Support cross-compilation

[jserv]

This pull request adds comprehensive cross-compilation and WebAssembly support. The core achievement is a unified build infrastructure that automatically detects compiler type (GCC/Clang/Emscripten) via mk/toolchain.mk and
configures appropriate flags, backend restrictions, and feature availability through Kconfig. For WebAssembly specifically, the branch introduces complete browser-based testing workflow with assets/web/index.html as the runtime interface, scripts/serve-wasm.py as development server with correct MIME types, and automatic post-build artifact deployment via the wasm-install Makefile target.

The backend architecture underwent major refactoring to establish a consistent interface where all four backends (SDL, VNC, fbdev, headless) now implement a standardized .start method and use twin_dispatch_once() for their main loops, ensuring work queue and timeout mechanisms function correctly across platforms. This architectural unification enabled significant API simplification where the new twin_run() function completely abstracts platform differences.

---

Summary by cubic

Adds cross-compilation and WebAssembly support with unified toolchain detection and a new twin_run API for consistent backend loops. Enables browser testing and improves SDL performance.

• New Features

  • Automatic compiler detection (GCC/Clang/Emscripten) and cross-compile prefix via toolchain.mk, Kconfig, and scripts/detect-compiler.py.
  • WebAssembly build path: SDL-only backend, Emscripten flags, wasm-install target, assets/web/index.html, and scripts/serve-wasm.py for local testing.
  • Unified app startup: twin_run(init) and twin_dispatch_once(), with .start implemented in SDL, VNC, fbdev, and headless.
  • apps/main.c now uses twin_run(init_demo_apps) to support both native and WASM.
  • Cross-platform guards for endianness and 64-bit fixed-point math fallback on 32-bit targets.

• Refactors

  • Standardized backend main loops with twin_dispatch_once() to ensure work queue and timeouts behave uniformly.
  • SDL: robust init/cleanup, reduced idle CPU with smart delay, and proper Emscripten main loop integration.
  • fbdev/headless/vnc: added start functions and fixed error paths and resource leaks; POSIX guards and stubs for headless on non-POSIX.
  • Build updates: use Emscripten ports for SDL/libpng/libjpeg, avoid building tools for WASM, and add macOS linker flag detection.

[cubic-dev-ai]

7 issues found across 18 files

Prompt for AI agents (all 7 issues)

Understand the root cause of the following 7 issues and fix them.


<file name="scripts/detect-compiler.py">

<violation number="1" location="scripts/detect-compiler.py:49">
CROSS_COMPILE should not override an explicit CC value; otherwise clang-based cross builds try to run a non-existent &lt;prefix&gt;gcc and detection fails.</violation>

<violation number="2" location="scripts/detect-compiler.py:69">
Split the compiler command into argv tokens before invoking subprocess so wrapped compilers (e.g., CC=&quot;ccache clang&quot;) are detected correctly.</violation>
</file>

<file name="src/twin_private.h">

<violation number="1" location="src/twin_private.h:190">
`_twin_xfixed_mul_32bit` loses carry bits when accumulating the 64-bit partial products, so the software fallback produces incorrect results for valid operands.</violation>

<violation number="2" location="src/twin_private.h:196">
The 64-bit fixed-point division fallback `_twin_xfixed_div_32bit` for 32-bit platforms is mathematically incorrect, implementing standard integer division instead of fixed-point division. This breaks path rendering in `src/path.c` on 32-bit cross-compilation targets, defeating a key objective of the PR.</violation>
</file>

<file name="backend/sdl.c">

<violation number="1" location="backend/sdl.c:263">
twin_sdl_exit now destroys SDL objects that were already freed in _twin_sdl_destroy (invoked on SDL_QUIT), leading to double-destroy crashes during shutdown.</violation>
</file>

<file name="assets/web/index.html">

<violation number="1" location="assets/web/index.html:98">
Only add the error styling when the message type is actually an error so informational logs aren&#39;t misrepresented.</violation>
</file>

<file name="mk/toolchain.mk">

<violation number="1" location="mk/toolchain.mk:162">
This detection invokes $(CC) with -Wl,-no_warn_duplicate_libraries but no input file, so clang/gcc exit with “no input files” before ld ever runs. As a result LD_SUPPORTS_NO_WARN is set to 1 even when ld rejects the flag, and the final link then fails on toolchains that lack -no_warn_duplicate_libraries.</violation>
</file>

_{React with 👍 or 👎 to teach cubic. Mention @cubic-dev-ai to give feedback, ask questions, or re-run the review.}

[cubic-dev-ai]

The 64-bit fixed-point division fallback _twin_xfixed_div_32bit for 32-bit platforms is mathematically incorrect, implementing standard integer division instead of fixed-point division. This breaks path rendering in src/path.c on 32-bit cross-compilation targets, defeating a key objective of the PR.

Prompt for AI agents

Address the following comment on src/twin_private.h at line 196:

<comment>The 64-bit fixed-point division fallback `_twin_xfixed_div_32bit` for 32-bit platforms is mathematically incorrect, implementing standard integer division instead of fixed-point division. This breaks path rendering in `src/path.c` on 32-bit cross-compilation targets, defeating a key objective of the PR.</comment>

<file context>
@@ -140,10 +140,96 @@ typedef int64_t twin_xfixed_t;
+    return neg ? -(int64_t) result : (int64_t) result;
+}
+
+static inline twin_xfixed_t _twin_xfixed_div_32bit(twin_xfixed_t a,
+                                                   twin_xfixed_t b)
+{
</file context>

[cubic-dev-ai]

twin_sdl_exit now destroys SDL objects that were already freed in _twin_sdl_destroy (invoked on SDL_QUIT), leading to double-destroy crashes during shutdown.

Prompt for AI agents

Address the following comment on backend/sdl.c at line 263:

<comment>twin_sdl_exit now destroys SDL objects that were already freed in _twin_sdl_destroy (invoked on SDL_QUIT), leading to double-destroy crashes during shutdown.</comment>

<file context>
@@ -188,23 +241,88 @@ static bool twin_sdl_poll(twin_context_t *ctx)
+    twin_sdl_t *tx = PRIV(ctx);
+
+    /* Clean up SDL resources */
+    if (tx-&gt;texture)
+        SDL_DestroyTexture(tx-&gt;texture);
+    if (tx-&gt;render)
</file context>

✅ Addressed in a330284