, waking the guarded
+--     block (it can never fire at runtime — compat.sdl2 has no X11 driver to
+--     produce an X11 window, so every delegation returns false); elsewhere
+--     supply the core::window stub the GLFW branch and upstream's release
+--     notes both describe ("on non-X11 platforms the delegation returns
+--     false and callers keep their programmatic fallback"), guarded by
+--     `__cplusplus` so C TUs (tray_bridge.c, native_bridge.c) don't choke
+--     on `namespace`. It must NOT be an interface define: consumers would
+--     inherit the lie on every platform.
+-- The rest is header/app-level: transparent framebuffer
+-- (DslAppConfig::transparent / framebufferTransparent), beginWindowMove /
+-- beginWindowResize (the consumers of x11_moveresize.h), DslAppConfig
+-- onStart()/appId(), GLFW 3.3/3.4 macro guards (we pin 3.4), platform.cpp
+-- dialog-diagnostics capture (mkstemp, POSIX-guarded), dropdown/slider
+-- z-order and OpenGL pixel-unpack fixes. No `3rd/` pin moves.
+--
 -- All `mcpp` paths are GLOBS relative to the verdir; the leading `*/` absorbs
--- the GitHub tarball's `EUI-NEO-0.6.0/` wrap layer.
+-- the GitHub tarball's `EUI-NEO-0.6.1/` wrap layer.
 package = {
     spec        = "1",
     namespace   = "compat",
@@ -187,6 +238,11 @@ package = {
                 url    = "https://github.com/sudoevolve/EUI-NEO/archive/refs/tags/v0.6.0.tar.gz",
                 sha256 = "11d0725bcc6f16abbbea6052b58c6949008cf78ceaa1645002ba1b3abf9772a1",
             },
+            -- 0.6.1: see the linux block — plain-string GLOBAL, no CN mirror.
+            ["0.6.1"] = {
+                url    = "https://github.com/sudoevolve/EUI-NEO/archive/refs/tags/v0.6.1.tar.gz",
+                sha256 = "cd0bd6bd437177db097da146fc20f4b8b01c70d0db561a6f8e26baa26dbbbf5f",
+            },
         },
         macosx = {
             ["0.5.3"] = {
@@ -234,6 +290,11 @@ package = {
                 url    = "https://github.com/sudoevolve/EUI-NEO/archive/refs/tags/v0.6.0.tar.gz",
                 sha256 = "11d0725bcc6f16abbbea6052b58c6949008cf78ceaa1645002ba1b3abf9772a1",
             },
+            -- 0.6.1: see the linux block — plain-string GLOBAL, no CN mirror.
+            ["0.6.1"] = {
+                url    = "https://github.com/sudoevolve/EUI-NEO/archive/refs/tags/v0.6.1.tar.gz",
+                sha256 = "cd0bd6bd437177db097da146fc20f4b8b01c70d0db561a6f8e26baa26dbbbf5f",
+            },
         },
         windows = {
             ["0.5.3"] = {
@@ -281,6 +342,11 @@ package = {
                 url    = "https://github.com/sudoevolve/EUI-NEO/archive/refs/tags/v0.6.0.tar.gz",
                 sha256 = "11d0725bcc6f16abbbea6052b58c6949008cf78ceaa1645002ba1b3abf9772a1",
             },
+            -- 0.6.1: see the linux block — plain-string GLOBAL, no CN mirror.
+            ["0.6.1"] = {
+                url    = "https://github.com/sudoevolve/EUI-NEO/archive/refs/tags/v0.6.1.tar.gz",
+                sha256 = "cd0bd6bd437177db097da146fc20f4b8b01c70d0db561a6f8e26baa26dbbbf5f",
+            },
         },
     },
 
