Fix: writeI53ToI64 crashes with -sASSERTIONS=2 when num >= 2^32

[49iohcy]

Hello.

I've recently been working on a WebGPU demo w/ Dawn WebGPU C/C++ implementation. When I tried to get the adapter limit in a node:24-slim devcontainer on my M2 Apple Silicon, I got 4294967296 for the maxBufferSize, which is a valid parameter for writeI53ToI64.

When -sASSERTIONS=2 is set, there's a trailing checkInt32 call in the makeSetValue implementation. Without masking the low word of num, checkInt32 throws in the current implementation of libint53.mjs, which I believe is not the desired behavior.

The following are relevant references for the most up-to-date implementations of Emscripten and Dawn.

https://github.com/google/dawn/blob/89b4626b4e838738227dc410209c4d03410d94a9/third_party/emdawnwebgpu/pkg/webgpu/src/library_webgpu.js#L581

emscripten/src/parseTools.mjs

Lines 494 to 502 in f7179e4

	function makeSetValue(ptr, pos, value, type) {
	var rtn = makeSetValueImpl(ptr, pos, value, type);
	if (ASSERTIONS == 2 && (type.startsWith('i') || type.startsWith('u'))) {
	const width = getBitWidth(type);
	const assertion = `checkInt${width}(${value})`;
	rtn += `;${assertion}`;
	}
	return rtn;
	}

emscripten/src/lib/libint53.js

Lines 21 to 30 in f7179e4

	$writeI53ToI64: (ptr, num) => {
	{{{ makeSetValue('ptr', 0, 'num', 'u32') }}};
	var lower = {{{ makeGetValue('ptr', 0, 'u32') }}};
	{{{ makeSetValue('ptr', 4, '(num - lower)/4294967296', 'u32') }}};
	#if ASSERTIONS
	var deserialized = (num >= 0) ? readI53FromU64(ptr) : readI53FromI64(ptr);
	var offset = {{{ getHeapOffset('ptr', 'u32') }}};
	if (deserialized != num) warnOnce(`writeI53ToI64() out of range: serialized JS Number ${num} to Wasm heap as bytes lo=${ptrToString(HEAPU32[offset])}, hi=${ptrToString(HEAPU32[offset+1])}, which deserializes back to ${deserialized} instead!`);
	#endif
	},

[sbc100]

Thanks for the tracking down!

I guess we are missing a test for this. Perhaps we can expand test/core/test_int53.c to include a test here.

[sbc100]

I think we normally write this as num >>> 0

[49iohcy]

Thanks for the feedback. I'll see what I can do with the test and let you know!