Fix nontrivial-memcall warnings in cute_graphics_sdlgpu.cpp

[pusewicz]

Two different situations here, so two different treatments:

cf_sdlgpu_make_shader_from_bytecode / cf_sdlgpu_make_compute_shader_from_bytecode used CF_NEW (default-initialization) followed by a memset to zero the plain C-array/scalar members that default-init leaves uninitialized — verified empirically that CF_NEW's new(args) (Type) form really does leave those as garbage. The memset also stomped the Cute::Array members, which were already properly constructed (harmlessly, since their zero-state matches empty, but still unspecified behavior on a live non-trivial object). Switched to CF_PLACEMENT_NEW with an explicit T() (value-initialization), which zero-initializes every member — plain arrays included — in one well-defined step, so the memset goes away entirely rather than being merely silenced.

Verified with a temporary 0xCD-poison + CF_ASSERT pass over every member (removed once confirmed clean) that nothing is left uninitialized after the change.

cf_sdlgpu_shader_swap_contents / cf_sdlgpu_compute_shader_swap_contents do a deliberate raw byte-swap of two live objects for hot-reload (both types are pointer-only handles with no self-references, per the existing comment) — that's intentional, not a bug. Left the swap as-is and just cast the destination pointers to void*, per the compiler's own suggested fix.

Verified: framework + test suite builds warning-clean, and ./tests passes the same 313/333 as master (remaining 20 are the pre-existing headless GPU rendering failures from #572, unrelated to this change).

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, you can upgrade your account or add credits to your account and enable them for code reviews in your settings.

[Copilot]

Pull request overview

This PR targets -Wnontrivial-memcall warnings in the SDL_gpu backend by removing memset on already-constructed shader internal objects and adjusting shader hot-reload swapping code paths.

Changes:

• Replaced CF_NEW + CF_MEMSET with value-initialized placement-new allocations for CF_ShaderInternal and CF_ComputeShaderInternal.
• Updated the shader/compute-shader “swap contents” hot-reload routines to cast destinations to void* for CF_MEMCPY calls.

Suppressed comments (1)

src/cute_graphics_sdlgpu.cpp:2651

• cf_sdlgpu_compute_shader_swap_contents has the same issue as the graphics shader swap: CF_ComputeShaderInternal contains non-trivial Cute::Array members, so the memcpy swap is UB even if cast to void*. Also, cf_sdlgpu_destroy_compute_shader currently CF_FREEs the object without running its destructor, which leaks the reflection/sampler Cute::Array allocations. Use a move-based swap and explicitly run the destructor before freeing (mirrors cf_sdlgpu_destroy_shader_internal).

	uint8_t tmp[sizeof(CF_ComputeShaderInternal)];
	CF_MEMCPY(tmp, pa, sizeof(tmp));
	CF_MEMCPY((void*)pa, pb, sizeof(tmp));
	CF_MEMCPY((void*)pb, tmp, sizeof(tmp));

[pusewicz]

Addressed in 7822c21 — swapped to Cute::swap (already exists in the codebase, does a proper move-based 3-step swap).