Defining what happens when a bool isn’t 0 or 1

[JohanEngelen]

I just learned that there can be subtle bugs when assuming that a bool value is either 1 or 0.
Read about it here: https://discourse.llvm.org/t/defining-what-happens-when-a-bool-isn-t-0-or-1/86778

We do this in (at least) two locations:

ldc/gen/dvalue.cpp

Lines 139 to 145 in 3acd761

	if (isOptimizationEnabled()) {
	// attach range metadata for i8 being loaded: [0, 2)
	llvm::MDBuilder mdBuilder(gIR->context());
	llvm::cast<llvm::LoadInst>(rval)->setMetadata(
	llvm::LLVMContext::MD_range,
	mdBuilder.createRange(llvm::APInt(8, 0), llvm::APInt(8, 2)));
	}

ldc/gen/dvalue.cpp

Lines 242 to 248 in 3acd761

	if (isOptimizationEnabled()) {
	// attach range metadata for i8 being loaded: [0, 2)
	llvm::MDBuilder mdBuilder(gIR->context());
	llvm::cast<llvm::LoadInst>(rval)->setMetadata(
	llvm::LLVMContext::MD_range,
	mdBuilder.createRange(llvm::APInt(8, 0), llvm::APInt(8, 2)));
	}

I recreated the example from the LLVM post in D:
https://d.godbolt.org/z/hsrErrGe3 (C code: https://godbolt.org/z/cofqzMooz)
The assembly is the same but somehow the bug does not show up for LDC... huh? (3rd executor should output "0 0 0 100" when the bug happens) I don't understand why it is not showing up while the assembly code is nearly identical...

[rainers]

The assembly is the same but somehow the bug does not show up for LDC... huh? (

You have selected a different compiler version in the executors (and missing option -O3). These are not set to use the latest CI build, but 1.40.1. If switched to latest CI, they show the same result as the C code. LDC 1.40.1 only uses the lowest bit as index.

[JohanEngelen]

Thanks Rainer! It seems to require LLVM 20.
https://d.godbolt.org/z/K168Y3T46