@@ -340,6 +406,49 @@ package = {
  * the SDL2 define, which is how upstream spells it too). */
 #if defined(MCPP_FEATURE_SDL2)
 #  define EUI_WINDOW_BACKEND_SDL2 1
+/* 0.6.1, SDL2 backend only: in window_backend.cpp upstream hoists the GLFW
+ * branch's sendX11MoveResizeForWindow OUT of any platform guard (the :608
+ * definition carries its own `#else return false`), but the SDL2 branch's
+ * copy sits INSIDE `#if __linux__ && !__ANDROID__ &&
+ * defined(SDL_VIDEO_DRIVER_X11)` (:223-298) — while beginWindowMove /
+ * beginWindowResize call it UNGUARDED (:567/:571). Three builds fail with
+ * 'sendX11MoveResizeForWindow' not declared, all measured:
+ *
+ *   * linux + compat.sdl2 — its linux config is dummy+offscreen only, so
+ *     SDL_VIDEO_DRIVER_X11 is absent (eui-neo-sdl2 member, local). Wake the
+ *     guarded block by defining the macro for THIS PACKAGE'S OWN TUs. It
+ *     can never fire at runtime: no SDL window is ever an X11 window under
+ *     compat.sdl2, so SDL_GetWindowWMInfo never reports SDL_SYSWM_X11 and
+ *     every delegation returns false — upstream's own "unavailable →
+ *     report failure" contract. The <SDL_syswm.h>/<X11/Xresource.h> it
+ *     pulls resolve through the same transitive compat.glfw → compat.x11
+ *     chain the GLFW leg already relies on.
+ *
+ * Both halves must NOT become interface defines — consumers would inherit
+ * the lie on every platform (on macOS, SDL_syswm.h would then demand X11
+ * headers consumers do not have). */
+#  if defined(__linux__) && !defined(__ANDROID__)
+#    define SDL_VIDEO_DRIVER_X11 1
+#  elif defined(__cplusplus)
+/* macOS / Windows: `__linux__` is false, so no macro can wake the guarded
+ * declaration — upstream's own CI builds the SDL2 backend on Linux only and
+ * never sees it (measured on both CI legs: use of undeclared identifier at
+ * :567/:571). Supply the fallback upstream's GLFW branch and its release
+ * notes both describe — "on non-X11 platforms the delegation returns false
+ * and callers keep their programmatic fallback" — as a core::window stub.
+ * Handle is void* in the SDL2 branch (window_types.h), so the signature
+ * matches exactly, and the later real definition cannot exist on this branch
+ * of the #if (it is linux-only), so there is no redefinition. Guarded by
+ * __cplusplus because mcpp_eui_backends.h is force-included into the package's
+ * C TUs (tray_bridge.c, native_bridge.c) as well as its C++ ones. */
+namespace core::window {
+inline bool sendX11MoveResizeForWindow(void* window, int direction) {
+    (void)window;
+    (void)direction;
+    return false;
+}
+}
+#  endif
 #endif
 ]==],
             ["mcpp_generated/eui_neo_platform_tu.cpp"] = [==[

tests/examples/eui-neo-app-main/mcpp.toml

@@ -30,4 +30,4 @@ name = "eui-neo-app-main-tests"
 version = "0.1.0"
 
 [dependencies.compat]
-eui-neo = { version = "0.6.0", features = ["app-main"] }
+eui-neo = { version = "0.6.1", features = ["app-main"] }

tests/examples/eui-neo-markdown/mcpp.toml

@@ -7,4 +7,4 @@ version = "0.1.0"
 # what switches components/markdown.h from its fallback to the real parser.
 
 [dependencies.compat]
-eui-neo = { version = "0.6.0", features = ["markdown"] }
+eui-neo = { version = "0.6.1", features = ["markdown"] }

tests/examples/eui-neo-sdl2/mcpp.toml

@@ -7,4 +7,4 @@ version = "0.1.0"
 # into a real libcurl-backed implementation. The render backend stays OpenGL,
 # unnamed — naming a feature no longer costs you the defaults.
 [dependencies.compat]
-eui-neo = { version = "0.6.0", features = ["sdl2", "network"] }
+eui-neo = { version = "0.6.1", features = ["sdl2", "network"] }

tests/examples/eui-neo-tray/mcpp.toml

@@ -25,4 +25,4 @@ name = "eui-neo-tray-tests"
 version = "0.1.0"
 
 [dependencies.compat]
-eui-neo = "0.6.0"
+eui-neo = "0.6.1"

tests/examples/eui-neo-vulkan/mcpp.toml

@@ -6,7 +6,7 @@ version = "0.1.0"
 # package resolves the exclusive choice in its own preprocessor from the
 # MCPP_FEATURE_* flags. The window backend stays GLFW, unnamed.
 [dependencies.compat]
-eui-neo = { version = "0.6.0", features = ["vulkan"] }
+eui-neo = { version = "0.6.1", features = ["vulkan"] }
 
 [build]
 cxxflags = ["-DHAVE_EUI_VULKAN=1"]

tests/examples/eui-neo-window/mcpp.toml

@@ -34,4 +34,4 @@ name = "eui-neo-window-tests"
 version = "0.1.0"
 
 [dependencies.compat]
-eui-neo = "0.6.0"
+eui-neo = "0.6.1"

tests/examples/eui-neo/mcpp.toml

@@ -3,4 +3,4 @@ name = "eui-neo-tests"
 version = "0.1.0"
 
 [dependencies.compat]
-eui-neo = "0.6.0"
+eui-neo = "0.6.1"