Crash in lower with a choice value: const_id.is_concrete() fails

[danakj]

Description of the bug:

CHECK failure at toolchain/lower/function_context.cpp:187: const_id.is_concrete(): Missing value: inst4E000330 {kind: AcquireValue, arg0: inst4E00032F, type: type(inst4E000013)} in specific<none> has non-concrete value runtime

choice VehicleType {
   Bike,
   Car,
   Bus
}
fn F() {
  let my_vehicle: VehicleType = VehicleType.Car;
}

Please report issues to https://github.com/carbon-language/carbon-lang/issues and include the crash backtrace.
Stack dump:
0.	Program arguments: /opt/compiler-explorer/carbon-trunk/bin/carbon compile --output=/app/output.s --asm-output <source>
1.	Carbon version: 0.0.0-0.nightly.2026.03.06+97af347
2.	Pending diagnostics:
3.	Filename: <source>
4.	<source>:6:1: Lowering function {}
          fn F() {
          ^~~~~~~~
5.	<source>:7:33: Lowering name_ref
            let my_vehicle: VehicleType = VehicleType.Car;
                                          ^~~~~~~~~~~~~~~
 #0 0x0000568df40a24e8 llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) (/opt/compiler-explorer/carbon-trunk/bin/carbon+0x866d4e8)
 #1 0x0000568df409fa74 llvm::sys::RunSignalHandlers() (/opt/compiler-explorer/carbon-trunk/bin/carbon+0x866aa74)
 #2 0x0000568df40a2f81 SignalHandler(int, siginfo_t*, void*) Signals.cpp:0:0
 #3 0x00007f29f1842520 (/lib/x86_64-linux-gnu/libc.so.6+0x42520)
 #4 0x00007f29f18969fc pthread_kill (/lib/x86_64-linux-gnu/libc.so.6+0x969fc)
 #5 0x00007f29f1842476 gsignal (/lib/x86_64-linux-gnu/libc.so.6+0x42476)
 #6 0x00007f29f18287f3 abort (/lib/x86_64-linux-gnu/libc.so.6+0x287f3)
 #7 0x0000568def4c5d36 Carbon::Internal::CheckFailImpl(char const*, char const*, int, char const*, llvm::StringRef) (/opt/compiler-explorer/carbon-trunk/bin/carbon+0x3a90d36)
 #8 0x0000568def3beffc void Carbon::Internal::CheckFail<Carbon::TemplateString<5>{"CHECK"}, Carbon::TemplateString<36>{"toolchain/lower/function_context.cpp"}, 187, Carbon::TemplateString<22>{"const_id.is_concrete()"}, Carbon::TemplateString<56>{"Missing value: {0} {1} in {2} has non-concrete value {3}"}, Carbon::SemIR::InstId&, Carbon::SemIR::Inst, Carbon::SemIR::SpecificId&, Carbon::SemIR::ConstantId&>(Carbon::SemIR::InstId&, Carbon::SemIR::Inst&&, Carbon::SemIR::SpecificId&, Carbon::SemIR::ConstantId&) (/opt/compiler-explorer/carbon-trunk/bin/carbon+0x3989ffc)
 #9 0x0000568def3beeda Carbon::Lower::FunctionContext::GetValue(Carbon::SemIR::InstId) (/opt/compiler-explorer/carbon-trunk/bin/carbon+0x3989eda)
#10 0x0000568def3cb024 Carbon::Lower::HandleInst(Carbon::Lower::FunctionContext&, Carbon::SemIR::InstId, Carbon::SemIR::NameRef) (/opt/compiler-explorer/carbon-trunk/bin/carbon+0x3996024)
#11 0x0000568def3bb3b4 Carbon::Lower::FunctionContext::LowerInst(Carbon::SemIR::InstId) (/opt/compiler-explorer/carbon-trunk/bin/carbon+0x39863b4)
#12 0x0000568def3ba871 Carbon::Lower::FunctionContext::LowerBlockContents(Carbon::SemIR::InstBlockId) (/opt/compiler-explorer/carbon-trunk/bin/carbon+0x3985871)
#13 0x0000568def3a9209 Carbon::Lower::FileContext::BuildFunctionBody(Carbon::SemIR::FunctionId, Carbon::SemIR::SpecificId, Carbon::SemIR::Function const&, Carbon::Lower::FileContext&, Carbon::SemIR::Function const&) (/opt/compiler-explorer/carbon-trunk/bin/carbon+0x3974209)
#14 0x0000568def3a7db8 Carbon::Lower::FileContext::LowerDefinitions() (/opt/compiler-explorer/carbon-trunk/bin/carbon+0x3972db8)
#15 0x0000568def3a3def Carbon::Lower::LowerToLLVM(llvm::LLVMContext&, llvm::IntrusiveRefCntPtr<llvm::vfs::FileSystem>, Carbon::FixedSizeValueStore<Carbon::SemIR::CheckIRId, llvm::function_ref<Carbon::Parse::TreeAndSubtrees const& ()>, Carbon::Untagged> const&, Carbon::SemIR::File const&, int, Carbon::Lower::LowerToLLVMOptions const&) (/opt/compiler-explorer/carbon-trunk/bin/carbon+0x396edef)
#16 0x0000568dee5318b9 void llvm::function_ref<void ()>::callback_fn<Carbon::(anonymous namespace)::CompilationUnit::RunLower()::$_0>(long) compile_subcommand.cpp:0:0
#17 0x0000568dee52df1e Carbon::(anonymous namespace)::CompilationUnit::LogCall(llvm::StringLiteral, llvm::StringLiteral, llvm::function_ref<void ()>) compile_subcommand.cpp:0:0
#18 0x0000568dee52aff0 Carbon::CompileSubcommand::Run(Carbon::DriverEnv&) (/opt/compiler-explorer/carbon-trunk/bin/carbon+0x2af5ff0)
#19 0x0000568dee51ff54 Carbon::Driver::RunCommand(llvm::ArrayRef<llvm::StringRef>) (/opt/compiler-explorer/carbon-trunk/bin/carbon+0x2aeaf54)
#20 0x0000568dee4f6ffb Carbon::Main(int, char**) busybox_main.cpp:0:0
#21 0x0000568dee4f47d4 main (/opt/compiler-explorer/carbon-trunk/bin/carbon+0x2abf7d4)
#22 0x00007f29f1829d90 (/lib/x86_64-linux-gnu/libc.so.6+0x29d90)
#23 0x00007f29f1829e40 __libc_start_main (/lib/x86_64-linux-gnu/libc.so.6+0x29e40)
#24 0x0000568dee4f4725 _start (/opt/compiler-explorer/carbon-trunk/bin/carbon+0x2abf725)
CHECK failure at toolchain/lower/function_context.cpp:187: const_id.is_concrete(): Missing value: inst4E000330 {kind: AcquireValue, arg0: inst4E00032F, type: type(inst4E000013)} in specific<none> has non-concrete value runtime
Program terminated with signal: SIGSEGV

