[siracozmen01]

Problem

mwinGamepad_GetTestRumble and mwinGamepad_GetTestTriggerRumble in include/maul-window/test.h report how many rumbles a gamepad of the test platform was given in a uint32_t* countOut. #376 made every count the API reports a size_t, and a count of calls is a count too.

Change

• Both functions take size_t* countOut.
• The test backend keeps the count as a size_t.
• test/test_gamepad.c reads the count into a size_t.
• test/test_outputs.c checks that a refused call clears the count, now apart from the other three outputs. It fills and checks those as one struct with no padding, and a size_t beside three 32-bit values leaves padding on a 64-bit build.

Verification

• Built with no errors or warnings: the gcc-14 tree, the clang-20 ASan, minimal and fuzz trees, MinGW, wasm32-wasi and Android.
• tools/check.sh changed passes, the gamepad and outputs suites included.
• With the count's clear taken out of the test backend's GetRumble, outputs fails on both counts, for a call with no context and for a gamepad that is gone.

BREAKING CHANGE: mwinGamepad_GetTestRumble and mwinGamepad_GetTestTriggerRumble take size_t* countOut instead of uint32_t* countOut.
Closes #488

[siracozmen01]

Review at 30b26e2: ready, with one point not verified. No findings. Read, not run.

• The clear-on-failure contract covers the wider type by construction. GetRumble clears with mwinClearOut(countOut, sizeof *countOut) (backend_test_pads.c:141), so the clear widens with the pointer's type to all 8 bytes on 64-bit, before any refusal. The stored mwinTestRumble.count is size_t too. *countOut = rumble->count and the rumble->count + 1 in backend_test.c:465 and 473 neither truncate nor change behavior. On 32-bit builds size_t is the same 32 bits as before.
• The test change is right. Rumble drops the count so the struct's static_assert of no padding still holds (3 × 4 bytes). The count is filled with the pattern and checked on its own after each call, under the same status as the other outputs. With the clear removed, the 8-byte count keeps the pattern on a refused call. That is the failure the body reports.
• test.h, the backend and test_gamepad.c take and pass size_t*, and the BREAKING CHANGE footer names both functions. That fits #371's ruling that a count is size_t.

Not verified: callers outside the five files. Any C caller still passing uint32_t* is an incompatible pointer type, which the warning-as-error builds would refuse. So the clean gcc-14, ASan, minimal, fuzz, MinGW, wasm32-wasi and Android builds cover everything those trees compile. Not covered are samples outside the build, the docs, and any JS or Java test hook that reaches these through a raw 4-byte slot. On wasm32 size_t is 4 bytes, so a JS slot stays right there. A search for GetTestRumble and GetTestTriggerRumble across the repository would settle it.