What did you do, or what's a simple way to reproduce the bug?

No response

What did you expect to happen?

No response

What actually happened?

No response

Any other information, logs, or outputs that you want to share?

No response

[danakj]

This might be a good first bug? I am not sure. cc @dwblaikie

[dwblaikie]

This might be a good first bug? I am not sure. cc @dwblaikie

yeah, never can be entirely sure until you get there - but seems worth a shot in that direction.

I'd guess we haven't figured out what storage of choice types (choice objects?) looks like...

[Ko496-glitch]

@danakj @dwblaikie i would love to work on this, any suggestions how should i approach or what storage would be better for choice ?

[Ko496-glitch]

@dwblaikie i am thinking like c++ we can add integer to variants, i dont know if that would be viable design choice, any suggestions ?

[danakj]

The design of choice is very much still in flux, but for what we have implemented right now, the choice is represented in semir as an empty tuple or an integer:

carbon-lang/toolchain/check/handle_choice.cpp

Lines 257 to 272 in 6304df1

	SemIR::TypeId discriminant_type_id = [&] {
	if (num_alternative_bits == 0) {
	// Even though there's no bits needed, we add an empty field. We want to
	// prevent constructing the Choice from an empty struct literal instead of
	// going through an alternative. And in the case there is no alternative,
	// then there's no way to construct the Choice (which can be a useful
	// type).
	//
	// TODO: Find a way to produce a better diagnostic, and not require an
	// empty field.
	return GetTupleType(context, {});
	} else {
	return MakeIntType(context, node_id, SemIR::IntKind::Unsigned,
	context.ints().Add(num_alternative_bits));
	}
	}();

[Ko496-glitch]

@danakj this might be a dumb question but i understand right now ConstantID = runtime and should be concrete ? Is the right fix to add an EvalConstantInst for the SemIR inst kind used by VehicleType.Car ?

[danakj]

ConstantIds can be concrete, symbolic, or runtime. It sounds like lower needs to emit a type based on the TypeId chosen above (but I am still somewhat unfamiliar with how lower works).

[Ko496-glitch]

thanks, i will then update EvalConstantInst for SemIR::AcquireValue in eval_inst.cpp to properly forward the constant id from the operand instead of falling back to TODO ? makes sense ?

[danakj]

Appreciate the enthusiasm but I might suggest contributing in a different way for now, or spending more time learning how the toolchain works. AcquireValue isn't related to choice types/storage, and eval is not part of lower. Perhaps spending some time reading the toolchain docs and implementation will help.

[Ko496-glitch]

yes i just realized that i have to read the doc for toolchains, will spend some time in that direction, thanks tho !

[Ko496-glitch]

@danakj just a quick update — I’m currently reading through the docs on check and lower, and it’s been really helpful for understanding the toolchain. I’m planning to spend the rest of this week digging into it and trying to come up with a solution for this issue. Just wanted to let you know I’m still actively working on it :)